GPUs are a dying breed

Relating resolution to GPU death is somewhat flawed. Sure, you can render a scene at 2560x1600 in say, World of Warcraft, and have no problem because the complexity of the scene is not that great. 

Try that in Crysis, and you'll be rendering at < 5 fps.

If you are rendering a scene with blades of grass, individual leaves, truly organic looking shapes (instead of the low poly fakeout texturing techniques they use today), then you have horsepower required that is partly independant of resolution. 

Plus, you are completely ignoring the real-time lighting problem. Sure, they have ok approximations for it now with shader tricks, but it doesn't compare to a full on ray tracing implementation. There's a reason why movies use radiosity rendering algorithms as opposed to polygon rendering. 

Granted, individual blades of grass won't add that much to gameplay in todays games. However, I think that real-time lighting will be of tremendous use to future games because of its ability to set mood. 

The GPUs of the future will likely be ray tracing implementations.

This post is a moronic simplification at best. Resolution is not and has never been the issue. The room for increased shader complexity, polygon count, multiple rendering passes (which increase the need for bit depth to avoid lossiness), etc is nearly infinite. Sure, in 2015 your low end Radeon will be able to play all today's games at1920x1200 - but will it be able to play 2020's games even at 640x480? Hell no. Ask any artist whether they'd turn down another million poly's per model - you'll get the same answer. Until you can't tell the difference between a game and real life, graphics have not caught up with your eyes.

The real question is whether the custom circuitry of the GPU will be subsumed by the many core CPU. I'm guessing not for a while - memory bandwidth and being able to architect for a one-way pipeline will probably keep custom chip vendors in business for years to come.

1000x in 4.5 years?? but that would mean a gf8800gtx is 1000x faster then a gf fx 5900 (2003) and it's not. Not even 100x. Ok maybe 100x :). Still, 1000x the power is probably more like a 15-20 year span.

Issues that haven't seemed to be taken account for is that games increase in scene complexity in any given resolution. A video card that can drive todays games in 1080p will not drive tomorrows game at this resolution. 

I would say sound cards haven't reached the holy grail until audiophiles start featuring them in audio magazines. So there is still room for a market that take sound seriously.

That is what DirectX 10, 11, 12, ...nth is for!!

Microsoft will keep the GPU companies in business!

Except that 1680x1050 LCDs are priced to move in 2007-2008. And then people will buy 1920x1200 in 2008-2010. And then there is Crysis and the like. When we can play a Transformer battle that looks like the movie, then maybe it is over.

I think you're way out of line saying that gpus are not going to matter anymore. Why? because aa and af are only the beginning. What kind of power are we going to need for ray-tracing? and then after ray-tracing im quite sure there will be some other form that will require even more power. ALso takes this for example. If i run an old game say counter strike with a 8800gt and i can completely max out my settings (aa af and rez) can i do the same thing with a game like crysis? i dont think so. which only leads me to say that as games get more complex and more realistic, there will always be more demand for power to make them run. 
ray-tracing, more complex games, im sure im forgetting a few things here but we will definitely always need more powerful cards

We need that power for the matrix, blimey bloke!

You are assuming that the software side keeps the same level for next upcoming years.

Where are the raytraced games at a decent res and frame rate?

Where are the ultra-high-insane ploy count for ultra-high-insane detailed models?

The hardware can scale up to the double every year but the lads at Epic, Id Software or Crytek will always figure out a way to put those GPUs working their *ss out for some decente frame rates.

Even when a GPU can give us a LOTR, Star Wars ou Matrix efects in realtime, someone somewhere will find a way to push things a bit far.

Graphics cards and graphics industry won't slow down until the day when we can render truly realistic lifelike images on the fly. That won't be until 2200. lol

The GPU will die when in a few years time Intel and AMD are pushing 128 core processors capable of processing realtime ray-traced graphics at high resolution making the GPU only useful for piping images to the screen with some form of on board memory to act as a buffer to reduce lag.

You don't get it. Future cards are not centered around color depth and the nonsense you've spouted. They're about allowing more geometry/shaders/etc. New games with more complex engines fuel new demand for cards. Sound is an entirely differently story, don't try to compare them. Even if games go down the tubes demand for GPU's as pure number crunchers in scientific analysis will be huge.

Sound card development slowed down because their output is now essentially indistinguishable from real life audio; this is not a good comparison.

When the visual ouput of a GPU is indistinguishable from real life visuals, then they will have come to the end of their development. Given that even the best non-realtime CGI doesn't even come close to real life, I'd say we have quite a ways to go.

Even once we've reached this point, it's quite conceivable that further graphical processing power will be wanted (for example in producing beyond-real-life visuals for some fictional setting).

i remember back when i got my first 486, it had megs and megs of storage space.

"im never gonna fill this thing in a million years"

If you think the industry is going to create the tools to make themselves obsolete,... think again.

Hardware designers rely on software desingers create unnessasary, over balloted aspects of software to try and force you to buy more hardware. 
Take the DirectX 10 situation and beyond.

It's neverending,... one hand washes the other. 

~The prospect of counsel makers to crush pc gaming is huge, problem is, they don't want to loose all that tasty pc gamin' $$$$.

"The hd3870 can run a modern game at 1920*1200?? Sounds like someone hasn't played Crysis yet.

I can't currently render the feature film Beowulf in realtime on an 8800GT* at any decent resolution, and as our ability to mimic reality with CG grows, so will our expectations from the visual portion of gaming.

When I can't tell the difference between what is happening in the real world (a recording) and what a computer is generating in realtime and displaying, then maybe we can talk. Even after that occurs (if ever), the GPU/CPU/whatever responsible can still be made exponentially more efficient.

I don't see R&D in such areas disappearing soon. Even if GPUs and CPUs are to combine, the march of progress must continue, and will, I believe.

The progress curve will taper off over the next few years. You say 4.5 years, I say at least 8 years until we reach a point of no discernible return on consumer-grade visuals.

However, I am certain there will be new mainstream visualization technologies by then that will drive needs for more GPU power. While unlikely to become mainstream any time soon, holographic display technology is one way of pushing graphics processing power requirements by another easy 1000x.

Disagree with you for once. This is only arithmetics...
Aren't you forgetting the large amount of compromise that is still being made by game developers so that todays' games dont kill both the CPU and GPU?

There is tremendous potential to use additional compute power to achieve more realistic physics modelling and 3D rendering. Did you notice how curved surfaces must be defined as a succession of straight lines to limit the amount of polygons? And look at damage modelling in most games, or the very basic lighting effects (vs movies)
Hollywood-class 3D is the next goal, and doing it in real time will take a lot more compute power than we have now.

Keep the good work, it is a pleasure to read the Inq. And Happy Xmas!

By the time polygon accelerators hit the wall, folks will start demanding accelerated raytracing engines for even more realistic rendering. Intel is already demoing 4-8 core raytracing in Quake 3 or 4, and it looks spectacular. Who's to say there won't be raytracing GPUs out there with multiple cores and the whole thing will just continue on?

As a computer animator, I have to say you couldn't have missed the mark more if you tried. The number of tricks, cheats, and compromises needed to get game geometry to play at even a semi-reasonable frame rate of the best GPUs out there, makes it very clear that they still have a very long way to go until they are anywhere near good enough to not need to upgrade anymore. Anyone with some 3D software and a spare weekend can come up with a scene that will choke a top of the line card down to 3 or 4 fps. It is only by having a team of people pouring tons of man-hours at the scene that you can get the poly count and shaders optimized enough to play at a decent framerate, and look passable. There is still a lot of room for improvement in GPUs for many years to come.

When you get to the point where just about any scene you can dream up has no problem being pushed at 30 fps, then you will have a point, but right now developers are having a hard time not choking a system just walking through a wooded path. We are not anywhere near that point yet.

Having built my own PC rigs for the past 11 years, I know there will always be a high end market for GPUs. PC's play dual roles still. 

Production wise - Working in CG animation, video/effects and graphics intense work always benefits from GPU progress. Then there's the whole science/medical/industrial fields. Many of which use the same development tools now and are always looking for a performance edge. 

From a gaming standpoint, 2008/2009 may be lean times for the PC industry. Consoles present a strong match and bigger market for game developers. That puts ATI and nVidia to compete for a market in contraction. 

But really, does it matter? There are dual cpu computers on the market being sold to consumers. A large percentage of the general public doesn't use software that actually uses the 2nd core. The same applies to GPUs.


The second we have enough capable processors that can ray-trace, gpu's are done. Ray-tracing at a high resolution is the holy grail of gaming. It honestly seems as if we're not too far from it, considering a recent project that was seen running at several frames per second.


... 2012 is the year to build my next rig. Gotcha :)

I loved the article, but it really did miss the point.

The hd3870 isn't even twice as fast as the x1950xtx, and the 8800gtx is the grandad of gaming that will be what? nearly two years old before it gets retired? so i struggle to support this doubling of standards every six months, indeed we are lucky if we manage this in twelve months.

With the current trend however, most of the above is totally void. Latest gen games arn't selling like they used to (look at crysis, and ut3), and consoles support this further (wii anyone?), so the target now really isn't going to be perfect reality, as its clear the closer we are getting, the faster we are loosing interest.

Features will start selling new hardware, proper virtual reality, powered by specific cpu's, body sensors, etc, building on the success of increased interaction of modern gaming.

There is room to go further with graphics, but its clearly dropped off alot. And with power requirement and cooling being an increasing factor now its getting near impossible to "stack" performance like previously.

quantum computing could see the demise of the gpu, recently demonstrated etc, or it could open up a whole new ball game.

did we also include ray tracing? thats a massive leap in its current form, let alone added game complexity in the coming years.

we will see new hardware concepts getting merged into current standards. gpu into cpu, physics into gpu and what not.

I have a Dell 30" and an 8800 GTX ad I an already play games like TF2 at 16x AA and 16x AF at 30+ fps without a hitch, so why should there be a wait?

