CUDA, PhysX Are Doomed Says AMD's Roy Taylor

[Someone Somewhere]

Um. You realise both AMD and nVidia's pro cards support OpenCL, with all the pro drivers?

And that turning on OpenCL does not mean turning off CUDA support?

 

[smeezekitty]

Um. You realise both AMD and nVidia's pro cards support OpenCL, with all the pro drivers?

And that turning on OpenCL does not mean turning off CUDA support?

Yes. I think that OpenCL can eventually displace CUDA but it may take quite a while

 

[InvalidError]

You're argument only works if AMD can take a larger portion of the discrete pie.

AMD and Nvidia are not the only tickets in town. You are forgetting the huge blue elephant in the room: Intel's Ivy Bridge and Haswell CPUs have IGPs that run OpenCL quite well so ~74% of PCs being sold come with OpenCL support by default regardless GPU brand or lack of a discrete GPU, courtesy of Intel's IGP. Once you add AMD's APUs, this percentage goes even higher.

As for CUDA on SoCs, that would be only on Tegra5 and beyond so no SoC CUDA until next year. If Tegra5 is about as successful as Tegra3 and Tegra4, we can expect almost no devices whatsoever to get to market and that renders CUDA on Tegra5 moot.

 

[somebodyspecial]

I am sure the professional cards support OpenCL as well.
Wouldn't it make sense to support as much hardware as possible to maximize sales? I think so.

You do realize CS6 already has preliminary openCL support. I can only see it improving from there

You're argument only works if AMD can take a larger portion of the discrete pie.

AMD and Nvidia are not the only tickets in town. You are forgetting the huge blue elephant in the room: Intel's Ivy Bridge and Haswell CPUs have IGPs that run OpenCL quite well so ~74% of PCs being sold come with OpenCL support by default regardless GPU brand or lack of a discrete GPU, courtesy of Intel's IGP. Once you add AMD's APUs, this percentage goes even higher.

As for CUDA on SoCs, that would be only on Tegra5 and beyond so no SoC CUDA until next year. If Tegra5 is about as successful as Tegra3 and Tegra4, we can expect almost no devices whatsoever to get to market and that renders CUDA on Tegra5 moot.

Sorry didn't realize Intel's iGPU could use ECC...Oh wait....So no pro market for them then...Your comment is moot.

T4 just started shipping in July. Not sure how you can call it a failure yet:
http://venturebeat.com/2013/06/03/toshiba-launches-tegra-4-based-tablets-with-beautiful-screens/
2 toshibas with T4 (and a 3rd with a T3)
Vizio has a T4 (and T3) coming also that they showed at CES
Also the Asus Transformer Pad Infinity,
HP SlateBook x2 & an AIO device (20in?) both T4, not to mention
Mad Catz MOJO (they are SEEKING deal T4 they said).
BungBungame's Kalos tablet uses T4 also
Xiaomi Mi3 handset
http://www.phonearena.com/news/Xiaomi-Mi3-scores-highly-at-GFX-Bench-thanks-to-Tegra-4_id45011
Looks like T4 owns the top so of the list right? Xiaomi phone is impressive.
http://www.nextpowerup.com/news/1949/acer-prepping-tegra-4-based-tablet.html
Acer with TA2 coming with T4
Of course shield, and now maybe a tegra tab7 Q1?
Surface RT rev2 coming shortly (I think RT kills anything, but it's another device)
Couple those with Ouya, gamepop, wikipad etc and I think tegra has a pretty good year lined up even before the T4i. Hp, Toshiba, Microsoft, Asus, Acer all showing tegra 4 support.
http://www.phonearena.com/news/NVIDIAs-rumored-Tegra-4-powered-tablet-visits-AnTuTu-blows-the-2013-Nexus-out-of-the-water_id46303
ZTE also has U988S phone coming. I'm sure I'm missing a few devices but these are already coming.

