So the guy that screwed up at NV is HIRED by AMD and will now help to make great stuff at AMD that will Kill NV...LOL. Right, so Mother Theresa will now come back to life and start a killing rampage too? Why would she be so different the 2nd time? INQ showing their NV hatred again. I'd suggest the title of the article be, NV sheds fool engineer who screwed up FERMI and is a BRILLIANT move gets him signed up at AMD...LOL.

To the SGI guy, I laugh at anyone who says they think anything about GPU's 18 months down the road. Parity? I'm sure NV's next chip won't be so hot and big (what 2 die shrinks by then?) and 18 months is a long time for EITHER company to screw up, or their partner producer (ahem, TSMC) to screw everything up for either or BOTH as happened this last go around. If Fermi sucks so bad, I'll just leave you with this. How much money did AMD make this quarter? How much money did NV make? Here's a better one. How much money has AMD made in the last 4 years? How about NV? Want another? How much money does AMD OWE in DEBT? How about NV?...ROFL. I don't have to be a rocket scientist to realize who has a SERIOUS upper hand whether anyone likes it or not. Same can be said about Intel cpu's vs. AMD's cpu's. Money, in the end, wins almost any R&D war. NV turned the corner on their bad TSMC chip fiasco 2 quarters ago (as seen in their profits) and are still debt free. AMD's problems will only get worse. They have great cards (I just got my 5850!) but they sell them for nothing and lose money or barely make a few bucks. They need to fire their whole board and get someone that knows how to price a product and make some money. This looks like SGI just trying to get better pricing from NV :) This crap is said all the time by all vendors.

I actually understood your argument completely the first time.

I just disagree.

Ps, Have you seen IBM,s webcast where they show racks upon racks of TESLA servers clustered together. I prefer to base my arguments on actual press release's rather than speculation, rumour, and conjecture.

And yes i Know HPC systems are used to run all sorts of calculations I was merely trying to point out the performance difference between an onboard chip and a discrete card.

I am a system builder who also deals with Nvidia and ATI cards all day long.
And although it is nice if the card runs cool in the first place there are ways to cool them down. Yes some servers have big aircraft fans too.

(Sigh)

As I said before, I work with Brainstorms (very HIGH END realtime 3d graphics for television broadcast)and all but one of my personal computers have Nvidias of various generations.

What I am getting at with my hypothetical HPC rack full of AMD/ATI Fusion CPU/GPUs AND what was the whole point of this and another Inquirer article concerning Nvidia is that as a manufacturer of HPC (High Performance Computing) solutions.... NOT WORKSTATIONS WHICH HAVE SPACE ENOUGH FOR A FIREBREATHING QUADRO OR RADEON....but racks upon racks of 1U or at best 2U spaced boxes chock full of chips that have to be cooled by itty-bitty fans rather than the wind-tunnel sized fans in your custom built Chysis killer box....as a manufacturer, I have to be KEENLY AWARE of total TDP (Thermal Design Power). If I cannot design rack systems that meet customers' TDP requirements (ala the Government and their Carbon Credit requirements) OR I cannot build racks SMALL ENOUGH so I can build them DENSE ENOUGH for the QUANTITY my customers' need (say...Google or some financial firm who needs them for high end algorithm processing for picking stocks)...then I have a problem.

Nvidia with their penchant for making the hottest GPU's generation after generation is painting itself in a corner in the HPC space.

HPC solutions are used not JUST for CGI but for ANY kind of heavy duty calculation of VERY LARGE data sets from data-mining, to simulations, to financial work. EVEN IF AMD/ATI's Fusion solution can't compete by itself with a single Fermi based Geforce card, that's not the point. 4,096 Fusion CPU's with built in DirectCompute and OpenCL capable GPU's WILL ALWAYS outperform a HANDFULL of Nvidia Fermies when it comes to OVERALL NUMBER CRUNCHING CAPABILITY. And on a 1 to 1 basis...4,096 Fusions with say 35 watts of TDP will ALWAYS beat 4,096 35 watt Intel CPU's PLUS 4,096 35 watt TDP Nvidia's (if they can ever get the Fermi down that low) when it comes to building a rack with the lower overall TDP.