Of course, I think I just proved your point. :/

Current raster graphics cards may be getting towards the end of their era but still 5 years to go is still a worthy investment for nVidia/AMD. Not to mention the fact that there are still many more raster-rendering improvements to be made in terms of true-to-life image reproduction (is it me or is bump-mapping kinda pathetic?)

Either way the next stage of graphics wow-ness is slated to be real time raytracing which from what I can tell only Intel is looking into.

When all else fails, go 3D. Add (virtual) depth to monitors. There is one monitor out already that does 3D, so that will probably be the next step in 5 years. Processing 3D depth of vision on the GPU.

Either that or the GPU guys will diversify by retasking their processors as physics or math co-processors. :)

Cheers,
John

Ever heard of power density?

You know that Moore's law doesn't apply anymore for a few years, right?

How much power and heat would a 1000x faster GPU consume/dissipate if it's made using processes expected for the next 5 years (32nm, 22nm)?

Also, the graphics of the games will have reached its limits when a person can not distinguish the game from reality. It's not about resolutions or antialiasing.

If we were 10 years in the past, you would have written this same lousy article claiming that in 5 years all GPUs would run Quake1 at crazy resolutions and AA, so they must be dying.

Brilliant.

What about the fact that GPU advancements are driving by the entertainment industry? PC games like Crytek's "Crysis" pushes GPU sales....and that game like no other games ever made actually mimic realism. Game developers have dreams of creating games that mimic realism but they won't get to in their lifetime or their grandson's life time.... its going to take a long long time for computer graphics to mimic real life. If GPU's end, than that must mean that the pc/console gaming industry will come to a halt?

Kevin

Well here's a challange, Virtual reality for the Gaming PC. 
http://en.wikipedia.org/wiki/Virtual_reality

or why not go for the big one, the holodeck :)
http://en.wikipedia.org/wiki/Hollow_deck

"Remember what happened to the thriving sound card industry?"
Yeah ,it was killed off by creative, what's your point? ;/
(I get your point, just having a stab at creative)


The thing this article does not take into account is photo realistic image generation. If you want true global lighting effects, you need infinite processing power. Today, we are able to 'approximate' what would look like a good compromise between performance and image quality. But we are still very far from having everything in realtime.. Just to name a few.. global illumination, volumetric, particules effects, raytracing, caustics, and all thoses other nifty features.. If you want to have an idea of what is coming, look at SIGGRAPH conference proceedings.

Computer animations, movies which take days to render haven't been able to fake a real movie, and those are made on server farms. I'm not exactly sure of the state of tech on graphics are, but I don't think we're remotely close. Maybe we just need better artists and animators but theres too much refinement needed to really fool us into perceiving a real enviroment. Maybe the graphics curve will level out soo much that vast improvements in gpu hardware won't reflect on the screen. We'll see but these block worlds and there texture laden surfuces seem full of areas to improve on.

I'm not so sure there will be a cliff, even if there was Microsoft would hand them a parachute, just think of DX9 to DX10 frame rate issues, in 5 years time we might just see a new version or two of this program as well as increasing resolutions with larger monitors, ray tracing, far more polygons, better lighting effects, etc.. not to mention graphics are far from life like at the moment and the possibilty of 3d monitors must be considered. 

Even now, Crysis and other DX10 titles would struggle with one graphics card from either manufacturer at 1920x1200 in DX10 with all the goodies turned on max, two 8800 ultra's might be playable but may not be able to stay above the 60FPS minimum we would like to see without a supersize overclock.

I think the end of the GPU will most likely come from future processors with lots of cores, unless the next few generations of Windows sap all thier power! Only time will tell.

You forgot games with support for 3D monitors, but it's just another raw number like you said.

Surely the ultimate aim of the GPU manufacturers is to provide equivalent quality interactive 3D which is on par with what ever hollywood can dish out. This is a moving target with each special effects movie being more impressive than the last. Take recent films like Transformers and you realise there is still a long way to go for the GPU guys and plenty of Graphics card upgrades for us. 

You are not taking into account that the new games will need more resources due two factors:

1) Easiest development tools, make development process faster, but they need more processing capacities
2) New tecnogies need more processing to make games more realistic. Maybe 1000x is very much for the todays games, but the very new games will need more processing power (even if we can't appreciate it, say DX10).

So, there is enough live to GPUs from now.

Excuse my english

If only it was all that simple. Anyone who really knows their GFX knows there is a LONG LONG way to go. For example, you haven't even included view distance, number of object's visible at one time etc etc. And without including them, all the other stats are meaningless. And frame rates are everything and have a long way to go too. This just sounds like someone trying to create a story where there isn't one. Nice try.

I disagree

In my opinion, game developers will continue to utilize more and more of the GPU power. A game designed in 5-years-time from now will not look the same as one designed now. Look at Crysis. You can’t play that game with today’s GPU even at modest resolutions. 

Then there is physics. Physics calculations have no limit (literally). Sure, after so many physics calculations per frame, you won’t notice a difference, but before we get there, we’ll keep needing GPU power to do more and more those calculations.

So you see, the end won’t happen in 4-5 years. Maybe in 10, probably much later. And when the end does appear to be in sight, Ray Tracing, or something similar, will demand a faster GPU.

And what about multiple displays? How about enough of them to go all they way around you? Can anyone say “holodeck”.

Don’t be like the fool who thought he could predict the future of IT, and has stated that 640 KB of RAM is all that anyone would ever need. Still, it was a thought provoking article.

Revisit this comment in 5, then 10 years. Let’s see who’s crystal ball is cleaner. Then again, perhaps both of our crystal balls are dirty.

Fatone

When you have enough performance the issue turns to quality.
We are still a (very) long way from photo-realistic realtime 3D graphics, It is getting better but there's plenty of distance to go yet.
We are entering a phase where purchasing decisions will be more feature-based than the ability of a a GPU to just be able to drive an even higher-res screen at all.
Honestly, 1920x1200 is more than enough. It is already a slightly higher res. than HDTV.
But there will always be bleeding-edge games drawing ever more power from GPUs for ever more complex lighting and volumetric effects as we approach the grail of photo-realisitic graphics that are visually indistinguishable from reality.

Isn't ray tracing going to add another level of complexity/realism. Once this has been mastered there will be something else better and more GPU intenstive.

You should know better than to predict the end so quickly. How many times has Moore's law been declared dead?

GPU and Game companies also need to catch up with the ever-more-popular widescreen resoultions...consoles are doing much better in that regard!

Someone could have written this article back in the day of the 286 processor. Back then it was easy to render a perfekt image of a non movable blue triangle at the screens max res in real time. Why whould we need to develop better graphics hardware?

When we get to the point where there is no point of having greater res there will still be lots of ways of modeling light in the physical world more accuratly. How about real time raytraceing a world with infinetly many moveing objects in real time.

There will always be things to calculate and throw clock-cycles at to make the gameworld look more realistic.

what a bullshit article.. sounds like a 10 year old just told me his perspective.

How about photorealistic ray-tracing delivered by retinal scanning display laser goggles? I would bet my annual wage that resolution and image quality would continue to improve exponentially up untill the Singularity :P

People are looking for high fidelity gaming, not just high resolution. So it seems to me that while you may be right about hitting some resolution brick-wall in monitors, there's still plenty of scope for GPU makers to duke it out over the amount of realtime post-processing (pixel shaders?) they can have in a scene... and all that other fangled cinematic stuff they nick from pixar and friends. 

Just as a clue to where competition could lie in the future. Harking back to the release of HL2, we saw how Nvidia's pixel shader implementation was bunk on its then high-end boards. So Valve recommended/forced those card owners down the DX8.1 path... whilst with ATI cards HL2 was (more-)playable at DX9.0.

So really wont it just be buisness as usual, we'll just find new aspects to squabble over?


The same reasoning can be aplied to almost all system components. What is driving CPU sales if not games? If GPUs become powerful enough to make further progress meaningless so will the CPUs. Technology can be enhanced only so much. But I'm sure there will be plenty of other applications to justify R&D so they can milk us out of our money.

Nice article but I am willing to bet that there is a thriving graphics industry well beyond 10 years from now... I don't think we'll even have to wait for holographic displays to find something really nice to waste GPU cycles on. :-)

I think your prediction is somewhat wrong. I do agree that there will come a time when GPU's will no longer need to become faster, but that time is a long way off. If screen sizes and resolution were to stop getting larger, graphics accelerator development would be around for at least another decade.

Game devs use a variety of techniques to simulate a realistic 3d environment within the scope of the current generation of hardware. I'll give you a couple of examples:
-Bump/normal mapping is used because rendering lots of high polygon models would decimate performance.
-In many games, level designers can't just place things around willy nilly. They have to construct visibilty blocks using walls/fog/items to prevent the engine from rendering everything. 
-Todays games (and next gen as well) aren't capable of using ray tracing due to performace. 

Such issues ensure that GPU development will continue for some time.

If 3d visor displays ever catch on, everything will need to be rendered twice--once for each eye. And once graphics cards are fast enough, they can be made continually smaller, quieter, cooler, and use less power.

I'd also argue that the soundcard industry went down the toilet because of Creative's monopoly.

I don't think one can reasonably argue that GPU power has doubled every year for any period of time. And the "holy grail" of graphics is ray-tracing i believe and doing that in real time with decent graphics is much further out than your predicted 4.5 years.

i've thought this before, and to an extent it's true, but you have to take into account whats being displayed on the monitor. how much graphical horsepower is crysis XX going to take to run at that dizzying resolution, with however much AA or AF you can run (something which may well be hyped to sell the top end gpu's im sure, 32xTSAA or something stupid im sure)? the graphics industry is always going to be striving for in the end, something that looks as real as the world before you... i doubt that will be in 4 years...

How can you say that GPU's are going to die. As long as there are graphics to be displayed on a screen there is going to be a need for a processor to process these graphics.
Even if future processors are capable of also processing graphics then the GPU will still exist, although incorporated with the processor.
As future software also gets more and more complex and require more and more processing power then technology will have to advance to keep up with it. Its a never ending cycle, software develops and current hardware cannot cope with it, then hardware is developed to cope, then the software advances further.
Its not rocket science and I am surprised and disappointed that this article has even appeared on the inquirer site, whose articles I normally trust.

