so disapointed compare with Octane

[Karonte]

Hi guys, you cannot compare nvidia CUDA with OpenCL.
Cuda is much faster... octane, arion, and much more ... they use CUDA.
Vray, for internal politics, no. It use OpenCL. So it's normal that is slower.
I see that you cannot compare the number of path because Cuda will be always faster and the images with less noise .
Vrayrt is good, in my opinion, not for production but fast previsualizations. For now....

[vlado]

Hi guys, you cannot compare nvidia CUDA with OpenCL.
Cuda is much faster... octane, arion, and much more ... they use CUDA.
Vray, for internal politics, no. It use OpenCL. So it's normal that is slower.
I see that you cannot compare the number of path because Cuda will be always faster and the images with less noise .
Vrayrt is good, in my opinion, not for production but fast previsualizations. For now....

Well, this might change when AMD offers better OpenCL performance

Best regards,
Vlado

[Paul Oblomov]

Here's V2 of my cornell box. Use it. If you want to...

http://rghost.net/3293663

P.S. I always thought, nvidia opencl wraps around cuda anyway

[Alan Iglesias]

P.S. I always thought, nvidia opencl wraps around cuda anyway

...And that's probably why OpenCL may be fundamentally slower than straight CUDA on nVidia hardware. Chaos Group chose OpenCL so that V-Ray/RT users can be free and not tied to a specific hardware company for GPU rendering. The downside of this will be some temporary performance incongruity while developers decide what, how heavily, and when to support standards in the world of GPU rendering.

Unfortunately, nVidia's chief competitor, ATI, does not have OpenCL implementation that is presently usable by RT. That should change and when it does, you will very likely see a very different picture out there. Just give things a little time.

It may currently seem that in the short term Chaos Group is not making the fastest GPU renderer out there, but knowing these folks like I do I predict that in the long term, V-Ray users will have the best in integration, compatibility, freedom to choose competing hardware, ease of use, and yes, speed in their GPU rendering solution.

Just my 0.02$

-Alan

[Paul Oblomov]

Hm. When I compared RT to Arion - I saw a difference in kind of noise they produce. It was the first released working piece of Opencl software that made me want RT so much

And RT was a lil slower, when using meshlight emitters. Vlado promised, it would be faster. One day.

P.S. It would be great, if RT will support AA filters.

[fraggle]

Ok, asap I got the time I will test again.
We bought the 5800 because of the 4Gb of ram, but at he end it's useless!

[vlado]

P.S. It would be great, if RT will support AA filters.

Yep; we will do that. No time for the initial 2.0 release, but it will be added in an update.

Best regards,
Vlado

[Paul Oblomov]

Hello ! Anyone ?

[Paul Oblomov]

You lazy ppl!

[vlado]

So back to the original issue with the 295 card; it looks like there is a problem with the latest nVidia drivers (260.99) and multiple GPUs, including the 295 card. The older drivers (258.96) seem to work fine though - see this thread for a similar issue: http://www.chaosgroup.com/forums/vbu...working-at-50-...

I'm not sure yet if this is a bug in the drivers, or something we have missed - looking into it now.

Best regards,
Vlado

[Jiri.Matys]

I will try 258.96, but please inform us about this.

[vlado]

I've done some changes to the code now for the release and it runs fine with the 260.99 drivers; I still think it's a bug in the drivers, but there is no indication when it will be fixed. In any case, from my tests, things are running fine on multiple GPUs with the new changes...

Best regards,
Vlado

[Paul Oblomov]

Ok, asap I got the time I will test again.
We bought the 5800 because of the 4Gb of ram, but at he end it's useless!

Any news ?

[fraggle]

Any news ?

We gave up on RT GPU for now as it slow down the process more that anything else. We just waste money with the FX5800, thanks to fake benchmark all over the net. What may works for exterior scene doesn't work for interior one.

[vlado]

The Quadro FX5800 is quite slow, it has only 240 CUDA cores (compared to e.g. GeForce 580 GTX, which has 512), and I'm not sure it's a Fermi architecture.

As for the general slowness when using V-Ray RT with a GPU, it might better to keep one GPU only for handing the user interface/viewports, and other GPUs for actual rendering; or another machine altogether through DR. Some of the interactiveness can be brought back by tweaking the ray bundle size/rays per pixel options as described here: http://www.spot3d.com/vray/help/rt10...m#load_balance

Interior scenes are certainly harder to compute, and you'll need a lot of computing power - see for example this thread: http://www.chaosgroup.com/forums/vbu...h-V-ray-RT-2.0

Best regards,
Vlado