ATI ends the physics argument
AT GDC A few days ago, AMD dropped a bomb on the GPGPU world by announcing a GPU-accelerated Havok implementation. It is technically nothing new, Nvidia teamed up with Havok years ago, but the 'how' side of the question means AMD just won the argument.
The big thing that AMD did was code the Havok implementation in OpenCL. This means that it is a standard, and any other OpenCL supporting GPU, Nvidia, S3 or Intel can use it on their GPUs as well. As long as you have OpenCL drivers, you will have Havok GPU-accelerated physics.
Samurai demo at GDC
There were two Havok demos, the standard train bridge demo and a woman in a red dress for cloth simulation, and one AMD demo. The AMD version is tentatively called the samurai demo, and features a martial artist in a stone square doing some acrobatics with a spear. Ribbons on the spear flow realistically, and flags blow in the breeze.
If this doesn't sound all that spectacular, it really isn't. The Havok demos are pretty standard, and the ATI one doesn't break new ground. The part that is spectacular is that it was all done in a handful of weeks. A quarter ago, this didn't exist, and now you have a full Havok implementation on the GPU, written in a standards-compliant way. It wasn't all that fast, yet, but it did work seamlessly.
Don't underestimate how big a deal this is, however. As soon as it is optimised correctly, you can parse the physics load between the CPU and GPU. If you have more of one than the other, you can still use physics in the way it was meant to be played. Oops, wrong slogan... but this implementation should actually do what the other side promises. The upshot is that game developers can use physics more liberally, they don't have to worry about minimum specs as much.
Correct use of resources, not bluster, ulterior motives and branding, will bring physics to the PC gaming market.
Given what was accomplished in a few weeks, expect to see this for real in a quarter or two, then the game changes for real. µ
Share this:
ShareBusiness Intelligence Software
Customer Relationship Management (CRM)
Enterprise Accounting Software
Enterprise Resource Planning (ERP)
Enterprise Systems Management
And nVidia cards will also support this method as well. The fact that nVidia has brought phsyics into the argument seems to be lost on you.
First you forget Charlie is biased.
And then yessss nVidia has CUDA, but it's so proprietary and closed they didn't even consider giving a license to AMD without a large sum of money.
So OpenCL, everything for free, software that works on S3 ... WAY TO GO AMD. nVidia SUCKS.
BTW, I own an nVidia card. That will change on next upgrade.
nVidia has been slapping it around showing how awesome it is. AMD is all "but you are using proprietary stuff; that's way uncool" and starts coding for the OpenCL (which is totally not the same as OpenGL). That means nVidia *could* dump their CUDA and make the switch but that would be admitting that AMD had the right idea instead of them.
Not likely.
Somewhere along the line nVidia fancied themselves to be the next Intel and started behaving the same way. The Physics/CUDA things is yet another example of such. Nothing like opening a can of whup ass on oneself.
Go AMD go.....
"...you can parse the physics load between the CPU and GPU."
I really really hope you meant "pass", because otherwise that phrase is total BS.
obout time to end that extra fluff on nvidia cards bs
The good part is it is an OpenCL implimentation which means GPU OR CPU can be used. This brings flexibility to the specs for a game. If someone has a crappy old CPU and a recently upgraded GPU, they can run physics. If someone has a beefy CPU but only shelled out 50 bucks for a bargain GPU, they can run physics too.
Yes you can parse the load, so arse to Horribloke.
.... and an open standard particularly with Intel coming forth with Larrabee soon is welcomed. You could run an opencl process alongside directx 10.1 and even program gpgpu's to load balance stream processors to best effect; and it will run on NVidia hardware the same. So perhaps CUDA as a locked doors enterprise became less amazing, but at the same time Stream didn't make any wins - everyone did.
"made it important"?
Hardly.
So you're saying if NVIDIA didn't take physics mainstream when they did, AMD would have first? Who cares if NVIDIA did it first, second or not at all.
Just because, in your opinion, AMD did it "better", doesn't make them "end the argument". It merely keeps physics on peoples minds.
The ONLY thing that matters is the competition benefits the consumer.
The next generation of DX11 cards will all support this INCLUDING nVidia, AMD, etc... I don't get the point.
So...
ATI made a implementation of Physics using OpenCL, so It's FRee, WORKS everywhere else and can use ANY processing resource avaliable (be it CPU be it GPU...)
