LuxrenderGPU

[vertigo262]

Thanks dade!!!!!!

it seemed to have worked. But I'm using windows 7 and I no longer get the opencl error. Although now when smallluxgpu loads it crashes. I'm using version 1.1 and maybe it's not compatible with windows 7. If you know where I can download a newer version I'd greatly appreciate it.

Thanks for all the help so far.

[Dade]

it seemed to have worked. But I'm using windows 7 and I no longer get the opencl error. Although now when smallluxgpu loads it crashes. I'm using version 1.1 and maybe it's not compatible with windows 7. If you know where I can download a newer version I'd greatly appreciate it.

viewtopic.php?f=34&t=3458 (but as SATtva wrote, you are going to be unable to use the x1400, you can just run with native threads or AMD CPU OpenCL device).

[Meelis]

WOW 4x GTX 295
It has 8 x 896MB = 7168MB memory, but can the total geometry be splitted between GPU's, soo if i need 2,5GB for geometry now then 896MB per 1 GPU aint going to be problem.
May it speed things up if all data sits in GPU's memory?

CPU and mobo may stay for internet browsing Or not?

@ KyungSoo
How about testing with 64 threats or something like that, or more than 8 - 16, 32
And update period 1-2 + sec.
And rise the ray depth.

[Chiaroscuro]

Indeed very interesting, the unrealized potential is huge; how to keep so many hungry beasts fed. KyungSoo, have you tried bringing in one GPU at a time using the opencl.devices.select SmallLuxGPU config file parameter to see at what point Load starts droping? So, given eight OpenCL devices: opencl.devices.select = 10000000 should be a single one, opencl.devices.select = 11000000 should be two, etc..

[KyungSoo]

Indeed very interesting, the unrealized potential is huge; how to keep so many hungry beasts fed. KyungSoo, have you tried bringing in one GPU at a time using the opencl.devices.select SmallLuxGPU config file parameter to see at what point Load starts droping? So, given eight OpenCL devices: opencl.devices.select = 10000000 should be a single one, opencl.devices.select = 11000000 should be two, etc..

Performance increase related to the number of GPUs.

1 : 1,484K
2 : 2,831K
3 : 3,284K
4 : 3,173K
5 : 3,332K
6 : 3,290K
7 : 3,343K
8 : 3,299K

opencl.devices.select = 10000000

opencl.devices.select = 11000000

opencl.devices.select = 11100000

opencl.devices.select = 11110000

opencl.devices.select = 11111000

opencl.devices.select = 11111100

opencl.devices.select = 11111110

opencl.devices.select = 11111111

[KyungSoo]

WOW 4x GTX 295
It has 8 x 896MB = 7168MB memory, but can the total geometry be splitted between GPU's, soo if i need 2,5GB for geometry now then 896MB per 1 GPU aint going to be problem.
May it speed things up if all data sits in GPU's memory?

CPU and mobo may stay for internet browsing Or not?

@ KyungSoo
How about testing with 64 threats or something like that, or more than 8 - 16, 32
And update period 1-2 + sec.
And rise the ray depth.

I think, the limit of geometry related to the memory size of GPU, will be solved, with composition method.
By feeding portions of geometries to the GPUs and compositing result (alpha) images.

Increasing CPU threads has no performance increase.

And the ray depth is a matter of quality not speed.

[Chiaroscuro]

opencl.devices.select = 10000000

That's interesting... one and two GPUs can be mostly utilized (although it still only says 85%?), but then at three the efficiency really starts to drop (individually; together they plateau) from there on (I wonder if the bus also gets busy?). Can you imagine if all 8 could be mostly utilized, you'd be getting over 12 million samples per second on that scene. Seems like it's going to be a challenge.

[KyungSoo]

That's interesting... one and two GPUs can be mostly utilized (although it still only says 85%?), but then at three the efficiency really starts to drop (individually; together they plateau) from there on (I wonder if the bus also gets busy?). Can you imagine if all 8 could be mostly utilized, you'd be getting over 12 million samples per second on that scene. Seems like it's going to be a challenge.

I think, it is possible, if LuxRay developers move some CPU task to GPU, something like rayBuffer feeding.
By removing the CPU dependency, CPU tasks will be done more efficiently, too.

[Meelis]

I really don't know how to measure quality, but may higher ray depth increase quality in shorter rendering times if it's done only by GPU. Should test with really complex hard to clear scene (very very small hi power light sources, many light sources, many reflective surfaces, many glass objects with cauchy b coeff, or Hi cauchy b coeff, more close to pure white materials than recommended).

[Dade]

That's interesting... one and two GPUs can be mostly utilized (although it still only says 85%?), but then at three the efficiency really starts to drop (individually; together they plateau) from there on (I wonder if the bus also gets busy?). Can you imagine if all 8 could be mostly utilized, you'd be getting over 12 million samples per second on that scene. Seems like it's going to be a challenge.

I think, it is possible, if LuxRay developers move some CPU task to GPU, something like rayBuffer feeding.
By removing the CPU dependency, CPU tasks will be done more efficiently, too.

While this is doable (and I have done it in SmallptGPU, btw, you could give a spin to SmallptGPU2 for a GPU-only Vs CPU+GPU comparison), it is not applicable to Luxrender so it is bit out of my scope. I can not port all the Luxrender code the GPU. Not only it would require an insane amount of time but it would not even work (too many materials/textures/light sources types, etc.). I'm looking for a solution where I can port the 1% of the code (i.e. only ray intersections code) and still have the 99% of the Classic Luxrender features.

Please, note, your system is highly asymmetrical (1CPU + 8GPU), it isn't really a surprise if a CPU+GPU architecture doesn't scale well there. Most users have just 1 CPU + 1 GPU or 1 CPU + 2x GPUs Your system configured with 2 CPUs would produce the awesome results Chiaroscuro was talking about (i.e. 12M samples/sec).

P.S. Are your 8xGPUs installed on the same PCIe 2.0 bus ? All 16x ? 8x ? I wonder too, like Chiaroscuro, if PCI bus performance have some influence. Both ATI and NVIDIA provide a sample application in their SDKs to evaluate PCI performance. It would be quite interesting to patch the application to work with multiple GPUs and to evaluate how the available PCI bus bandwidth changes with multiple GPUs.