Jump to content
3DCoat Forums
jwiede

OpenCL/CUDA for Mac in v4?

Recommended Posts

Andrew & co,

Any word on whether Mac users can look forward to OpenCL/CUDA support for v4? Lack of GPU "acceleration" is the one remaining area where Mac is solidly behind Win & Linux in 3DC, and it'd sure be nice to see that addressed.

Thanks!

Share this post


Link to post
Share on other sites

So, at least at time of V4 release, there is currently no intention of supporting either OpenCL or CUDA on Mac in 3DCoat yet? Oh well.

Is 3DCoat Mac GPGPU-acceleration at least something planned to be addressed at some point during the V4 version cycle for Mac?

Share this post


Link to post
Share on other sites

Personally I feel CUDA has been more a hindrance than anything else. I feel the CUDA builds are much less stable, and prefer to simply use the DX non-CUDA builds on Windows. FWIW, I also use the Mac OS build (and sometimes Linux, to a lesser degree), it feels great to use and is stable, much like the DX builds on Windows.

Be careful for what you wish for.

Share this post


Link to post
Share on other sites

While I don't think he has updated/recompiled CUDA for 3D Coat, Andrew did tell me recently that he did something to CUDA that has made it MUCH more stable in the past several builds. I agree. It crashes much, much less than what it used to. I still wish he'd re-compile it to CUDA 5, as there are some amzing new technologies that NVidia opened up for CUDA and Kepler (or newer) cards (600 series +), that could make a huge difference. I also wish he would get CUDA or OpenACC involved in the Paint Room, with larger brush radius' and large texture maps.

Right now, 3D Coat is not that viable for Game Cinematics or Film work...because it can't handle the big brush sizes and large maps, very well. Getting the GPU involved somehow, would affect a big change in those markets, as many studios would opt to pay $350 for a seat instead of $2k+ for Mari.

Share this post


Link to post
Share on other sites

Is Mari, zbrush, mudbox cuda or opencl accelerated? I'm sure zbrush is not but its fast with its 2.5d process.

I don't like these accelerated gpu software. Some of them are just hype. Its fine to use them in simple parallel processing software like 3d renderers and video encoders(some are crap even with cuda).

But for complex software like 3dcoat its not good. Lots of crash and lots of effort to support a hardware code that is constantly changing and maybe defunct in a few years.

Share this post


Link to post
Share on other sites

Mari is CUDA accelerated, and Mudbox uses the GPU to accelerate both sculpting and painting. Prior to MB 2009, it relied on the CPU. After switching to GPU acceleration, it took the performance crown over night, and maintains it today. Mudobx chews through pixels, large brush or not, like it was nothing.

Share this post


Link to post
Share on other sites

Do they have a way to turn on/off cuda for comparison?

When I was testing some cuda accelerated appz and turning off cuda to see difference, it was maximum 10% difference. Not so big for such a big effort to support such code.

Share this post


Link to post
Share on other sites

Mari is CUDA accelerated, and Mudbox uses the GPU to accelerate both sculpting and painting. Prior to MB 2009, it relied on the CPU. After switching to GPU acceleration, it took the performance crown over night, and maintains it today. Mudobx chews through pixels, large brush or not, like it was nothing.

Incorrect. Mari does not use CUDA. Mari is GPU accelerated, but it does not use CUDA.

Share this post


Link to post
Share on other sites

Whereas I haven't had many crashes on the CUDA build for Windows. That said, I'm not sure I can really compare it to anything else, as it's all I use. The term "many" is relative.

Share this post


Link to post
Share on other sites

Incorrect. Mari does not use CUDA. Mari is GPU accelerated, but it does not use CUDA.

There have been videos co-promoting Mari and NVidia cards, and from the article at awn.com, it seems to imply CUDA usage, although not specifically. Maybe it doesn't, but if not, then I can't understand what NVidia is acutally selling? CUDA is their proprietary parallel processing/compute language, and thus if they are promoting the usage of OpenGL or DX acceleration, they are certainly hiding it very well. In such a case, they aren't offering anything different than their competitor, AMD/ATI. If that is indeed true, then it's very misleading on their part.

Could very well just be marketing speak..."Mari utilizes NVidia technology"....and if so, I stand corrected. But it does seem implied. Imagine if Nvidia did the same thing with 3D Coat, but there was no CUDA implementation. Just OpenGL or DX. "3D Coat is driven by NVidia Quadro cards to offer accelerated performance." That implies a specific, proprietary technology from NVidia...which OpenGL and DX are not.

http://www.nvidia.co...deo-view05.html

http://www.awn.com/a...zation/page/2,1

Share this post


Link to post
Share on other sites

There have been videos co-promoting Mari and NVidia cards, and from the article at awn.com, it seems to imply CUDA usage, although not specifically. Maybe it doesn't, but if not, then I can't understand what NVidia is acutally selling? CUDA is their proprietary parallel processing/compute language, and thus if they are promoting the usage of OpenGL or DX acceleration, they are certainly hiding it very well. In such a case, they aren't offering anything different than their competitor, AMD/ATI. If that is indeed true, then it's very misleading on their part.