AND...
nVIDIA, yes, it started it out, brought us the concept of physics, BUT can't make it work correctly, and for that broken implementation it will charge you some dosh so you can use it in your software...
OH YEP. Now I see all the programmers going to the nVidia side of physics... NOT A GOOD IDEA.
But in the end nVidia may end up with a lot of big games under it's physics implementation for one single reason: $. They pay the game developers to do so. ... But will it last for how long?...
In the mean time, go on, get your NEW GTS880!!! The sameone I had back on the 8800... 9800... 9800gtx... 9800gtx+... Whoa, it's starting to make my head spinning.
Open Cl, First published as Oh, Charlie might have broad base, yet in theINQ, wasn't Open Gl 3.0 mention bit back? How could that Monster Have Artificial Inteligence? Graphics Loop 3, like triple charlie, only More Charlies. so Cl is closed loop or Dedicated Loop, Not Limited to Graphics Archetecture. therefoe cpu/pu/gpu &?Yahoo interchange9its female thiang). I Love Chocaolate YaHoo! See , off couch charlie, mambie, well You guess Nam.e drashek
that it was Ageia that started the whole physx/physics thing?
What would make this thing is the ability to offload physics to an integrated GPU while playing on a discreet card, IE 780g/790g and an extra graphics card (which happens to be my current setting). Right now I can use the onboard graphics to hang 2 more monitors on my computer, but it doesn't help while playing (I could do Xfire only with a 3400 not with the 4850 I have) and being platform agnostic I could even use a nvidia card.
So just a question or two that hasn't completely been answered. Does this Havok OpenCL basically remove any PhyX or CUDA from the playing field? GG Nvidia?
It is time some of you posters did a bit of reading.
The importance of the article is that ATI have what appears to be a working "OpenCL driver" for their video cards.
It has absolutely nothing to do with Havok or PhysX or anything else.
Any code that operates on a large set of similar data that does the same operation on each of the data is what OpenCL is all about. Physics is only one small area that will benefit from this. It is also time that people understand that there are a shit load of other physics libraries out there other than just Havoc and PhysX. This article is just that ATI are currently the first ones to demonstrate a working OpenCL drive with an OpenCL implementation of a physics library (Havok in this case). There is no reason why the PhysX physics library will not also at a later date be made OpenCL compliant and can then run on any compliant video card, not just Nvidia.
People ....DO SOME READING BEFORE LETTING THE WORLD KNOW HOW IGNORANT YOU REALLY ARE WITH YOU FANBOY BIAS POSTS....
ATI usually does thing smart, not just fast or flashy, but very smart. This move was very smart.
You guys can't seem to figure out the importance of this move, so here, I'll help.
What ATI has just done is offer an alternative to Nvidia's PhysX that EVERYONE can enjoy regardless of what OpenCL compliant GPU they've got, as opposed to PhysX working only on Nvidia cards.
Guess which solution now would seem the most attractive to developers? Guess which solution will most likely get overlooked now that this thing is around?
Nvidia should take notes from ATI.
Give me a break. Devs could care less about OpenCL or the PC for that matter. What they care about is cross platform support across the consoles and other platforms. OpenCL is a non-factor on the consoles. So if the Havok tools suck ass for cross platform, but have OpenCL support, no dev would touch them.
Devs also don't want their shit to look cookie cutter. that is why they spend a lot of time selecting the right tools for the job. Just because Havok showed cloth running on OpenCL doesn't mean a hill of beans. If all Havok has is cloth, then big fucking whoop. I'll be more interested when someone shows me their full SDK running accelerated on the GPU and providing tangible results to the end consumer.
And even if NVIDIA doens't port to OpenCL, then doesn't that mean that those folks who have GeForce cards could accelerate both Havok and PhysX? That seems to be a win for those folks.
I thought physics was unimportant?
I thought it was only for useless eye candy?
Now that ATI has physics, it's suddenly relevant again?
BAH!
"In the mean time, go on, get your NEW GTS880!!! The sameone I had back on the 8800... 9800... 9800gtx... 9800gtx+... Whoa, it's starting to make my head spinning.
posted by : Erick, 06 April 2009"
Dont forget the 250gtx or the 250 mobile graphics card
You have no idea what you're talking about.