Even if you do need a final rendering of whatever data you crunched then throw that to an outboard Tesla box and be done with it. But it still would be all those Fusions doing the calculating grunt work that would allow for a Tesla render.

Anyway, I think I have explained myself well enough. I am not an AMD/ATI fanboi. I just happen to agree with the gist of the article that Nvidia has a potential problem in the HPC space because they continue to NOT address their persistant heat problems.

I should know....I work with Nvidia chips and cards every single day at home and at work.

I guess IBM don't know what there doing then, and forget gaming have you done much cg on you intergrated chip. And as far as linux goes I'd rather have a proprietary driver that works than an open one that doesn't.

No Chris you missed the point of this article. It's about HPC (High Performance Computing) not gaming. Of course gaming rigs don't care, most have primary fans the size of small jet planes to cool hot chips from either ATI or Nvidia.

In the HPC space, power requirements, cooling solutions, and (if it is a government run HPC project) carbon credit calculations are very important. The last point is where the Fermi Tesla GPU supercomputer for Oak Ridge is in dire straights to be cancelled before it is built.

Sure integrated graphics for games suck. But when AMD releases Fusion and after Menju Hedge who founded Ageia and invented PhysX helps customers in the HPC space take advantage of it, then as I said before can you imagine a very dense HPC cluster with 4,096 Fusions CPU's each having a built in GPU, built in HyperTransport memory controller and built in PCIe 2.0 controller.

Now imagine how it could be possible to make AS DENSE and AS COOL (relatively speaking) a cluster with 4,096 CPU's AND 4,096 Nvidia GPU's.

In a word, impossible.

Oh and as far as ATI/AMD and Linux, yeah, ATI's drivers have not been up to snuff like Nvidia. But at least they have open sourced their drivers to let the community help in that process. Nvidia is still proprietary.

Didn't I just read that IBM was going to be using Fermi instead of their own core processors in their new HPC machines.

Ps. I also use linux so you couldn't pay me to run an ATI card.

When was the last time you played a game on an intergrated graphics chip.They may have access to more bandwidth but at the moment there still no competition for a discrete card. And they wont be for a long long time.
And i think you completely missed my point in the fact that I actually said one year ATI has the hot cards the next year Nvidia has the hot cards. For all you know ATI's 6000 series could be real cookers.

Hey man, I'll let you borrow my Quadro 5600fx to light your crack pipe.

Fermi at 28 nm ?!!?!! Maybe in 2012.

Nvidia is fixing Fermi's heat problems NOW....by lowering the number of CUDA cores and lowering the speed.

Nvidia's other problem will be OpenCL. Once that becomes more established CUDA becomes less attractive particularly from a licensing standpoint. Which means a decrease revenue stream at least from that vantage point.

Plus, like everything else in the history of electronics, the name of the game is miniaturization and integration.

For the vast majority of computers sold today have integrated graphics both on the desktop and the laptop. Intel will have a CPU/GPU combo at least in the same package. AMD will have the Fusion APU which will have 2-4 CPU cores PLUS a DirectX 11 GPU in silicon.

What, pray tell, is Nvidia going to integrate with....an ARM core? Oh yeah, that's right they do already in Tegra. One slight problem....Tegra is NOT X86. So once again Nvidia will have an increasing problem soon from an integrated GPU standpoint since they are out of the chipset business AND they do not make X86 chips like Intel and AMD.

Manufacturers already on razor thin margins will look at a BOM sheet and ask, "Why should I pay for either an Intel or AMD CPU AND an Nvidia chip for this motherboard when I can just use an integrated CPU/GPU combo from Intel or AMD alone?"

Less cost, less room taken on the board and less heat.

Yeah, right....this means nothing. LOL!

TSMC is going to 28nm soon and that will fix fermi's heat problems. Not that it really matters though, one year ATI has the hot cards the next year NVIDIA has the hot cards. As long as you have good cooling who really cares.

Since you felt compelled to repeat your post from the story yesterday concerning Nvidia announcing their mobile Fermi GTX480M, I feel compelled to copy and paste my own rebuttal to you.....