Really not quite sure if all these devices will fail, but they're coming. Your statement is fantasy at best. Can you name that many S800 devices yet? T4 doesn't even have a modem, so imagine the use of T4i when it finally gets certification shortly and is aimed at $100-200 china market etc (expect Q1 devices and announcements by xmas or so probably for those products). These are devices that are going to sell in some quantity whether you like it or not. I expect Google's console to come with T4 also as they'll want to take advantage of the only company pushing mobile games (TEGRAZONE) right now: Nvidia.

Rumor is that tegra tab7 I mentioned is kepler...So Q1 is not far off. We're talking 5 months and they already have working silicon they've demoed with Ira & Island demos. If they told the truth it uses 1/3 the power of ipad's gpu, which would be very impressive and I'm sure used by many devices. There isn't one article on kepler mobile that was not impressed.

 

[Someone Somewhere]

Who said the pro market is the only market.

Actually, yes it can. There are Xeon E3 chips with enabled iGPUs.

Name many T3 devices. Or T2. I'd expect similar uptake. There's several, but they didn't take off much. With the exception of the N7, and I'd guess that would do well with most chipsets. New N7 dropped nVidia too.

I expect Mediatek to get the $100-200 china market. They mostly already have it.

 

[Mate Banchilo]

You say PhysX is not used and UDK and Unity uses physx, just look what games use these engines... As for CUDA, it is more powerfull then OpenCL, and used in some cases for specialized rendering systems like Octane Renderer, VRay RT and others... I think it will continue to be used in nvidia based cloud systems. Even if most of consumer applications go to OpenCL...

 

[Someone Somewhere]

As for CUDA, it is more powerfull then OpenCL

Like to see a source for that.

Many of those examples are beginning to move to supporting both IIRC.

 

[Mate Banchilo]

As for CUDA, it is more powerfull then OpenCL

Like to see a source for that.

Many of those examples are beginning to move to supporting both IIRC.

There are differences between CUDA, OpenCL and DirectCompute... I have been making really simple programs to see what is what, and what is easier to use... Some of CUDAs advantages over OpenCL are templating and math-library...

 

[Duke Wayne]

Blender FAQ:

What renders faster, NVidia or AMD, CUDA or OpenCL?

Currently NVidia with CUDA is rendering faster. There is no fundamental reason why this should be so—we don't use any CUDA-specific features—but the compiler appears to be more mature, and can better support big kernels. OpenCL support is still being worked on and has not been optimized as much, because we haven't had the full kernel working yet.


Roy Taylor, GTFO.

 

[Someone Somewhere]

I'm currently looking at the Blender FAQ, and I see that nowhere. Are we talking about somewhere other than here: http://www.blender.org/education-help/faq/

 

[smeezekitty]

Blender FAQ:

What renders faster, NVidia or AMD, CUDA or OpenCL?

Currently NVidia with CUDA is rendering faster. There is no fundamental reason why this should be so—we don't use any CUDA-specific features—but the compiler appears to be more mature, and can better support big kernels. OpenCL support is still being worked on and has not been optimized as much, because we haven't had the full kernel working yet.


Roy Taylor, GTFO.

That is almost certainly due to the way the code is optimized (or lack of)

 

[Unto Re]

Still, the physx usage on The bureau is just AWESOME, how would they achieve something like that without physx?

 

[renz496]

so physx is dead?

http://physxinfo.com/news/11741/call-of-duty-ghosts-will-be-enhanced-with-gpu-physx-effects/

http://www.geforce.com/whats-new/articles/gamescom-2013-call-of-duty-ghosts-to-be-enhanced-with-nvidia-txaa-and-nvidia

 

[Mousemonkey]

so physx is dead?

http://physxinfo.com/news/11741/call-of-duty-ghosts-will-be-enhanced-with-gpu-physx-effects/

http://www.geforce.com/whats-new/articles/gamescom-2013-call-of-duty-ghosts-to-be-enhanced-with-nvidia-txaa-and-nvidia

Of course its dead, that's why they're ghosts! :lol:

 

[ubercake]

so physx is dead?

http://physxinfo.com/news/11741/call-of-duty-ghosts-will-be-enhanced-with-gpu-physx-effects/

http://www.geforce.com/whats-new/articles/gamescom-2013-call-of-duty-ghosts-to-be-enhanced-with-nvidia-txaa-and-nvidia

Of course its dead, that's why they're ghosts! :lol:

Ha! That's good stuff Mr. Mousemonkey.

