swaaye wrote:

It's a complex problem. One cause has been that NVIDIA was the only company taking GPGPU seriously until recently. They had the […]
Show full quote

HunterZ wrote:

Using proprietary APIs to gain advantage over competitors always annoys me greatly, because it's a cheap attempt to lock software developers into your platform.

It's a complex problem. One cause has been that NVIDIA was the only company taking GPGPU seriously until recently. They had the best tools and they are willing to help.

AMD just bullshitted about it and did almost nothing. Also, in reality their hardware was not well designed for GPGPU. Their brand new GCN architecture is much more well rounded for it.

Also, from what I gather, it has been a major bitch to write this GPGPU code, and what runs well on NVIDIA doesn't on AMD. So unless NVIDIA would come in and do it for the developers, the developers didn't really care about it. But now there are emerging programming langauges like C++ AMP and new APIs like DirectCompute and OpenCL (which is similar to CUDA). Standardization will make this stuff more accessible and useful.

Anyway, personally I've found GPU physics to be ridiculous in practice and unworthy of the performance hit to say the least. 😀 Get back to me when it's more than fluff in the air, gaudy particle explosions, little weird balls of water/smoke, etc.

Actually, AMD is the GPU supplier of choice for Bitcoin mining and that is pretty intense where GPU computing is concerned. nVidia video cards don't even come close.

I've read about the AMD Bitcoin phenomenon. Bitcoin is doing a SHA256 hash and AMD has an instruction that helps immensely for it. It also works nicely with the old VLIW4/5 architecture which really lets it scream because the architecture has huge potential if your task can utilize it.

NVIDIA has had much lower peak potential but is easier to utilize fully and has fewer nasty catches. But AMD GCN is a pretty excellent GPGPU chip apparently.

good thread
http://forums.nvidia.com/index.php?showtopic=203078

HunterZ wrote:

3dfx rested on their laurels, overconfident in the developer lock-in power of their proprietary Glide API.

That really was a shame, not just in the sense of 3DFX screwing themselves in the long-run, but depriving the market as a whole from having a good alternative API. At least from what I understand, Glide was specifically designed as an OpenGL inspired stripped-down gaming-oriented API that could have catered very well to a variety of competing GPUs in addition to 3DFX's (cleaner/more efficient than OpenGL or especially Direct3D of the time).

From 3DFX's perspective, making Glide an open (or licensable) multi-platform standard would have had advantages too. Glide may have been adopted as a true standard alongside OpenGL and D3D rather than just a popular proprietary format. There'd be stronger and more consistent overall support from software developers, and while it would have aided competition in 3D game compatibility and possibly rendering performance, the 3DFX's own GPUs would still have the advantage of specifically catering to the feature set supported by the API and thus be more optimized (not to mention their drivers), at least initially.

They also released cards that were just scaled-up versions of previous products.

Scaled up evolutionary developments of older hardware can be a good thing though. You've got potential for hardware level compatibility to cater to any applications/drivers that depend on low-level optimization, then there's the savings in R&D costs, and the potential for still quite competitive performance.
Granted, there's also cases where scaling up is impractical, handled poorly, or not pushed far enough, allowing the designs to stagnate or end up hacked together and cost ineffective and/or buggy.

And in the specific case of 3DFX, the Voodoo 1 to 2 to 3 evolution was rather good for the market demands overall, though dropping SLI with the 3 cut out a market sector the 2 had catered to, among other problems with products moving on from the voodoo 3 in general.

At the same time, multiple competitors sprung up (especially nVidia) and drove Direct3D and OpenGL development until it was on par with Glide, forcing developers to abandon the latter in order to ship games that a wider base of customers could run. 3dfx had over-optimized their cards for Glide, so their weaker D3D/OGL performance then hurt them as well.

Ironically nVidia is now making a lot of the same mistakes that allowed them to beat 3dfx in the first place. Focusing on marketing over hardware innovation and trying to push their proprietary PhysX API around are two of the big ones.[/quote]

kool kitty89 wrote:

3dfx had over-optimized their cards for Glide, so their weaker D3D/OGL performance then hurt them as well.

How optimized?

kool kitty89 wrote:

Focusing on marketing over hardware innovation and trying to push their proprietary PhysX API around are two of the big ones.

There is no alternative to PhysX, it is exactly the area where Nvidia carries burden of innovation alone.

still, how do you guys get 3dmark99 run under 2000/xp?

I ran 3DMark99 MAX under XP last weekend with a DirectDraw patch. I'll upload the file here if you can't find it.

PS: nevermind, just found it again: http://www.hardwareecke.de/overclocking/downl … _99_max_fix.zip

batracio wrote:

I ran 3DMark99 MAX under XP last weekend with a DirectDraw patch. I'll upload the file here if you can't find it.

PS: nevermind, just found it again: http://www.hardwareecke.de/overclocking/downl … _99_max_fix.zip

Downloaded! 😁
I could never get 3dMark99 to work in XP and never knew there was a fix for it

Putas wrote:

kool kitty89 wrote:

3dfx had over-optimized their cards for Glide, so their weaker D3D/OGL performance then hurt them as well.

How optimized?

That quote is from HunterZ, not me. 😉

Anyway, that's a good point too . . . I missed replying on that particular claim, but I ended up including the general topic anyway.

I'm not sure this was a problem at all. It may not have been that 3DFX "over-optimized" the Voodoo GPUs for Glide, or that their OpenGL or D3D drivers were weak, but rather optimized Glide for voodoo (and gaming in general) as an inherently more efficient GUI. Again, my impression was that Glide was more or less conceived as an OpenGL-lite of sorts, cleaner and simpler but less feature-comprehensive than OpenGL itself.

It may very well be that, even with the most optimized GL/D3D drivers, the Voodoo would still fall significantly short of what Glide allowed, not just because Glide catered to the Voodoo feature set, but as it was fundamentally simpler/cleaner/more efficient than contemporary OpenGL and D3D APIs.

This is the very point I was making above with my comments on the potential of Glide as an open-market standard rather than 3DFX proprietary. Having a cut-down, more gaming-oriented API could have benefited many competing cards and offered a more concrete alternative standard than GPU/company-specific APIs. Or, if not Glide, a similar cut-down/bare-bones gaming API developed specifically for the general market.

With a simplified general-purpose gaming API, many contemporary companies may have had much better functioning drivers than ended up for D3D and/or OpenGL. (ATi, S3, Rendition, Matrox, etc, etc)