OTOY Forums

BorisGoreta wrote:Thanks for your report. Problem is that even if I set the global CUDA options to use only 1 GPU I still get OpenGL 1.1 which are only some basic OpenGL functions. When you boot into windows please check your OpenGL version.

Do you have regular nvidia drivers installed or some CUDA special drivers ?

CUDA swithces work for CUDA but have no effect on OpenGL.

I don't understand how can your system have good OpenGL with so many GPUs.

Hi BorisGoreta,

here is a screenshot of the OpenGL Extension Viewer after startup with all my gpus "online".
The first GTX580 is the one in my PC, all others are in the turbobox. As you see it shows Open GL Version 4.3.
I have only normal NVIDIA driver installed, no CUDA SDK or similar.

The program seems only to detect the first GTX580... so only this one might be in use for Open GL?
What does your extension viewer show up when all 6 cards are active? How about the render test?

Here is the render test of mine... comped two screenshots together to show it only used the GTX580... at about 30% load, getting about 1200 fps average with Open GL 3.0 - 4.2.

This is what I get with SLI off. Just OpenGL version 1.1 which basically can't do much.

I really don't understand how can you have good OpenGL. It must be something with the mix of cards you have. You're lucky !

And this is what I get if I disable two cards. You can see in device manager that 2 cards are disabled. Now I get proper OpenGL version 4.3.

BorisGoreta wrote:This is what I get with SLI off. Just OpenGL version 1.1 which basically can't do much.

I really don't understand how can you have good OpenGL. It must be something with the mix of cards you have. You're lucky !

The attachment Capture.PNG is no longer available

So there is no GTX listed in the dropdown menu, you only get GDI Generic by default (which is basically no hardware acceleration)? Maybe it has to do with the Titans layout? Is it somehow limited in terms of Open GL?
Or it is a limitation of your CPU or Mobo, I have read about PCI lane limitations of CPUs lately, but did not understand fully how it works. As I think, the CPU or bus system might drop down PCI Express 3.0 or 2.0 to 1.0 or even slower when having too much physical GPU cards installed on one board (too much PCI Express lanes used)... and maybe then Open GL cannot work proberly?
I have 5 physical graphic cards with 8 GPUs, you have 6 physical cards with 6 GPUs... maybe it makes a difference, how much PCI Express Slots are used?
Do you have the same problem, when removing 1 of your 2 Titans in the PC, leaving only 1 Titan in the PC, and 4 Titans in the turbobox?

Did you also check which PCI Express Slot of your mobo the turbobox expansion card is installed to? Is it 16x, 8x, or maybe less? (Not all PCI Express Slots are the same). Did you try different slots?

Is someone using Octane on Linux?
I'm curious if the performance with the latest proprietary Nvidia Linux drivers is worse than with the latest Windows drivers.

ChrisVis wrote: here is a screenshot of the OpenGL Extension Viewer after startup with all my gpus "online".
The first GTX580 is the one in my PC, all others are in the turbobox.

Hi Chrisvis,
What mobo do you have?
Seeker

(GTX 690) + (GTX 690) + (GTX 650 Ti) The 650 is used as the display card.

When all GPUs are been used I get a score of 17.22 Ms/sec Path Traced

nice setup!

finally this theme got back on track =)

Thanks SamQior for posting your rig setup, I had been searching long ago for someone that actually had their display card on the last pci-e slot. I only had found references that it should work albeit a slower bus speed and no longer direct connected to full speed to cpu, but all I could find were bios options for onboard vid or PCI/PCIE option but would only go straight to first pcie slot. doh! I had to connect both cards at the same time to a monitor and change it in the display setup haha brainfart, I kept only 1 DVI plugged in before at bottom and it wouldn't work back then on bootup (figured it should detect which card had the monitor plugged in).



My earlier watercooling hardware dry test setup where I had wanted my main display card on bottom and the render gpus's on top.

PCIE 1 GTX 580 3GB (GPU 1)

PCIE 2 GTX 580 3GB (GPU 2)

PCIE 3 Quadro 5000 2.5GB (display)

Ended up swapping the cards top and bottom. It is currently a spaghetti mess of power cables at the moment. Just got my nxzt power extension cables to clear that up hopefully will do that over the weekend.

Ahhh yes it wasn't without it's troubles...My motherboard will prioritise the GPUs in the PCIe slots in the order of top to bottom, However I found a work around. Both my monitors have dual input, so I can plug in DVI as well has HDMI. Both 690's are plugged in to the monitors(HDMI) as well has the 650(DVI)...I set the 650 as my Primary display in Nvidia's control panel and disabled the 690's, So when I boot up the 690 is the primary display but when windows takes over it switches to the 650

Nice setup by the way!! how fast is the Quadro 5000 in Octane? really curious about what they can do!