Why not throw that crazy parallel processing power into other areas that badly need work like: 

physics, AI, getting me a coke, and social skills?

I'm quite sure that GPUs will travell sound card path. Of course today is still alive some fraction of sound card market, but average computer user is 100% happy with integrated part. And there is no reason to question that in several years average computer user will be happy with integrated GPU part. And addon GPU bords will be sold mainly for special use - for those who has some special, super high quality screens, or super special needs.

You haven't taken into account the increase in polygons games will use... And this is really unpredictable, developers can just keep using more and more polygons to define shapes better, enhanced shaders, and like the other guy says, direct x 11 12 13 14 etc.

Look at the hit DX10 has. These cards 3870 etc. cannot handle these games.

Look at Crysis, sub 20fps at 1600x1200 without any AA at all...

It'll be a while before GPUs really become peripheral to a gamer.

GPUs have a LONG way to go yet, before they get to the holy grail of graphics...reality. I'm thinking 15 years or so down the road. Until they are capable of rendering 10,000 individual units that are photo-realistic, running at 60 fps simutaneously on the battlefield, have no "fog of war"... that's when they can honestly say... gentlemen...we are finished.

GPGPU programming is seeing a lot of interest from researchers.

You have an algorithm A, for a GPU, and algorithm B, for a CPU. They both take the same inputs and produce the same output.

Algorithm B is much less efficient than algorithm A. You still choose to run algorithm B on the GPU as opposed to algorithm A on the CPU because for the same price, the GPU hardware is a heck of a lot faster and it makes up for the difference and then some.

In the near future, there will still be a need for more computing power, so we are not nearing an end. (Bill Gates and 640K RAM.)

Off the top of my head, applications include robots being able to see and identify objects, time-consuming encryption algorithms, and essentially most decoding/encoding algorithms.

If I am wrong and none of this matters, there will still be the war of making a commodity GPU cheaper and cooler and quieter and more available and more compatable etc...

Since when has there been a "doubling" of performance every 6 months? There hasn't. In fact we've previously gone as long as 3 years before we've seen such an improvement in performance. I have no idea where the author got this idea from.

I'm also completely baffled by the argument that the HD3870 has acceptable performance in 1920. Perhaps if you crank some of the details down, then you can get nice framerates in a decent number of games, but in a considerable number of current titles this just isn't possible. Of course this all depends on what your perception of "playable" framerates are. If you have the hand eye coordination of a rubber boot then 30fps may be smooth to you. However we're talking about "limits" here aren't we? Therefore we need to think of it in terms of not only how high the visual quality can go set against human perception, but also framerates. this means you need at LEAST 60fps ALL THE TIME.

This whole article is just nonsense quite frankly. These arguments aside, claiming that we are close to meeting the limits of human perception is incredibly short sighted. I remember when similar things were said of Mario 64. There's always a new fancy visual trick to be pulled off, and when it is we'll all wonder how we lived without it.

Theorycrafting on this topic is really a moot point. Have we got 100% fully realistic visuals in any game yet? No. We've hardly got a (single card solution) card that can run Crysis on Very High with full options in DX10 - and though the graphics in that game are stunning, they are a Far Cry (no pun intended) from being realistic in the definition of the word.

Like the many posters above me have said, resolutions and AA/AF mean squat if the game itself doesn't have good models etc.

This article reminds me of a quote:
"64kb ought to be enough for anyone" - Bill Gates

One BIG miss in your analysis is the desire for more realism. We are not even close. All those pipes and pixel pushing advances are merely approximating reality and we're not even close. We are shooting cartoons are 60 fps. 
What American game players really want is to shoot life like British. There is a long way to that. Ironically, the conclusion may be similar because the holy grail is vector graphics - not raster graphics which is what is done now.

Look to multicore (hint Intel) to get us to the next level of advances - vector graphics. Now we're talking

Wow, just wow. This must be the most clueless article I ever read. Author clearly has no idea of what video cards actually do.

I saw a video recently where someone stacked 3,000 barrels in crysis then knocked them over to watch the physics. It got me thinking: this has got to be the future of games.

Ultimately, for a true simulation of the real world, you'd have to model each atom and their physical interactions with each other. How's that for an end game? Being 20, and given the exponential curve of number crunching, I'd say there's a slim chance I'll see it before I die :)

Jacob Nielson did the number crunching for you.

"...Screen size: Of course, even a "big" monitor is much smaller than, say, a newspaper spread, so we need monitors to be at least three times wider and two times taller, or 12Kx8K pixels in 300 dpi resolution. Given that we want 1200 dpi resolution, we can conclude that monitors should display 48Kx32K pixels.

Scan rate: Refresh rates of 60 frames per second are sufficient to minimize flicker on the conscious level, but we really need at least 120 frames per second for perfect image quality.

A 48Kx32K pixel monitor with 24-bit-per-pixel color needs 4.6 GB videoRAM. Given a refresh rate of 120 fps, a 4.6 GB screen will require about four tera bits per second bandwidth without compression, or one Tbps with some compression..."


It's a good thing you didn't write this article in 2002, or you'd feel mighty silly right now.

While your ultimate conclusion is of course true (there reaches a point in everything where the human is constraining the system), your time frames and the numbers games you play are completely absurd. There are a couple things to consider.

First, take a look at games like Oblivion and Crysis. Oblivion came out in early 2006--we're about 3 months away from its 2 year anniversary... and yet even the highest performing graphics cards today can only push 30 fps at 1920x1200, max settings, no AA, and 8X AF. To get this 21-month-old game running at the optimal 1920x1200@60fps 8xAA/16xAF, if we use your calculations you'll need 2x the performance to get your framerate up, another 8x the performance to get the AA, and another 2x for the AF--which comes out to 32x the performance altogether, or 57x the performance if you want to do it on a 30" display. Even with your insanely idealistic "GPU performance doubles every 6 months", that's still nearly 3 years to go, making it nearly 5 years between the release of Oblivion and the day when a single, $500 graphics card can play it at the highest settings and resolution available at launch. Or what about Crysis? Even with a nice, high-end card you can expect dog performance. 10 fps on a high-end single-card system at 1600x1200 with 4x AA and Very High settings? You'll need 6x the performance to get the framerate up, another 2x the performance for the AA, and 2.13x the performance to go from 1600x1200 to 2560x1600. That yields a much more pleasing result--only an increase of performance by a factor of 25 is needed to get good framerates and good settings at the highest resolution, which is comically less than the graphically inferior Oblivion.

Anyway, both of these games demonstrate one very important point: if you give a game designer performance, he will give you larger worlds and finer detail. In 2001, we all thought that Max Payne looked great. The textures were so realistic, there were cool particle effects when you shot things, and the clothing and faces of the characters looked like (and in some cases, I think were) photographs! We were happy as clams. Then we realized around 2004 that we could use shaders to do some pretty sweet real-time effects, like heat lenses, fluids, and glass--we also started to heavily use bump and specular mapping. These days we're dabbling with subsurface scattering, multi-kilometer view distances, volumetric soft self-shadowing, motion and depth-of-field blurs, and more. And as performance numbers indicate, this is immensely taxing. What else could we tack on to our impressive list of technologies? Well, if every enemy's vision was an individually rendered video feed, and image recognition was then carried out in concert with highly advanced AI, it would result in a disgustingly slow, but theoretically realistic simulation of the world. More likely changes can be identified by watching the cinematic trailer for Starcraft 2 (any Blizzard cinematics including or after Warcraft 3) or any recent realistic CG movie--until games have the level of detail exhibited in these tediously slowly pre-rendered videos, there's still room for improvement. Until the simulation and rendering of an in-game world is negligibly different from the real world, there's room for improvement. If I can't see the reflection of an enemy sneaking up on me on a nearby doorknob, or track someone by looking for subtly crushed folliage, or see the finest wrinkles in a character's face when they frown, or note the different reflective properties between the varnished surface of a door and the wooden splinters that result when I cut it in two, there is room for improvement. And I'm betting it's going to take more than 4 times the current level of graphics performance to accurately model all of the effects and optical properties of a game from 2020 compared to a game from 2007 (at the same resolution).

To make a long story short, there's a lot of room for improvement before you can't tell the difference between a gameplay video and a live-action video. Until we hit that point, graphics cards will benefit from increased realism. After that point, when we've got IMAX dome-like screens with 300 DPI at 90 frames per second, then yes, graphics cards will only really benefit from being cheaper, smaller, and more power-efficient. But I'm guessing this will be a matter of decades, and not 4.5 years.

Apparently it did for the author too, only he got fooled and hit the pub to get drunk enough to produce this article.

People have been predicting the end of Moore's law every fifteen minutes since the words came out of Gordon Moore's mouth over 40 years ago.

That hasn't happened.

Completely realistic virtual environments where the viewer cannot recognize the difference between what is displayed on the monitor and an analogous real-world environment are many orders of magnitude more complex than current graphics technology permits. 

Even with exponential growth in the graphics area, we are still years away from anything that can meet those challenges.

Demand for more transistors and better GPU architecture will be consistent for as long as the manufacturing processes allow.

When you can play a game that looks like LOTR and play's at 60fps, then yes, the GPU is dead. Since it currently takes 1-2weeks to process 1second, I think there is some room for growth. You say valve thinks all video cards are fine, they know they are lying. Can you play with more than 32 people at the same time on TF2? no. How about +200 in a FPS map? nope, not even close.

Lots and lots and lots of room for growth for the next lifetime.

While not so much a dying breed, can we agree that there may be a curve of diminishing returns? Audio made great strides and now for most purposes the ear can't hear any better, so there not much to innovate in the realm of sound these days. 

With graphics, the same fate will befall them, in a way. Our eyes can only see so well. While there is probably much more time needed to hit that point, it will happen one day.

The three main things I see are resolution of screen, the quality of the image in term of rendering quality, and frame rate. I would take a bet that with in two or three years, we will have monitors in mainstram capable of displaying more then enough resolution for human eyeballs.