 

[Mousemonkey]

Sometimes someone needs to point out the obvious! :lol:

 

[17seconds]

And Call of Duty: Ghosts is going to sell exactly how many copies?

 

[Mousemonkey]

And Call of Duty: Ghosts is going to sell exactly how many copies?

Depends how many cards are sold I suppose.

 

[somebodyspecial]

As for CUDA, it is more powerfull then OpenCL

Like to see a source for that.

Many of those examples are beginning to move to supporting both IIRC.

Supporting both and opencl being faster are two different things. AMD doesn't have the money to fund lots of apps getting opencl and besting cuda support which has been fostered for 7+yrs.

Show me something where cuda and opencl are supported and opencl comes out ahead. Lovers of an opencl world seem to refuse to compare them (tomshwardware included). Tomshwardware repeatedly shows adobe benchmarks but can't seem to figure out how to compare opencl and cuda though both are supported. Same story for video/photo/rendering apps. They prefer fake stuff like basemark instead of just benchmarking photoshop/premiere with say opencl for amd vs. cuda for NV. The same for rendering when they'll show one app on cuda and another for opencl when you can pick one app and use a rendering plugin for BOTH sides to compare. The only excuse I can see for doing this is they are afraid to show cuda rocks. Anandtech seems to do similar crap with home made benchmarks etc for opencl instead of simply comparing opencl to cuda in popular apps.

Octane and luxrender etc have plugins for many popular apps that you could compare (and have been benched by both sites alone, cinema4d, blender, 3dsmax, maya etc) but still toms etc refuse to do it. This is because the whole OpenCL is great argument goes down the toilet the second you see what REAL MONEY does when applied to software optimizations. AMD hasn't done squat really for 7yrs (you can't fund much R&D losing ~6Billion during that time) while NV has constantly put money into cuda and getting schools to teach it massively across ~500 universities.

Let me know when this story changes 😉

 

[somebodyspecial]

Blender FAQ:

What renders faster, NVidia or AMD, CUDA or OpenCL?

Currently NVidia with CUDA is rendering faster. There is no fundamental reason why this should be so—we don't use any CUDA-specific features—but the compiler appears to be more mature, and can better support big kernels. OpenCL support is still being worked on and has not been optimized as much, because we haven't had the full kernel working yet.


Roy Taylor, GTFO.

That is almost certainly due to the way the code is optimized (or lack of)

Incorrect...It is due to hardware that can't do the job according to them:
http://wiki.blender.org/index.php/Dev:2.6/Source/Render/Cycles/OpenCL
"OpenCL support for AMD/NVidia GPU rendering is currently on hold. Only a small subset of the entire rendering kernel can currently be compiled, which leaves this mostly at prototype. We will need major driver or hardware improvements to get full cycles support on AMD hardware. For NVidia CUDA still works faster, and Intel integrated GPU's are unlikely to give any speed improvement over CPU rendering. "

"NVidia hardware and compilers support true function calls, which is important for complex kernels. It seems that AMD hardware or compilers do not support them, or not to the same extent. "

"Note that V-Ray RT at this time also does not support running their full OpenCL kernel on AMD (only an older and simpler version), and Luxrender with OpenCL is also running into kernel size issues when adding more features. So that's a good indication kernel size is the main issue here."

Ok then...You need some new hardware or more R&D to figure out how to get around the issue in software. Translation: Cuda wins. Correct? Straight from the devs themselves.

This is why Cuda/NV own 90% of the pro market. R&D spent on what devs need to get the job done. Also note the jab indicating Intel has no hope...LOL.

 

[sKorcheD]

If they put all this energy into they're CPUs, maybe they would stop getting roasted by intel year after year. Nvidia or AMD for gpu, either a good choice but CPUs, your always 2nd place as long as your AMD inside, says the "industry".