" What with Intel falling back to making subpar integrated graphics after the Larrabee bust and with Nvidia more and more becoming a space heater company ("Hey, warm your room this winter AND surf the internet at the same time! Now THAT's multitasking!"), AMD has a golden opportunity to claw back major market share in the integrated motherboard market for desktops AND thin and light notebooks with their Fusion hybrid CPU/GPU. Just from a reduced BOM cost for motherboard makers should be incentive enough much less the thermal characteristics compared to the Fermi NRPU (Nuclear Reactor Processing Unit).

Can you imagine the size of the power brick for the notebook this beast is crammed in? It really WILL be the size and weight of an ACTUAL brick.

HHHmmmm....Nvidia on top...yes, maybe in the HPC space. But they are already losing market share everywhere else.

Could it be because they are a year late to the DirectX 11 game?

Could it be because the manufacturers already know that their chip will be too hot to put in most thin & lights where the bulk of the market is going?

Could it be because Big Players like HP, Dell, and Apple are still gun-shy from "Bumpgate"?

Could it be that as manufacturers you look out a year in advance and see that AMD will have higher performance cards and chips that STILL run cooler not to mention the Fusion hybrid CPU/GPU which will go 32nm next year.

Can Nvidia hope to significantly increase market share and profit by just catering to the HPC crowd? Seems a little "rare atmoshphere" to me.

Don't get me wrong. I work with dozens of Nvidia Quadros in high end 3D graphic machines for sporting events. I have an Nvidia 9500GT PCI card in my old converted IBM server turned workstation because Sparkle makes the highest performance video card there is for old PCI and its an Nvidia.

But damn....every single one is do DAMN HOT! The temp sensors in my servers caused all six fans to jump up in revolutions just to cool the damn thing down. But it beat the alternative in AMD PCI which was a Radeon X1300.

But don't be such a fanatic. In this day and age of increasing energy demands but faster DECREASING amounts of CHEAP energy, Nvidia will have a problem on their hands explaining just why they can't make a high performance AND a cooler, more energy efficient chip. "

If the sales really are 10:1 I assume the bulk of that comes in the low end and potentially mobile markets. ATI has a complete range of 5000 series products while nVidia is only selling pricey gfx boards. A better statistic would be to compare the sales of its 2 products to the similar AMD offerings. I am quite sure the sales will be reasonably close.

@Fermi, first comment. "The record sales of the Fermi-based Tesla show exactly that." I wouldn't call a few thousand cards sold a record sales. On the contrary ATI's HD5000 series is outselling Nvidia's GTX400 series by 10 to 1. The correct description would be a sales bust.
Just yesterday Nvidia lost one of its top engineers for CUDA and PhysX to AMD. Manju Hegde was hired by AMD to bring similar features to AMD's Fusion.

@Fermi, first comment. "The record sales of the Fermi-based Tesla show exactly that." I wouldn't call a few thousand cards sold a record sales. On the contrary ATI's HD5000 series is outselling Nvidia's GTX400 series by 10 to 1. The correct description would be a sales bust.
Just yesterday Nvidia lost one of its top engineers for CUDA and PhysX to AMD. Manju Hegde was hired by AMD to bring similar features to AMD's Fusion.

The article is, badly misinformed and just plain wrong.

If this is the view of CGI, I'm not at all surprised they have gone bust, twice!

It's not only a problem with heat.

First off all: the Fermi's are not three times faster compared to the new ATIs in terms of DP. It's not. Look at the specs. Real tests show it's similar for the high-end HPC cards.

It gets worse: Nvidia has limited the double performance of the desktop GPUs, while ATI has not. There's now a factor 2-3 difference in performance between the two in favor of ATI. So for people experimenting with GPGPU, ATI will soon the best choice.

A second problem is that Nvidia HPC Fermi cards are expensive and the company is unwilling to negotiate on larger quantities.

A tremendous amount arrogance has slipped into their behavior. Limiting performance to force customers to take the expensive high-end cards is a silly route to take if your competitor does not.

I have always been a massive fan of ATI GPU's. For one main reason, they actually work.