However I could see it take a decade or two easy for the output rendered by card in real time to be equal to say the "Beowulf" movie. (Been showing in the States for maybe a month).

So my short point is this article should be rewritten as a focus on monitors as I see them hitting there hard cap much, much sooner.

You might be right in saying that it's only a matter of years before hardware can easily render current software effects at great resolution and framerate but you're severely short-sided if you think that's the end of it! Where's the imagination? Sure we have great looking "textures" now but there's really no texture to them. Modern GPUs still can't handle rendering a true sphere or circle. There are work arounds we've come to accept but there are still basic challenges yet to be met!

I find it a little humorous to see this article so soon after another article on the inq criticized recent PC games for being too demanding on hardware.

Seriously, though, like some of the other comments here, I cannot agree that GPUs will die in the coming decade. I will agree that upgrading will eventually become a game of diminishing returns, but games will *never* become so realistic as to be completely indistinguishable from real-life. As such, there will always be a market composed of die-hard enthusiasts willing to put a second mortgage on their house to get slightly better water ripple effects, or whatever.

Also, as GPUs (and PCs in general) become faster we will see games and other programs become increasingly easier to produce, thanks to managed code ala .NET. The more powerful these future libraries become, the more overhead they will require, and the result is yet another force that is driving the adoption of faster hardware. 

And why stop at 2D gaming? I wonder how many GPUs it would take to power a holodeck... :)

Moore's Law predicts maybe an 8x to 16x increase in transistor count over those 5 years. To achieve 1024x increase, the Gfx vendors only have to figure out how to get another 100x out of frequency, architectural improvement, and die area. Yeah, they will only have to come up with a wonder architecture that is 2x faster for the same resources, increase frequency by 10x (10GHz), and increase die area by 5x. That's all. And for $250. Yay!

I agree with just about everyone else who's commented on this article. Point taken about soundcards- they reached a level where there wasn't much discernible difference between what they can reproduced from real life and they'd reached industry standards (Dolby DTS) and whatnot.... GPUs on the other hand are only limited in life if someone comes up with another technology to handle the GPU's job (I can't even speculate on what that may be). 

'Perfect' sound reproduction (as far as the human ear is concerned) is done and dusted, graphics on the other hand can ALWAYS be better. Until you make a machine that can process an unlimited (which is impossible) number of polys and all the other effects, something more powerful to handle graphics will be there.

Sorry mate... you're wrong.

Is it just me or does Direct X 10 lower framerates with (curently) little visible benifit. Maybe some time down the chain Direct X 11-12 or whatever all that wonderfull GPU power will be struggling pumping out 25 FPS. And thats on top of all the physics the GPGPU's will be doing ^_^

I remember reading a voodooextreme editorial in 1998 (right when the 3dfx voodoo 2 SLI became very popular). Basically adding cards until it wasn't necessary. The author guessed 2002 would be the year that graphics cards would be at the end of their usefulness.

Bill gates thought 640k was enough ram. IBM thought 5MB hard drives were large enough, and there was a global demand for about 10 computers.

We can now add this silly (and baiting) column to the scrap heap of history. Congrats. I will quote it every time someone else with limited imagination starts contemplating the end.

The whole premise for this article is a joke, it sounds like the applied logic of a child (or very ignorant adult)

Its so flawed on so many levels that I cant even begin to bother to pull it apart, I know what Ill do, Ill start a thread in Beyond3d and let the laughs begin, I cant wait to see responses from over there about this ridiculous article, 

only problem is, I might get flamed for starting a thread about a subject that is so technically incompetent, that ill be accused of degrading the quality of the forum. 


This truly is a low point for the inq.

both intel and amd are going to make gpgpu's?

This reminds me of when people proclaimed 64K of RAM would be more than enough for anyone...

We haven't even gotten close to the limit of graphics rendering. So many shortcuts (deferred shading, z-buffers, simple lighting models, parallax mapping, just to name a few) are used right now just so that modern games will run on modern hardware. Even the most advanced CGI movie is still easily recognization from real life.

Everyone talks about bigger and bigger. With today;s faster computers, users are more inclined to multitask, surfing the web, reading e-mail, chatting, running iTunes, writing a document and analyzing a spreadsheet.

Just check the financial services and software development industries. Two monitors are a minimum, with some having as many as six or eight.
Yet only a few video card manuafacturers have caught on to this trend: Matrox, maybe. For the rest,their multi monitor cards are relegated to some niche somewhere.

Soon we'll have laptops with two fold up screens, one in the normal screen and the other is at the back: used as a tablet when closed, but a second screen above the normal one when open.


Graphic cards will start taking over things like physics, and take them to a level well beyond Ageia Physx. We will eventually add smooth, non-polygon shapes, like spheres, which take a lot more work. Also character generation has a lot of way to go before we see things like realistic emotions, blemishes, wrinkles, scars and wounds.

We're just now starting to talk about real-time ray-tracing, and no one has mentioned radiosity. Basically you follow the light from the source until it hits the virtual camera. It produces much more wonderful lighting effects, and I'm sure, even though you can currently run it in real-time, it would be willing to eat up tons of GPU cycles for a complex scene with lots of lights, especially if you took away their ability to pre-compute the surface characteristics and used surfaces that could dynamically change (like getting mud on a window).

All these renderings look rather poor, but you can see a comparison of the radiosity lighting here:
http://www.suwiki.org/suwiki/index.php?title=Radiosity

Check out geomerics.com or this video for a better example of what it provides: http://www.youtube.com/watch?v=FB8Nu8_rsDM

Hi,

Well it might not be true...but this quasi quote comes to mind..

"640KB is the most memory a PeeCee will need"..

http://en.wikiquote.org/wiki/Talk:Bill_Gates

Build it and they will use it.....

Movies as software running in a rendering machine....

Interesting discussion..."When it comes to technology, most people over-estimate it in the short term and under-estimate it in the longer term." Dr. Rodney Brooks

http://moderndragons.blogspot.com/2006/04/review-singularity-is-near.html

Frank

Charlie thinks GPUs are soon dead, but hmm, even the best PC graphics I see today really still look rather synthetic, and generation to generation changes are quite evolutionary at this point. Oh yeah, and the graphics architects I know think they have decades of work ahead to give us perfect reality -- you know, like what you see when you look out your office window. 

Gee, I can't seem to figure out which one of them is right and which one sounds like a complete idiot.

Hi,

I am not sure about doubling performance every 6 month but as said the time might very. However you assumed that one screen is what the GPU has to do. There are some usage models that might require more power. Ray tracing was mentioned though CPUs for now seem to be better suited.
Physics on the GFX might add some years to the doom day.
I also think once the GPU power is up we will start see games develpped for "surround GFX" where you can look to 2 other screens to left and right for realism.
I also think that screens of a "dome" formation might appear especially with developments in the flexible screen
So do not rule them out just yet traditionally when more power comes out of the HW people think what to do with it.
I also believe the GFX giants have (or at least they should have) smart people thinking on usage models that will make us want to buy their goods.

We've already reached the tipping point of good enough graphics in the computer hardware wars. DX10 is a joke, Vista is a failure, and Crysis is a drug searching for an addict. Consoles are more attractive for developers and customers due to the appeal of optimizing graphics through yearly software improvements instead of yearly hardware upgrades. Poor nerds outgrow overclocking, tweaking, and patching, while rich nerds outgrow benchmarking, early adopter price gouging, and system instability. One already has switched to console gaming, the other is waiting for a few more price drops. Which one are you?

I like how the first post is pro microsoft: it will most-definitely will, it is the easiest platform to program for (right now anyways). And the 1990s dude I feel you on that one, lol. 

But the article is way off the mark though. I can't believe they actually ran it. there should be a point system for the articles, this would be one of those at the bottom of the pile.

You are dead wrong.

We still haven't got

Stereo display
Ray tracying 
Global illumination
Unlimited polygon count

And who knows if GPU would be required to super high resolution images directly to your brain in 20 years?

Crisis should prove your coment wrong. We don't even have hardware that can run that beast. Not in all it's glory. And in 6 months or so another game will change the rules. It's not resolution that is key, it's realism. Once we have flawless photo realistic gameplay at 60fps then it will be over. While Crisis looks fantastic. It's far from photorealistic. It's still looks like a game. So the end game might be 20 years off, but it's a long time off anyway.

What about ...
Better realtime noise removal.
And other secret ideas even i can think but cant disclose.

What a short-sighted lame article. The nonsense you spouted would only hold true if cryengine2 was the last engine ever developed until the end of time.

Wrong... we've heard this doomsday predictions plenty of times before and hardware just keeps finding ways to improve.

Software definitely needs to catch up with hardware. I agree wholeheartedly.

with true 3 dimensional displays (not display screens), there'll be the added depth to consider about which would be way more demanding than current 2 dimensional displays

"I think there is a world market for maybe five computers", Thomas J. Watson head of IBM in 1943.

"640K ought to be enough for anybody", Bill Gates in 1981.

It seems Charlie is trying to add himself to this list.

A nice long article, Charlie, but it would seem that you were feeling a bit pessimistic when you wrote this. We all get like that sometimes. What this gets us is articles like "Is the PC gaming going to die in the near future?", "How much cores does the average user really need?" and now this.

Chin up, Charlie, as long as there are artists and game developers looking to chase the ultimate realism, the GPUs will not be out of work. Maybe when nVidia can do their excellent Human Head demo level graphics realtime in a game with hundreds or thousands of unique character models we will be close to it.

Until then, think of titles like Alan Wake and Far Cry 2, both of which are fairly near-future affairs and will probably require a lot more than a 8800 GTX to run smoothly at a hi-def rez.

I completI agree with the author of the article! GPU will die! I am developing real time graphics software since the first Silicon Graphics Reality Engine Super computers and I consider myself a bit an expert in this field. So far every post in this forum was against the author and I have to comment!

The author stated correctly that at one point higher resolution, color depth and frame rates will not make sense anymore, because you will not be able to see the difference. So graphics card performance will be high enough and more is not needed, but honestly I do not think it will ever happen, because before that GPUs will die and will be replaced by CPUs. Why?