Could very well just be marketing speak..."Mari utilizes NVidia technology"....and if so, I stand corrected. But it does seem implied. Imagine if Nvidia did the same thing with 3D Coat, but there was no CUDA implementation. Just OpenGL or DX. "3D Coat is driven by NVidia Quadro cards to offer accelerated performance." That implies a specific, proprietary technology from NVidia...which OpenGL and DX are not.

http://www.nvidia.co...deo-view05.html

http://www.awn.com/a...zation/page/2,1

The reason for missing official nvidia support can only be answered by the "marketing department" at pilgway. I don't know when or whether pilgway contacted nvidia. There can be several possible reasons for the current situation:

- nvidia is not interested in pilgway, because they are not popular enough (Foundry is very popular an makes a good marketing job)

- pilgway never contacted nvidia for support or a collaboration

I don't know. I guess it's not that easy, as we sometimes imagine. Don't forget pilgway is small, and the situation in Ukraine is not comparable with western countries. I think they already do a as much as they can do.

Best wishes

Chris

Share this post


Link to post
Share on other sites

The reason for missing official nvidia support can only be answered by the "marketing department" at pilgway. I don't know when or whether pilgway contacted nvidia. There can be several possible reasons for the current situation:

- nvidia is not interested in pilgway, because they are not popular enough (Foundry is very popular an makes a good marketing job)

- pilgway never contacted nvidia for support or a collaboration

I don't know. I guess it's not that easy, as we sometimes imagine. Don't forget pilgway is small, and the situation in Ukraine is not comparable with western countries. I think they already do a as much as they can do.

Best wishes

Chris

I think it is both, as you mentioned. Size and it seems that Pilgway doesn't even try to communicate with Nvidia. Stas said some time ago, that NVidia was initially interested, but seemed to lose interest and never contacted them back.

I personally think it's probably because:

1) CUDA doesn't appear to make an overly dramatic difference

2) The implementation in 3D Coat isn't broad enough and probably isn't optimized, nor re-compiled to take full advantage of what CUDA can deliver.

There is an app within the CUDA Toolkit that analyzes the application's performance, identifies the bottlenecks and provide suggestions to optimize the code, in order to overcome those. It's called NSight. I doubt Andrew has used this. It's why I wish he would contract a specialist to do this for him.

Share this post


Link to post
Share on other sites

I'm not convinced by either position.

1) Cuda is not making a true difference.

Cuda is not up to date or "introduced" in the right places and it's something that needs to change.

2)Cuda is not making a true difference.

Cuda is creating many of the bugs we see in the geo handling in 3dcoat and should be removed.

I can't decide at this point.

I'm not sure it would be beneficial with those datas in mind for any other platform than windows...

Maybe the solution resides in the update, or maybe 3dcoat's architecture isn't optimal for cuda at all.

At this point I've the feeling the best option would be to completely forget cuda and rewrite the code for a proprietary gpu accelerated engine at east this way when it doesn't work, you can understand why easily. But I talk easy ;)

Share this post


Link to post
Share on other sites

Whereas I haven't had many crashes on the CUDA build for Windows. That said, I'm not sure I can really compare it to anything else, as it's all I use. The term "many" is relative.

It's not crashes that are easy to spot (I can't really blame CUDA on that) but artefacts in sculpting are common on cuda enabled version which is MUCH less frequent on simple version.

Share this post


Link to post
Share on other sites

Yeah, I can't speak to that at all. I've seen artifacts, but I get that in just about every app I've messed with. I solve it in 3DC via a combination of resolution changes and smoothing techniques.

Share this post


Link to post
Share on other sites

No, not talking about that kind of artefact (refresh), but the surface/voxel getting mad. The kind that creates spikes and lines of voxels. It's only found in 3dcoat in my experience ;)

This is not only a visual thing, the mesh gets actually modified :)

Share this post


Link to post
Share on other sites

It's not crashes that are easy to spot (I can't really blame CUDA on that) but artefacts in sculpting are common on cuda enabled version which is MUCH less frequent on simple version.

Interesting. I've never tested it to switch between cuda off or on and see what happen with artefacts. I have always cuda on, because it makes all my scenes extremely faster.

Share this post


Link to post
Share on other sites

×