OpenCL is supported by consoles, which means this is VERY relevant to any developer thinking of making a cross-platform game with some sort of physics acceleration system implemented. This is not just for the PC.
Your second paragraph just made me laugh. Shows quite a bit of naivity and ignorance, especially when you think all Havok has to offer is cloth physics. Do you even know what you're on about?
Also, "Devs also don't want their shit to look cookie cutter."?? LMFAO!
Anyway, spare me the crap about wanting to see benefits for the end consumer. If you're supporting PhysX ATM and yet still saying that, then you're full of crap. PhysX has no benefits for the end consumer so far, either. Every game it has been implemented in has been completely useless, barely showed a noticeable difference (aside from a map in UT3), bogged down graphic acceleration enough to make it not worth anyone's while. I'll pick the extra FPS from disabling PhysX over some barely noticeable eye-candy with extra strain on my GPU any day of the week.
Not saying that OpenCL Havok will be any better in that regard, but we don't know that since it hasn't been implemented in a game now has it?
Bottom line, OpenCL Havok is openly available to developers and works on the widest range of hardware, which means no one has to get excluded. Unless Nvidia ports PhysX to OpenCL as well, then they may as well just forget about it altogether because now that there's an alternative, PhysX won't seem as good.
I doubt very much that Nvidia will dump it's PhysX anytime soon and start implementing OpenCl Havok or even a new OpenCl PhysX port into their drivers. As mentioned before it would be like admitting they made some bad decisions (yet again), and we all know they NEVER admit making bad decisions. And even if they do, they usually make extremely poor attempts at rectifying them.
This is clearly a win for everyone BUT Nvidia as far as I can tell. It's still not too late for them to turn around tho.
From the Nvidia site
"NOTE: All binary PhysX SDKs are free for commercial and non-commercial uses*"
From the openCL wikipedia page:
"OpenCL was initially conceived by Apple Inc., which holds trademark rights, and refined into an initial proposal in collaboration with technical teams at AMD, Intel and Nvidia."
And finally, from bit-tech.net
"However, an intrepid team of software developers over at NGOHQ.com have been busy porting Nvidia's CUDA based PhysX API to work on AMD Radeon graphics cards, and have now received official support from Nvidia - who is no doubt delighted to see it's API working on a competitor's hardware (as well as seriously threatening Intel's Havok physics system.)
So enough of this ATI fanboy cr*p, Nvidia are just as involved in OpenCL as AMD, but they seem to have no problem with ATI using Physx and aren't charging for the SDK.
I'd be more concerned that Apple have their claws in OpenCL. They're not exactly the most "open" company around are they.
What matters is what Microsoft dictates. If it is behind OpenCL it will conquer. If it supports a DX style windows implementation, this will prevail.
Either way, CUDA will die faster than Glide, without having ever reached its importance. But in away, Ageia (and Havoc) legacy will prosper and physics will become a reality.
I am so tired of people saying Charlie is biased. Why is it when people merely state the facts it's being biased? We live in a world where facts and undeniable truths are subject to conjecture. It makes no sense it's totally illogical!! Nvidia has done nothing short of deplorable things of late, both to it's customers and not embracing the latest API standards. Nvidia are dying and they can't even got out in style like SGI! By the time they are done everyone will be glad they are gone.
Havok is used by several games and afaik it allows also interaction with the accelerated objects (like bumping a cloth will make it move and give a feedback to the application the objects collided). Nvidia will at some point port Physx to OpenCL and if PhysX starts fading in time nonetheless they will spin it as Nvidia being the "proud pioneer of physics".
Porting it today would be bad as Nvidia cards would loose their "extra feature". Intel on the other side has nothing to loose not having a premium GPU yet. So free Havok for everyone. AMD was smart to play the "good guy game" by porting Havok GPU acceleration to OpenCL (maybe this was even a shocking intel-AMD collaboration) and thus opening Havok for everyone to use.
Wasn't Havok used by HL2 for the physics? I wonder if this "driver" will offload those calculations from the CPU and allow for some stunning mods using the HL2 engine.
I just found the solution for Nvidia: port PhysX to OpenCL with some bloatware and maintain a clean CUDA support as well for N-cards.
charlie - you used to be one of the best writers for inq. once i saw title and your name i already guessed (correctly or not) the contents and skipped reading it. please write about seomthing else. you used to be pretty funny.