CPUs will be so fast and have so many cores that real time ray tracing will be possible at high frame rates and high resolutions. 

Todays real time rendering techniques are all about workarounds! Every real time graphics book you read (OpenGL or DirectX) shows how to render refractions, reflections, shadows, global illumination and multiple layers of transparencies … but all these methods, as clever and smart as they may be, are workarounds to render realistic pictures on graphics hardware. A graphics card composes textured triangles into a frame buffer. Done! That’s it! It has been like that 15 years ago and today it is still the same! Faster, higher quality, more sophisticated algorithms, but still the same principle! The problem is: Polygon rasterization is not well suited for realistic images which need e.g. shadows and shadow rendering in OpenGL or DirectX is hell and has many limitations.

Not so ray tracing! It simulates the real world and everything can be computed elegantly without work around! It just needs a LOT of CPU power. The good news is that it scales fantastically well on multiple CPUs or cores. 2 CPUs will ray trace an image twice as fast one … almost :-) Take a look what Intel can ray trace today on current CPUs and play the number game for CPUs. In 10 to 15 years we will have a XXXX Cores CPU running at YY GHZ? Don’t laugh! It will happen! And then you can ray trace 1080p with 16xAA and every feature you can imagine (shadow, reflection, global illumination …) in real time! This will make graphics cards obsolete! 

Almost everything that is done today on GPUs which is not rendering polygons is completely useless! Ray tracing on the GPU? Give me a break! What for? A ray tracer is one of the worst things you can run on a GPU! It is not built for it! Is somebody still seriously trying to compute physics on the GPU? There are some exceptional algorithms that can exploit the way a GPU works which run much faster on a GPU than on a CPU, but these are very very rare!

Now one can argue that the GPU features will be expanded, shaders will be much more flexible than they are today and then you CAN run a ray tracer on the GPU. True, but then it will be like a CPU or actually many CPUs on a graphics card … if you still can call it that way.

I think that Intel will scale up the CPUs faster than Nvidia can build CPU features into their GPUs, so Intel will win! AMD should be a candidate for success after they bought ATI, but what I have see from them recently does not give me much hope! I still think Intel will crush them all.

And here is one more: DirectX and OpenGL will both die! Why? DirectX and OpenGL are simply APIs pushing data into a graphics card. No graphics card, no APi needed! A ray tracer works different, new APIs will emerge! Well, they are already here, but used in non real time rendering.

When will this be? My conservative guess is 12 to 17 years from now.

Have fun and don’t forget to sell your NVidia stocks early enough :-)


To me, advancing real time graphics mean we can have decent Global Illumination, Sub-Surface-Scattering, 128 bits per channel (including alpha), true soft-body-dynamics, skin deformations, dynamic mass interaction and more of this stuff at infinite frame rates and impossible resolutions. 
For me, the simulation limit ends (probably) at the point we have something like Cpt Pickard's Hollodeck (and maybe there's still room for more). Till then the increase factor is something like ten in the power of some seven figure number, so there's still too much room for development.
On the other hand, I still enjoy a game of tetris on a 1st generation Game Boy screen...

Wow, why would you write an article on a subject you know nothing about?

Two words: Global Illumination


Yeah, can really relate to what somone said about the 486, had my 486 DX66mhz, man it had 4mb of ram and 520mb hdd! Wohoo! I was the king of the world, then internet came....

As for the GPU being dead part, well thats basicly nonsense, even if software wouldnt be able to show off the raw power of GPU's people would still aim to get them.

Games now are pretty sweet looking but however they can push limits waaaay much further.

I back in the voodoo 2 days remember sitting on IRC, chatting and some guy in the chat started talked about some new unknown crap company that was gonna make something they called a GPU, it all sounded alot like jibberish, something that just couldnt be done.

Like why the hell would anything be better than a Voodoo card that you connected to your graphicscard?

A card standing on its own? Nah, cant be done.

He sent some link to a page with some smowoff demo movie files from what they believe the card could do.

If the memory serves me right it was some firetruck, some trees with some leaves. My mind said to me, thats impossible, that can't be done.

Showed the same demos to my friends and that said it was impossible aswell.

And when it was released, I went for it and it basicly blew me away. And my friends was pretty sad they didnt go with the new king of the hill instead.

As for gaming memories....

Still remember back when I first saw Unreal, well that felt unreal seeing the waterfalls at the 2:nd level, and the shiny floors on level 1.

Even Quake II felt pretty amazing once.

The Morrowind came along and blew me away, man can games look this good?

Farcry showed me yes they can, but your computer sucks.

And the story just keeps on going.

Each step along the way in the beginning made me feel, is this even possible?

So all in all what was the meaning of this long comment, well just to say as long as its possible to make things more powerful it will be done, and it will be utilized. Because people always crave for more, who settles with less than the best if there is a choice that won't hurt them or somone else.

We should just be glad as it just gives the industri more jobs and creates a nice market.

Just se all the oppurtunities to get a job in the gaming industry these days, not that long ago a game was made by a small team of just a handful of people.

These days with more complex models it takes more time to create all content, witch means a thriving gaming industry.

No way I believe this will slow down in the first place.

C'mon, everyone knows that GPU's have been driven by visual quality in games for many years now. And that really has very little to do with screen res, and everything to do with overall scene depth, object complexity, lighting models, and post-processing effects. To say that since we can now process each screen pixel individually, the GPU is dead is way overstating the case. There are always uses for more graphics oomph.

The root of the problem is more that we've already reached the point of no return in terms of perceived visual quality. If you look at games like Warcraft 3 or Half-Life 2, despite being a few years old they're still compelling and more surprisingly they hold up very well from a visual quality point of view compared to today's titles, on today's graphics cards and also on the graphics cards of then. From a practical standpoint a modern RTS is competing on a level playing field with WC3, which certainly is not true for, say, the original Command and Conquer from a few years before that.

It means that there is very little motivation to buy a top-of-the-line GPU anymore unless you're a competitive multiplayer gamer where 80 fps vs 25 fps actually matters. For everyone else, a mid-range GPU in the $50-$150 price range will do fine. On the other hand, the good news is that people will always need a GPU ;)

Wow, you must not know anything about games, maybe you should leave the "GPU" catagories to someone who really knows what their talking about, damn. I take it you've never played Crysis, and yes, it will get more realistic than Crysis. Nice try at sparking an article that'll get views tho.

You obviously know nothing about gaming(you're on a laptop 1366x768 at that). It's not "resolution" that anyone cares about, it's the "game reality" that's important. Adding "more of everything" into new games to make them seem like realistic environments. This is what gamers care about. Your argument only makes sense if no new games will ever be made- then yes - you can only go so far. BUT look at the reality in Pixar movies and Ratatouille, moving fur hairs, depth of field, shadows, reflections, light perspective, wind, clouds, rain, textures and shadows at a high level- etc etc... games are no where "there" yet, not even close- Now think of the required GPU power to compute that much info -real time - in a "real" 3D world...with one computer(not a hugh aray that Pixar uses) We have a long, long way to go...
What was the point of your article?

What about realistic game physics? Most modern games lacks this. And that could consume vast amount of horse power, so could realistic ray-tracing rendering. Look eg. at the TES4 Oblivion: the grass is still made of sprites (and that's why it's ugly). What I need is movie-like graphics & physics experience in games, but I'm not sure that even ATI/AMD HD9999 would be capable of doing that.

CPUs can't reach 4 GHz, they've said, Pentium 4 is the last step of Moore's Law... Well, things are changing.
Also, let me remind you, that the "holy grail" of graphics is not resolution, or frame rate. The real "holy grail" is ray tracing, and the GPUs are far from being there...
The shader models get more and more complex, new versions of DirectWhatsoever need more and more power. Just like in the field of CPUs, makers of GPUs will have to solve the power issues somehow. Anyhow, the GPU race is here to stay, and this article will be written right next to the sayings like "640K ought to be enough for anybody", along with the greatest misunderstandings of the computer world.

Your article is long winded utter rubbish.

It's abundantly obvious that 3D has a long way to go still. Yes, each generation looks progressively better, but it's still not real.

High res video decoding? Still not here. It's only just with the 8800GT that H.264 can be decoded with minimal processor load.

Then there's the driver issue. Boy, there's driver issues... Nvidia's cards are capable of supporting multimonitor and SLI without fiddling around, non plug and play or fixed frequency monitors, should be able to handle high colour, desktop spanning across multiple cards plus utterly flawless 2D acceleration.

Are any of those true? Nope - they're all broken (except for desktop spanning, which is a Quadro only feature). They can't be bothered to fix the drivers on XP, and are even less bothered on Vista. All so you can get 3fps faster in UberShootemUpWithMarginallyBetterGore 2000. Clue to manufacturers - fix applications *before games*.

That's even before you add increased functionality. Such as built in video inputs and video processing by default, running bits of a game engine (actual logic, collision detection) on the card itself

As many of you don't seem to know, GPUs can't do raytracing, or radiosity, which is what movies use to create lifelike images. The GPU will die and all graphics processing will be done on multicore CPUs.

Until our games include completely dynamic environments, GPUs will still have their place. Think about, do any games really have completely dynamic environments? Can I bore a hole through walls in a game to get to a destination? No I can't.

This comes from the fact that physics software/hardware is very primitive, mostly limited to simply collisions or pretty effects like the Havok engine. It's also because these maps are designed to be statickly lit in order to render quickly. Sure, we have shaders and such, but they're still very much statickly lit.

Come back and tell me GPUs are dead when Dreamworks can release movies to be rendered at runtime.

why did you think in 2D let's think like this : 1920 x 1200 x 1920 !!! shold be nice............

The reason why LCD panels aren't improving is the "legacy OS known as windows" doesn't work fine with anything other than 96 dpi. Vista should have brought resolution independence... but it didn't. So even on a meagre 150dpi screen fonts look small. 

I mean, if nearly nobody is able to profit from 300 dpi, why will LCD manufacturers push for it?. In any case, with 300dpi, the importance of AF decreases dramatically, and this would favour your argument.

