OTOY Forums

Hi all

I have a question regarding multi gpu. I;ve come across this evga board that supports 2 x xeons and silly amounts of ram. The question here is if it has 7 slots and say 7 single slot cards how much performance loss is there across each additional card?

Also it is possible to add 4 gtx 580 (3gb)... Can anyone recommend the right PSU for the job (around 1600 watts)... remember it has 2 x xeons and 24gb of ram... the reason for this is I want to splash out on a serious setup for rendering and video editing HD. I looked into the quadros but render speeds to price are more important.

Does anyone know which CAD or 3d programs support quad sli?

Not sure if I should wait until the new nvidia cards arrive considering the expense and energy costs.

Any help here would be welcomed and if anyone has a dual or triple gpu rig please share the benchmarks.... because of octane I've got a new job

Hi,

The most reported GPUs seated on a single board with one single x16 PCI-E wide bus, with octane 2.4 and GTX5xx GPUs is 6 GPUs.
This setup works and speedup is near linear, i don't have extensive benchmarks handy but it was approx 5.5x the speed of one GPU.

As the difference in convergence only improves per square of samples in time, i would advise 4 GPUs as the best solution.
Going over 4 will give you more MS/s, but the difference in noise for final frames will not be that high.
The next best situation after 4 would be 8, but then you'd need a board with 8 PCI-E slots which i don't think is available right now,
or you should get extension boxes and put 2 in each box and then connect 3 to a 4slot board, which is very $$$ expensive

The 6x GPU setup was reported using linux as the OS, unfortunately we have no reports of any windows or mac systems beyond 4 GPUs.
With a high amount of GPUs, quadro/tesla boards provide better scaling currently.

Radiance

radiance wrote:Hi,

The most reported GPUs seated on a single board with one single x16 PCI-E wide bus, with octane 2.4 and GTX5xx GPUs is 6 GPUs.
This setup works and speedup is near linear, i don't have extensive benchmarks handy but it was approx 5.5x the speed of one GPU.

As the difference in convergence only improves per square of samples in time, i would advise 4 GPUs as the best solution.
Going over 4 will give you more MS/s, but the difference in noise for final frames will not be that high.
The next best situation after 4 would be 8, but then you'd need a board with 8 PCI-E slots which i don't think is available right now,
or you should get extension boxes and put 2 in each box and then connect 3 to a 4slot board, which is very $$$ expensive

The 6x GPU setup was reported using linux as the OS, unfortunately we have no reports of any windows or mac systems beyond 4 GPUs.
With a high amount of GPUs, quadro/tesla boards provide better scaling currently.

Radiance

Does this mean that 2 GPUs would converge less than twice as fast, despite giving twice the MS/s?

No,

I just meant that the benefit of 6 to 4 GPUs is less visually apparent than 4 to 2, and due to the additional cost that is usually much higher (very exotic MB and PSUs), i would leave it to a max of 4 per machine.

Radiance