Originally posted by: josh6079
No, no. Don't misconstrue me. I think I understand how ATi picking Havok physics is contradicting behavior when comparing Havok and PhysX alone.
It's just that I think I can understand their "picking between two poisons" if Havok was willing to give their SDK via OpenCL, especially if ATi is assuming CUDA will be non-existent in the near future.
I'm not saying I agree or disagree with their assumptions, but at least they gave reasons for why they had them based on history instead of sentiments.
And like I said from the very beginning, I wouldn't have a problem if AMD came out and said they didn't want to support a competitor's IP and they felt the horse they were backing was better. That happens all the time in this industry, no problem. Just don't claim you're not supporting it for reasons that are disingenuous and then contradict yourselves by doing the exact same thing you just condemned.
I understand you feel the reasons gave were deceptive, and that's well within enthusiasts' rights. Others feel the same about Ubisoft's reasoning for pulling DX10.1 support. I'm not wanting to go astray into that can of worms; I'm just using it as an analogy of official claims vs. enthusiast doubt.
Heh, again with the Assassin's Creed example. The difference here is Ubisoft's reason for pulling DX10.1 was valid, there were in fact verifiable rendering errors and omissions with AMD's DX10.1 parts. Nvidia has denied influencing the decision in any way and there's really no reason to believe otherwise.
Curiously absent is any statement from AMD denying Ubisoft's claims about AMD's DX10.1 problems, affirming their belief that their DX10.1 path is rendering properly. Also curiously absent is any promise from AMD to work with Ubisoft in order to properly support DX10.1 in the future.
Assassins Creed Controversy Discussion @ Tech Report
So again, Ubisoft's reason is valid based on verifiable facts, whereas AMD's reason for not supporting PhysX is contradictory, hypocritical and deceitful. Big difference there.
But again, had PhysX been available through OpenCL for the aforementioned demo, why wouldn't they have used it?
Probably for the same reasons they didn't adopt PhysX when the spec was ratified in December or in the months since: it didn't "make sense" yet or they were too busy working on their own "closed and proprietary standards".
I think many would agree that, currently, Havok is superior in many regards outside of GPU-accelerated physics. And that advantage is deteriorating with the arrival of OpenCL.
Perhaps, but that doesn't change the fact Godfrey Cheng's statements were contradictory and deceitful at the time when he claimed they wouldn't support PhysX because it was "closed and proprietary".
Yes, I understand that it doesn't need CUDA to function. But it running on x86, PS3, Xbox360, Wii, and etc. are all examples of non GPU-acceleration, are they not?
Beings how nVidia used CUDA to enable PhysX's GPU-accelerated functions, did they not couple both PhysX and CUDA together when offering it to ATi's Radeons?
They were allowing complete exclusion from CUDA if ATi were to use PhysX?
Look, I've already broken it down numerous times to both you and others. CUDA is not tied to PhysX, in any way, shape or form. I don't understand why people don't understand this or simply refuse to. Its like they're clinging to this bit of FUD as if it were true or meaningful, when its not.
CUDA is simply the architecture that includes the API, driver, and runtime libraries that sit between the hardware (GPU) and the application (PhysX). Nothing was stopping AMD from creating that entire HAL with their own API, drivers, and libraries if they wanted to, to suit whatever API or compute architecture they were pushing at the time. Again, there's plenty of examples showing this to be the case with x86, PS3, Wii, XBox 360 support of PhysX using a different HAL by necessity to interface different hardware. This has nothing to do with CUDA, as many of these PhysX solutions pre-date CUDA.
But that's not what bit-tech claimed three days ago. They claimed that nVidia
was coupling PhysX with CUDA as far as GPU-accelerated physics are concerned.
Link
bit-tech:
Nvidia has so far guarded its GPU PhysX technology behind a large CUDA-shaped wall, meaning that anyone who wants to use it has to use CUDA too.
Also,
isn't Havok's physics free? Or does the tool-set lie separate from the source code as far as cost is concerned?
And that's certainly one author's interpretation, probably based on similar misinformation from AMD quotes and interviews, despite all other information and technical resources indicating otherwise.
Also from what I've seen, PhysX is free to use whether personal or commercial, 50k for the source code to port, troubleshoot or customize. Havok is 20k per commercial license, not sure if that's per seat or what, but also includes the open source code.
That just furthers my questioning though. With all of that support for GPU-accelerated physics and TWIMTBP titles being such a large chunk of the recent gaming library AND the G80 < being the dominant GPU for 2-3 years, why have features pertaining to physics remained so static?
Demos from GDC 09 was the first thing I've seen in a while that's been remotely impressing, imo.
Again, I find it extremely difficult to reconcile the fact you ask these questions, yet don't see any fault in AMD's actions and statements over the last 9 months. Look at some timelines and it should become clear why. Nvidia acquired Ageia about a year ago and released CUDA PhysX drivers in July/August 2008, which should give you a reference starting point for GPU accelerated physics.
Also, you can find similar PhysX demos in the Nvidia Power Pack download like the
Nurien demo. Havok's application is undoubtedly more impressive, but the underlying technology is the same and only enabled through the superior hardware capabilities of GPUs used for physics calculations.
That's probably true. Could one also say that OpenCL was necessary because of Havok and it being the primary choice of the majority of developers?
Absolutely not, as the scope and purpose of OpenCL extends far beyond physics or even GPUs. Also, as has been stated numerous times already by Ben, the spec was initially submitted by Apple, who has no direct interest in consumer-level physics or GPUs. You'll also notice Microsoft's support is not-so-curiously absent from the OpenCL standards board.
Oh no, no. Like I said, I can see how their unwillingness to allow CUDA has been a factor in delaying GPU-accelerated physics.
But, I also understand that business is politics, and nVidia was unwilling to cleave PhysX from CUDA when regarding Radeon adoption.
Had nVidia been willing to let ATi go down the "path less traveled" in getting it to function through Stream, whose to say it wouldn't have happened sooner?
To me, these combined factors explain the delay.
Had I been in ATi's position, I would have probably bitten the bait. What Godfrey Cheng termed as "...artificially disadvantag[ing] our platforms in relation to theirs" seemed more like fluff. He really needs to clarify what exactly he meant by that.
Well again, everything I've linked to, including numerous interviews, SDKs and open source downloads, technical documents etc all indicate nothing was stopping AMD from porting PhysX to [whatever API or architecture they wanted to] if they actually wanted to. Instead they chose to support an API that didn't yet exist along with a closed and proprietary standard with Havok. No problem with that, companies do that all the time. But if AMD does that, don't turn around and condemn exactly the same thing they're doing with closed and proprietary standards and claim what they're doing is somehow better, especially when they don't actually have anything to show.
At this point, based on the most recent quotes and developments I'm just glad GPU accelerated physics should see cross-support from both IHV and middleware vendors. But that certainly doesn't mean AMD is going to get off the hook for the last 8-9 months, especially now that they're promoting GPU accelerated Havok as the greatest thing since sliced bread.....just as I predicted they would months ago. I guess they're "ready" now and it finally "makes sense".