There is an interesting range of comments here, most of which are true in some way or another.
But those saying that GPUs are going to go because there will be more things to put on the screen are missing the point : that is not actually a GPU problem, it's a global performance problem. You need the whole PC to be working really fast to deliver more stuff to see - today's high-end GPUs are actually waiting on the CPU for data.
Another comment I have an issue with is people saying that sound is not the same argument. Yes it is actually. Sound card makers have pushed real hard to get to CD-audio quality, then stood around dumb for a few years until Dolby Surround took off.
Now they have a few more years milking EAX stuff until they finally become totally integrated into motherboards and we won't hear about them any more.
Well graphics IS going the same way - be it for 128-core chips or simply discreet chips on the motherboard. And touting scientific apps as an excuse for their continued activity is like saying Nvidia has an iron lung waiting for it - we're all happy that it will have life support until someone unplugs it. Won't change the fact that raytracing or whatever else is, someday, going to motherboard-based.
That day will be when the lowest-end, bottom performer motherboard and humungo-core CPU will be able to put up whatever the app/game/film wants in whatever rez the user wants without any slowdown whatsoever.
Personally, I think discreet graphics will actually be dead before that. Once Intel (sorry, but I just don't see AMD pulling that one off) gets a "platform" spec that can outdo Nvidia's or ATi's offerings, the writing will be on the wall.
And since FSB and DDR specs are set for a massive hike, it might be sooner than one would expect.

[Just in case this comment didn't get through the first time because of a technical glitch, rather than being censored for being dull and rambling, I'll try again...]

Charlie, I think you're underestimating increases in visual quality. No computer game has visuals which anyone would mistake for reality (although there's no doubt that they're getting prettier). Graphics in film scenarios (with a lot of off-line rendering) are improving, but few scenes are completely convincing. Completely computer-generated imagery? We're a way off.

Remember when everyone was talking about how computer games were going to look as good as Toy Story when the graphics cards got faster? They may have got there, but Toy Story, while very pretty and definitely having a "look", is in no way going to be mistaken for reality - even if it wowed the audiences at the time. Pixar may have called their product "Photorealistic RenderMan", but artists are extremely pushed to make anything other than a few frames of abstract objects that are going to be mistaken for photos (of other abstract objects). Every now and then someone mentions how close real-time ray tracing is, and how it's going to make everything increadibly photo-realistic; it's not (at least, in terms of competing with raster graphics for game purposes), and it won't (of itself).

I doubt games manufacturers are going to stop trying to make the graphics ever-more spectacular and realistic. This is an asymptotically complex problem, complicated by an audience who rapidly become jaded by the last level of virtual reality. Correctly handling every last particle in the scene and every path of indirect lighting just requires a lot of computing power; as soon as you have enough, someone will want to make the scene more complex.

Monitor resolutions may be relatively static (my working theory is that the CEOs of monitor manufacturing companies have failing eyesight, which is why the pixels keep getting bigger rather than more numerous and - apart from Toshiba re-badging it - nothing's got near the T221 in the last six years), but that only holds until the technology gets past a stepping point. Start allowing multiple stereoscopic views and you can scale up the rendering demands arbitrarily to try to make the 3D effect smooth. Start rendering holograms and the demands on the silicon go through the roof again.

Eventually we'll hit a limit, probably where it stops being worthwhile for designers to increase the quality of the image further compared with the time they're spending designing the scene (which may, in turn, be when game production costs stop increasing). I can't see us being anywhere near it yet, though. It's true, the majority of customers may be perfectly happy with non-realistic graphics of current games - even Flight Simulator on a Spectrum gave a kind of immersive reality - but for so long as one title is compared to the others on the basis of the visual effect, we've still got an envelope to push. There'll come a time when it's easier to throw work at the graphics card than to spend time optimising it at the game design stage, so I can see a bit of inertia in the graphics market even then.

I've no doubt that there's an end, but I don't think it's in sight just yet.

I think people are forgetting the console bottleneck, is it any wonder cheaper cards can handle higher resolutions when most games are targeting 2 year old console hardware? A decent 8800 will pretty much run anything that comes from consoles these days.

Ray tracing will start the whole game all over again. You just cannot duplicate with shaders all that is done with RT. There is still a lot of room at the top! 
However with ray tracing the GPU and CPU will start to merge more and more which I am sure Intel will be quite happy about. So in that sense the GPU will start to disappear. 
However I think there will always be a need for some external graphics processors since some people will always want more graphics power than others. Maybe games development will move more towards consoles and that could spell the end to GPUs in the PC side (witness the poor performance of Crysis that perhaps demands too much from hardware).

u sound like the guy who said pcs would nv make it to consumer markets...

this article is a waste of space...so many ways to prove it crap

Actually, I'm inclined to think the comparison with sound cards is valid. Consumer level sound cards topped out years ago. There's still technological progress being made on converters, it's still possible to buy a £1000 sound card, but the basic sound card is pretty static.

What really limited the sound card quality was speakers. They stopped being premium items when every last john-jack one of them was good enough for the average computer speakers. The comparison with monitors is actually quite close. (Incidentally, if you think even a great set of 5.1 speakers is comparable to real-life, you need to go hear an orchestra play live.)

Equally, Crysis' release has rather confirmed the main argument. Its affect on hardware sales looks to have been slight at best. Try the following experiment: take an ordinary member of the public and show them doom and quake. They'll be able to instanty see the improvements. Try the same with Half-Life 2 and Crysis.

The most impressed my GF ever looked at graphics in a computer game was the train at the start of Half Life 2. "Finally" she said. "It actually looks like a real place." The advances in gaming visuals are going to have a lot more to do with artistic composition and a lot less to do with pixel pushing.

Maybe the increase in GPU power will allow a move away from polygons into something closer to Adobe PostScript. Isn't it time to do something so that a curve doesn't look bumpy? We'll never have virtual reality if they can't draw a simple circle.

The author is making a really good point about the limits of what the eye can perceive. At some point, there are diminishing returns. His projections are for a 1000 fold increase in todays processing to meet rez, AA/AF and polygon requirements. That is pretty much going to be realism.

The three major CPU/GPU companies are all working hard on the replacement to discrete GPU's. AMD/Fusion, Intel Larrabie and Nv is looking to acquire x86 licensing. Can you imagine an AMD/Intel GPU running at CPU speeds?? Crysis does look pretty good now - add some CPU power to address the physics processing it wants, and then draw those polygons and it will rock. But that could be done with a 2x or 4x increase in GPU processing. Part of the Crysis problem is the NV8800 family is borked on AA, its a design problem.

Maybe raytracing will throw a curve into this - that just shifts the same slope later in time.

It was a good point

I don't trust the Steam survey you cited here. When I ran the Steam survey on my PC, aside from the fact that it pegged my 3200+ processor at 2.199Ghz and refused to install anything requiring a 2Ghz version, it happily detected my 20" LCD as a 14" CRT. I have to wonder how many other mistakes it makes and how many people fail to check the information before they send it on. There's no way to amend the information that is collected nor to leave a notation of the errors, so the system has a rather significant fault.

The comment about ray tracing is spot on. Once games can perform real time ray tracing at the target resolutions and frame rates it's all over for the GPU industry.

That said, the IBM Cell BE chip comes remarkably close to managing this particular trick. It has mammoth FP performance and yet is sophisticated enough top be more than an FP accelerator.

Should be interesting to see whether Intel takes some of the lessons of the Cell design to heart. Imagine an Intel Nehalem paired up with 8 or 16 SPE style units with a fast on-chip interconnect. GPUs will be relegated to what they are actually good for, displaying pictures. Perhaps they will retain some use as scalers and anti-aliasing machines, but once the SPU has the grunt to ray trace in real time, it's game over for the big GPUs.

Regardless whether we're still rasterizing our graphics or using ray tracing in a few years, there's a lot of work to be done to make more sophisticated lighting models practical for interactive rendering of large datasets. Ambient occlusion lighting is just beginning to become practical in real-time now. I'd argue that we already have mostly-adequate resolution, but that we're still using overly primitive lighting and shading models. Indirect lighting models have computational costs that scale worse than linearly with the number of rendered objects, in some cases much worse. In order to continue achieving interactive frame rates with large datasets, we'll probably need to tack on a couple more orders of magnitude to the performance requirement stated in the article. Beyond that, the article ignored the fact that GPUs are beginning to see use in many non-graphical calculations, whether physics, signal processing, compression, encryption, etc. This is an area of significant opportunity for the GPU vendors to grow into. There may be a few "killer apps" waiting to be discovered yet...

But first, a few comments:

"Is there a grand challenge? Sure, it is frame rates, ..."

Hell, no. (Do you feel like you really want higher frame rate when watching 60 Hz TV?)

The grand challenge is to make things like hair in the wind, snowstorm, facial mimics etc. look real. REAL. And I think days when we'll be no longer capable to tell (never mind resolution and fps -- on a TV screen!) if we're watching camera footage or rendered animation of anything above are definitely not in the near future. 

Plus, display as we know it is damn too far from what I'd personally love to gaze at. Think if instead of looking at a panel in front of you, you're surrounded by a hemisphere and can literally look left and right, up and down. That slight movements of the head affect the picture, revealing perspective. That sharpness of the scene changes as you focus your eyes either on close or remote objects. Or, better yet, that in order to discern objects, eyes must focus at corresponding distance.

"So, the problem facing GPUs is that in a few years, they will catch up to monitors,"

I think the main problem GPUs face is if gamedevs suddenly stop taking care of making GFX in their games ever heavier, it will suddenly make no sense to buy a new or more expensive card to enjoy higher framerate and/or better visuals in a new game. Much as anyone would love this to happen, I seriously doubt it ever will. :)

"Remember what happened to the thriving sound card industry?"

Because percentage of people who would pay more for multichannel/3D audio or higher SNR was way too small -- most were happy with headphones or two speakers. And when there's nothing consumers either care or are FORCED to care about (like said frame rate or visual quality in games) and lots of competition, well, there's one way to go. 