And the ATI fanboys and the Nvidia fanboys are at each others throats again without knowing what they are arguing about.
Here is a link about OpenCL and its relevance to Nvidia.
http://www.theregister.co.uk/2009/03/17/opencl_demo/
Please read it so that some of these stupid comments will stop.
nVIDIA also have demoed an OpenCL Physx and OpenCL for Havoc API at various shows throughout the last 12 months. You quite obviously don't have a clue what you are writing about and your pro-AMD bias is appalling and unprofessional.
you nailed it
I think microsoft should tidy all this confusion up in the same way that directx did for d3d and opengl. they could just add it to the next directx version and the physics get done on gpu through directx regardless of what brand of videocard you have. scrap havok and ageia.
at dyno, me too
that this implementation will use spare cycles on CPU or GPU or BOTH.
If Nvidia are working on that, they certainly are very quiet about it.
This doesn't mean that ATI make better cards.
It means that (once again) ATI treats the consumer better thatn NV does.
link?
the point isn't that Nvidia is behind Ati, in fact it is ahead of Ati in many ways. The point is that opening the Havok engine to OpenCL ruins the "exclusivity" Nvidia has/d.
afaik Nvidia demoed OpenCL running on top of CUDA non Physix on OpenCL or Havok on OpenCL. Could be wrong of course, that's why we discuss to learn more. :D
http://arstechnica.com/apple/news/2008/12/gpgpu-opens-up-with-opencl-1-0-spec-release.ars
"Parse" means (roughly) to read and process some input according to a grammar or syntax. I should know, I've written a few parsers.
The sentence in the article therefore doesn't make sense.
Don't blame me, it's not my fault you're wrong.
"It means that (once again) ATI treats the consumer better thatn NV does."
no it does not. NV treats her own customers better as they get GPU Physx long before ATI customers do.
ATI giving late GPU Havok to all, because they don't have another way of doing it besides opencl. not because they love NV customers.
so if NV customers get Havok & PhysX while ATI customers get Havok only,
do the math yourself.
For those looking for a more balanced view on what this all means, you might like to check out the EA presentation from the OpenCL section of the Khronos developer session held at GDC (should be on the Khronos and GDC sites IIRC).
OpenCL is a designed to enable heterogeneous processing. That is it allows you to target multi core CPU, GPU, DSP or whatever the hell you have that's OpenCL supported from a single programming interface and paradigm.
As with all things there are hidden subtleties to deploying code on these devices even through an abstraction layer like OpenCL, some of which may not be immediately evident in a carefully crafted trade show demo from a manufacturer keen to flog more graphics cards at $700 a pop...
All in all though OpenCL is a good thing so go check it out, but I'd recommend a more educated source in future.
@Charlie... Would it kill you to do some basic research before you write articles you lazy git. The mans got no pride!
OpenCL is a "design by committee" specification that still has a long way to go. There is no platform that supports it at the moment. Apple beta is mediocre at best. Microsoft is not going to support it and any 3rd party which will want to introduce an OpenCL framework to Windows will have a real challenge sharing a GFX card between directx and their OpenCL. I wonder if ATI demo had a single GFX card or 2 cards where 1 is dedicated to OpenCL.
DirectX is "designed by committee" you know! Oh yes, they have a secret little advisory panel made up of industry stakeholders. They don't always listen to them mind...
MS not supporting OpenCL... well that would be an issue. Didn't MS say it wouldn't support OpenGL ES in Win Mobile either and look what happened there... market forces are a powerful thing.
Similarly it probably is a challenge to get an OpenCL implementation to play nice on directx, but wait wouldn't CUDA and Stream already have had to have solved that issue to be deployed now?
I would also expect AMD to be pretty confident of there ability to solve this issue, not just on XP, but Vista and Win7 as well, before showing a roadmap for Windows SDK in late '09 on there slides at Super Computing '08 (okay so the slides they posted on the web had the roadmap removed, but it was on there honest!)...
Nvidia will still have a working OpenCL implementation in a working driver at least a year ahead of ATI I think it's safe to assume, so they win anyway.
My god Charlie is dumb. nVidia demoed OpenCl on their GPUs last year:
http://www.youtube.com/watch?v=r1sN1ELJfNo