The amount of times I've had nothing but trouble with nVidia cards is unbelievable and has lead me to never want to have another one in my PC.

My first attempt with an nVidia card was way back in the days of 3DFX. I bought the nVidia equivalent of the 3DFX Voodoo 3000. I'd seen the Voodoo 3000 and was amazed at how good it was, but couldn't really afford it as I was at Uni at the time.

Within 2 days of owning the nVidia card I'd got totally fed up of the amount of times Windows BSOD just for making a simple change so I took it back to the shop, and forked out extra for the Voodoo 3000. The 3DFX card was a dream, it ran perfectly, no BSOD's, just really good 3D graphics.

Then when 3DFX were bought by nVidia I thought ok maybe they've made improvements I'll try again.

No such luck, yet again BSOD's all over the place, so I put my 3DFX card back in, it was a bit dated but at least Windows would run with it.

Eventually I switched to an ATI Radeon by Sapphire. That died within about 2 months due to serious overheating problems (mainly due to the fact that Sapphire hadn't put any cooling on the GPU), so again I tried nVidia. It worked without any BSOD apart from most games were running way slower than the ATI. Graphics were jerky and really didn't look very good.

Back to ATI. Since then I've stayed with ATI, even going from shop to shop to try and get a laptop with ATI in it rather than nVidia (or even worse Intel).

More recently though I fixed a PC for a friend, and her PC would not play blu-rays to the TV because she had older drivers for her nVidia card. I updated her drivers for her nVidia card to the latest ones, and the screen wouldn't come on. Went for the lower version that would still run blu-ray to the TV. The screen now come on, still however no blu-ray on TV, and now even better the computer wouldn't shut down properly.

Tried to remove the drivers and put old drivers back on, still not shutting down, tried removing all nVidia software and going back to VGA compatible drivers still not shutting down. In the end the only solution I could come up with was to re-install the whole of Windows. A bit drastic for some stupid driver that didn't work properly, but I did and it worked fine.

After doing a search on the internet I found the problem, the firmware on her graphics card is not compatible with later drivers. But instead of the later drivers doing the proper thing and querying the firmware before installing (as would be expected) it just installed them and messed the entire computer up.

It really is annoying because she has now had to go out and buy a blu-ray player for the TV because thanks to this firmware issue she can't install drivers that will allow blu-ray output to the TV. I have got no problems with my computer playing HD Blu-Ray and mine runs ATI, and my ATI card cost half the price of the nVidia. I have also never had problems updating drivers or anything like that. The GPU might be slightly slower than the nVidia's but the nVidia's is pretty useless if half of it's functions don't work due to bad firmware.

INQ hates NVIDIA, and it is kind of oddball how they repeatedly beat on NVIDIA. Let's take a realistic look at ATI/AMD's vs. NVIDIA in terms of using new hardware features: compared to NVIDIA ATI/AMD drivers are still bad, no way near the train wreck 2 years ago.. but still bad.

INQ is a bit of a MS-Windows basher, you would think they would check to see how terrible using ATI drivers on Linux is compared to NVIDIA.

Heck they are a major Vista and to lesser extent Windows 7 basher. Do they know that all graphics features of NVIDIA cards are exposed in GL4 for XP?

I have to repeat that Fermi is 3 TIMES faster when it comes to the all important, double precision floating point performance. And this is comparing the theoretical performance. On top of that Fermi has a general purpose architecture with caches, ECC memory, large amount of registers, CUDA, C++, Fortran, and lots of libraries ported to it. The architecture makes Fermi another 3+ times faster than anything ATI can offer. So Fermi ends up being 10 times faster on general purpose computing tasks. There is no contest. Simply, there is no other choice. The record sales of the Fermi-based Tesla show exactly that. ATI is nowhere to be found in the HPC space.

And yes, NVidia's Fermi cards will be in servers, workstations, desktops and laptops giving them around 10 times more computing power than their outdated x86 processors. Every PC will have one, it's a general purpose co-processor after all. Rejoice, finally we got something faster than x86 in the general purpose computing space. It hasn't happened since the days of Digital.