HOWEVER:

as for GPUs being a dying breed -- I would agree, but for a completely different reason! 

Radical reduction of comm latency enabled by putting GPU and CPU in close proximity (or even extending x86 with relevant insn set) will give rise to widespread use by gamedevs of a bunch of tricks doable even now, but totally useless for GPUs sitting on expansion boards. 

So yes, I do think GPUs as we think of them today -- a plug-in card, that is -- are definitely threatened by GPU-CPU convergence, but IMO things like "monitors and eyeballs having their limits", resoluions and frame rates have very little to do with it.

AMD has already started working on the technology in its Radeon HD 3000 series, according to sources at graphics card makers. The first product will be the Radeon HD 3870 X2 which will feature two RV670XT GPUs and will launch in January 2008 with a price set between US$299-349, noted the sources.

The Radeon HD 3870 X2 uses a PCI Express bridge chip, the PEX6347, to let the two GPUs work together on the card

Just as I mentioned X2 3570 last month, Whoopes, there it is on schedule. obviously if you want everything on constantly there would be no switches. So just put whole bunch of dem X2 cards on machine sli quad with quad core integrated of ye time & POW, Right in Face, Kisser. KAPOW. BAM, ZAP. just use wise setting & GOD Will appear. Also crossfire part is being incorporated latter into GPU 700 itself, so no extra outside crossfire is needed at all.
thomas s von drashek

"nobody will ever need more than 640k!"

This article is entirely based on a premise that is wrong. The fact is that the games will never stop its evolution. After Crysis we'll have FarCry2, and then Doom4, and then Half Life 3, and so... There's A LOT of thing to improve, and even thought we finally reach the 'reality look', who tells that reality is the end? Virtual world will go much further, just like some movies. ;)

want to know why 640x480 pixels of the proper source material looks better then your latest game at 2400x1800???, i.e. the grand canyon, or the pyramids, ......................it's because those 307,200 pixels (640x480) were derived from the real world, which compared to the world a video card can create, has infinitely more resolution and color. When the "then current" video cards can approximate reality, then people will put off an upgrade. I think the author is smoking some left over lawn clippings.

Fresh from the channel armada! Ol' Chap, might I sell you an Amigo 64? It is a virtual necessity for the halographic 2nd Life, and comes bundled with 3D Pong! You'll need a S-ATA 6 Gb/sec bus to sport it!


First off, there's only one "o" in the word "loses". Looses passes a spelling check, but not a grammar check.

Anyway, GPU's are starting to top out in what they need to do graphically. But they're still great, massively-parallel co-processors. Why not throw some physics problems and such at them? I'm pretty sure that's the way the world is heading, start having multi-core vector style co-processors to do various jobs that a general CPU won't do very well. Right now it's primarily graphics, but in the future it could be a lot of acceleration needs, and not just for gaming. AMD made a good purchase with ATI, it'll just take a few years for it all to shake out.

Just like how netburst got us 20Ghz computers that have ended cpu development?

(p.s. don't forget a doubling, because of need of 120+ FPS on Crysis 2012 for "3d shutter glasses" and 'real' 3d effects. Then double it again if you don't want a headache after 30 min of playing) so maybe around 2018 (add some years for power efficiency and price) Or 2020, to work the bugs out of multi-chip sli etc drivers for it.. Then make it work in a 120F environment (global warming by then)... and then add a few years for ... ... ... then Factor in how long release cycles will become after AMD folds. so add......

"There is nothing else that can be invented, everything that can be invented has been invented, it's time to close the patent office" U.S. Patent office 1889

Any family relation?

I am surprised to find this article on here, the technical variables chosen are narrow, and the market direction is misrepresented.

this guy has no clue what he is talking about. he just wanted someone to read it. waste of time so don't bother reading it. also i have a feeling you are possibly stupid nothing ever stops, NOTHING. they might slow down buy stop they will not. even if we get movie like cinematic games something new will come out and something will need to support it therefore newer tech needs to be released. i have a feeling if gpu`s stopped the world as we know it will change drastically.

You forgot to discuss giant size HD monitors, such as a 100 inch model that fills your living room wall. We could be decades away from a GPU that could handle 32 bit color, 2560*1600 rez on a 100 inch monitor at 60 frames per second.

The best graphics cards @$500 have barely playable frame rates on todays 30 inch monitors @ 2560*1600.

Intel only makes integrated gpus. Intel does not make discrete gpus. So, Intel does not push the gpu industry.

Just this week ATI reported that they will not be releasing the R700 GPU until late 2009. So, ATI is not pushing the GPU industry all that fast either.

The GPU industry has room to grow for at least 10 more years, probably more.

Until looking at a monitor produces an image that matches looking out a window I doubt graphic card vendors will ever be out of a job.

Graphics need more power all the time since we can easily get more demanding systems for GPUs to handle. Like modular screen with 9 or 16 units of 1920x1200 screens with 1-2 mm borders attached together or a quad version of that 5 yr old 9,1 mpxl screen that IBM and Viewsonic used to sell for 35 kUSD (and is now available in Japan for much less money). That would be around 36 megapixels, and then you might need more GPU power than any system can handle for next few years. Do I need one? Hmm, maybe, at least I want one (for work, that is;-).

Well, as soon as the monitor manufacturers come up with a new way to jam more pixels in the same place the GPU manufacturers will have something to look for to. I believe the human eye will lose clarity of the pixels at about 7000*4000 on a 21" monitor. I remember that figure from somewhere. That would completely eliminate the need for AA and put more focus on pixel pumping. Also, texture. One thing most things lack is something truly like texture. Monitors can use light and dark to mimick texture very closely, but it's very hard for a monitor with a glossy finish to mimick tarmac, rock or unfinished rough surfaces. Also, as soon as they do reach the physical roof that is the human eye, they can work on a true to life physics engine. Heaven knows, that could take many years of developement to get somewhere. They could almost spend as much time on that as they have with graphics. As much as we don't want to think so, physics engines in computers still have a far way to go.

I think we are just begining to develop the 3D viewing and that will make a hell of a new arena and power need.

There's still progress to be made. Remember the 3D monitor reverse engineered in the Bennifer movie "Paycheck"? What about Halodecks in Star Trek?? I'm sure hardware requirements for these will be exponentially higher than anything we use today. What happens when we establish direct brain connections. Computing will continually get faster and more efficient, and with that, endless possibilities. GPU cards won't die anytime soon. They'll just morph into something better.

When we can do reflection, refraction, diffusion and diffraction with sound and light with 1024 (or more) source and destination points at +1080p with +16AA/AF +128bit color enabled, in real-time (+60fps) and in addition, sub-pixel physics (fluid ie blood/water) with fully destructible and interactable environments and human level AI. Now put all that into multi-player online, over fiber optic cable and last but not least, make it bug free and stable. Then no... we have not hit the limit for gaming/entertainment/simulations technology. 

Imagine for a second, the amount of transistors and RAM needed to perform and store physics calculations for every grain of sand on a Crysis beach for the entire length of the game, just so that when you run on it, the sand/grass displaces foot prints and mixes with your blood for your enemies to follow and the system never removes those “decals” after you turn the corner for a few mintues. We are talking Exabytes of RAM and processing power here, on top of that, the AI would have to be running footprint/blood recognition software in the background to filter out what are rocks/mud/dents and in the meantime calculate the navigation path around obstacles etc etc. 
Rendering those pixels is the least of our worries

In the long term, general purpose game/simulation acceleration hardware technology is where we are headed; whatever you want to call it CPU/GPU/PPU/DSP. Technical Resources and Money are more of a limitation than are imaginations and public delivery schedules.

I've had this discussion with a lot of folks. In particular, everyone vastly underestimates the amount of progress there is yet to be made in terms of scene complexity. Yes, the human eye can only see so pixels and colors to a finite resolution before the distiction is relatively meaningless. However, our current ability to reproduce realistic scenery is staggeringly inadequate, and realistic interactions within those environments are even more depressing.

There will come a day when we'll be doing full scale modeling off massive systems of small scale components. Imagine fully interactive, mutable environments capable of rendering thousands of models down to the dirt under their fingernails all taking a swim in a live fluid dynamics based water model. Walls that, when destroyed, actually explode into millions of individually represented particles that discipate into and interact with the rest of the scene. Racing games with actual gas models that calculate the entire molecular content of the atmosphere to define turbulent drag across the body while simultaneously handling collision detection and deformation down to the millimeter of every panel and screw on the frame for a field of 50 vehicles while 40000 fully modeled spectators jostle about in a live collision model amongst themselves in the stands with equal graphic detail.

We are a LONG way from where we can and will be, and even then, there will be plenty of room to move.

We need processing power, and TONS of it, and the graphics vendors will have their hands full keeping up. Believe me, the moment they feel like they've outpaced the industry, we coders will pin them to the wall, you can be sure of it ;)

OK, this article is so far from the truth it hurts. I know almost nothing about graphics cards, but it only takes a 4yr old to tell that games look and play nothing like real life. I mean...we're not even remotely close yet. I can't wait for the day games have the same visual realism as the first Pixar movies....

Bring on the holodeck

BillG: no one will ever need more than 640kb.

Charlie Demerjian: GPUs are dead.

One of your main sources was valve's steam, now alot of people who use steam play counter strike and the norm for any CS player is a crt monitor because of its refresh rate. As for the gpu part your dead on saying their will be a point where it will life like. But i dont seeing that anywhere in the near future. i havent seen a good fabric model yet on a game chariacter. The games arent even close yet. Not even crysis comes close to real life.

Although i respect your thoughts mr Charlie Demerjian, i do not agree. And as you see noone does. Your ideas seem stuck to resolution and color depth which is not the case. This is a simple matter of fillrate. What about the raw power needen for realistic model? I will point out the transformers that a friend above mentioned. When we will be able to see graphics like that in real time the we will be ok.Lets also not forget physics. We need huge amounts of power in order to emulate a collapsing building for example. So imagine a bomb devastating a block. That would be interesting. Although it is not clear who will finally adopt and evolve physics (cpus or gpus) both companies (nvidia and ati) have shown much interest. So dont worry if they will need new business they will simply make us feel that we need it.

AS for microsoft and dx10 i can say they can shove it. It is 66% slower and the result is the same! Jesus the must really think we are THAT stupid/blind.

Baka ne. Obviously you don't have a clue what you're talking about. Image fidelity expressed in bpp and screen resolution in Mpixels are a necessary but not sufficient condition for achieving photorealism. And photorealism, make no mistake about it, is the Holy Grail of computer graphics, be it real-time or off-line. Seeing how primitive hardware accelerated rendering is in achieving photo-realistic robust algorithms, your feeble misguided argumentation is an exemplary piece of journalistic ignorance.

(for the writer of the article)

Don't you read "the inquirer"?

we are still far from "ray-tracing", physics are still begining to be developped. I may be wrong, but i guess that today's graphics industry will invest their resources on that, if, for some reason they don't progress anymore.

Resolution and number of monitors hasn't been the issue for a long time now... Multimon support is only of interest to a fairly small fraction of the market.

For the longest time now, and ongoing - the issue has and still will be - framerates and pure number crunching power. You tried to play Crysis any time recently?

Whilst I'll concede that there has been a slowdown in new engines of late, of which most semi-decent graphics cards will happily play maxed out - engines that push the graphics cards will continue to be made and yet ever more power will continue to be needed.

Resolution and number of monitors hasn't been the issue for a long time now... Multimon support is only of interest to a fairly small fraction of the market.

For the longest time now, and ongoing - the issue has and still will be - framerates and pure number crunching power. You tried to play Crysis any time recently?

Whilst I'll concede that there has been a slowdown in new engines of late, of which most semi-decent graphics cards will happily play maxed out - engines that push the graphics cards will continue to be made and yet ever more power will continue to be needed.

You only need 640k of memory - Bill Gates

Just think of that back when he said it and now. Now you see 16 GB RAM on 64-bit systems and 64-bit Windows would support over 100GB of RAM.

You are an absolute retard. Who in the world allowed you to even start to write technology based stories. Get out of the industry go back to Jack in the Box. Goodness do you really think GPUs are about nothing more than resolution and color depth? If a GPU were responsible for playing movies sure... but the GPU is about RENDERING not playback.

Never let this guy write anything ever again, please!

this seems like a very simplistic view. the author is completely ignoring the game itself and increasing complexity in the game world, which translates to higher polygon counts along with the more complex shading algorithms. can we currently render let's say a city in extreme detail? we can't even model human hair or fur properly without faking it. most games don't have fight scenes of 100's of people in extreme detail. so in short i can't currently see a limit for the polygon counts and our ability for perceiving them. and that's FAR from ultra-realistic currently.

the more likely bottleneck will be the humans creating the games. as the games become more realistic it takes longer to create the super detailed models and game worlds. that is a problem that's much harder to solve.

It's not about the pixels, idiot, it's about creating a realistic world to shove into the pixels.

I can feed you a perfect rainbow at 128 bit, 16000x9000 pixels, 120FPS RIGHT NOW with custom-created tech. That doesn't make it as convincing a world as Quake at 320x240.

The polygons (or voxels, or shaders, or rays, or anything) are where the exponential increase in GPU power has been, not the PIXELS. Yes, this requires game development - you suggesting that it doesn't exist?

"Think of it like this, if an ATI HD3870 can play most games adequately at 1920*1200, AKA 1080p, right now, that is fine."


1080p is 1920 x 1080, and 1920 x 1200 is WUXGA.
Just so ya know...

Ray tracing will no doubt be brought in, even if it does little to image quality, and this will require a lot more processing power. Also remember that GPU's are approaching the limits of conventional semiconductor size, as CPU's have already begun to, tricks like the high-k gate can only work for so long.

People who aren't qualified should not be
allowed to write articles on subjects they're
not qualified to write about.

My whole career has been in CG.
Dude, you're actually trying to tell me that
CG has come close to reality and there isnt
that much more to do? geez. We're not 
even at the quality pixar movies renders
in their movies much less at photo reality.

There is decades of work left to make
synthetic images look like real life. 
pixel depth and resolution are not everything.

Lighting/textures/motion physics/object
complexity have a long way to go as do
frame rates at the complexity we need.
antialiasing has a ways to go especially
texture aa. Edge aa automatically gets
better with higher res. Intel is pushing
ray tracing which changes the picture
a littlel by going with ray tracing rather
than raster based graphics but were still
nowhere near rendering realistic images
at frame rate.


man i laughed . first go and have a go at crysis with all settings maxd out .on high res, even with a cple of ultras in sli then come back and tell us u still agree with what ur saying.
im not saying ultras cant handle crysis just not that well .
as a gamer all i can say is WE NEED MORE AND ARE WILLING TO PAY FOR IT. we need better performing gpus. 

As long as consumer demand for more realistic graphics is maintained, and as long as the requirements for more realistic graphics demand an increase in performance, chipsets will continue to be developed that push performance boundaries. QED. If you honestly believe that graphics hardware going to flatline any time soon, I have to question how you got this job. As a proposition it is prima facie absurd.

No demand for more realistic sound in games = no great developments in sound cards for gamers. Audio technology for sound production, though, is a thriving and constantly developing industry, albeit niche.

Your maths is screwed as well, but a lecture on that would be a pretty dull affair.

"The world only needs 5 computers" - Thomas J Watson
"640K should be enough for anyone" - Bill Gates
"GPUs are a dying breed" - Charlie Demerjian
(Only one of the above is a genuine quote, but they're equally daft regardless)

Early this year, an 8800GTS 640MB could handle just about everything with pretty high detail levels at 2560x1600 (30" LCD native resolution).

Now we've got Crysis. With minimal AA/AF, 1920x1200, and "High" in-game settings, three 8800 Ultras in SLI are no longer fast enough to maintain an adequate framerate.

A pair of Geforce 9800 series cards (or whatever Nvidia ends up calling the next generation) will probably deliver playable framerates on a 30" LCD at "High" settings. By the time we hit the Geforce 10 series, even running "High" settings with 16x AA + 16x AF on a single card could be possible.

Of course, by the time that happens there'll be newer games out which stress the Geforce 10.

As has been said above, when people can no longer tell the difference between games and movies (or between games and photos, for that matter) then GPU development may slow down a lot. Until then, ATI and Nvidia (and Intel, starting late next year) will probably keep producing faster and faster GPUs.

Charlie you got this all wrong, How long have you been in the IT industry. You may have heard other ppl say the same things about memory, lcd, HD etc b4. You said ATI X1300 low end card can drive two 30" monitors, sure but what new games can you play on them....none. The GPU market is driven by gamers...and gamers alone. One day you gona say tht CPU is dead as well..wats the point of goin over 3.0g right... the CPU market again is driven by gamers and overclockers...read other ppl comments plz. You got it all wrong mate.

I remember reading a very similar article to this back in the days when i had a tnt 2

You missed Motion Blur. The real thing is more than just stupid posteffect current-gen games has. A real Motion Blur requires about 20-30 (or more) frames to build a final MBed frame you see ingame. So for 60fps gaming you need 60*30= 1800 internaly rendered fps. Current high-end hardware can't render that fast even in old dx7 games. So there is something to work on for some time.

I agree that monitors are the next obstacle. I just don't see the res being why.

Over 60% of the people surveyed use a refresh rate of 60Hz. Given the numbers on resolution I assume they are using LCDs.

I use a 21" trinitron CRT. I recently researched the fastest panel for gaming hoping to replace my obese CRT for some clear text and be able to retain my frames for the first person shooters.

The quality of the panel "best reviewed" was horrible. The quality panels that show text and color well, with good viewing angles, are too slow for gaming/movies. The faster panels have horrible viewing angles and the text is worse than a CRT.

Why would the majority of people accept this inferior looking technology for gaming is beyond me. Especially when I can build a superb gaming machine and ruin every ounce of effort by viewing the finished product on an LCD.

For desktop use they are great. This comment was aimed at the gaming machines, the people buying these cards.

Your statement is so wrong that I believe you're just trolling your own group to generate controversy and traffic.

You could only be correct if the cards merely displayed 2D photographs. It's not even about raytracing, photorealism, or resolution; those can be done by throwing GPU at them.

The problem, as you knew very well when you wrote this flame-bait, is that these cards model a 3D virtual world of ARBITRARY complexity.

I've never played a shooting video game; I only do second life. But even with my 7900 GS, I have to tell the system to stop rendering at 64 meters in order to get a playable frame rate. It always looks like I'm on an island, and when I walk, buildings pop up out of nowhere right in front of me.

You can always soak up an arbitrary amountt of GPU because GPU speed increases linerally, but in-world rendered area increases with the square of the max distance rendered. So if you double your render distance, the area modeled quadruples in size, --and the render distance can approach infinity.

And it's even worse than that.

SL models are so crappy because, due to GPU and memory limitations, no curved objects exist there; it's obvious that everything is made of polygons. 

Fixing THIS sorry situation is even more sure to grind any GPU to a halt because as the wireframe grid mesh becomes finer, the number of triangles needed to render a 3D volume increases with the CUBE of the amount of detail, and the amount of detail can be arbitrarily small -- as small as the pixel size.

Finally, for any given scene, it is always possible to add objects to make it more realistic, and the frame rate can be arbitrarily large.

And I haven't even mentioned physics shaders, the fractal complexity of clouds, or rendering individual drops of rainwater and particles of smoke to generate proper behavior.

Yes, there are practical limits to how much work the GPU might have to do someday (for instance, the triangle size cannot shrink below the pixel size), and in a 24th-century holodeck, I'm sure these limits will have been hit. But the exponential nature of 3D models insures that no conceivable GPU in this century will simulate a second lfe scene of extreme complexity out to 100 kilometers using pixel-sized triangles, with all objects interacting realistically.

GPUs a dying breed? Hell, we don't even have anything close to what the Matrix can do yet.

And I'm really looking forward to spending some "quality time" with the woman in the red dress.