Two cards the same speed at one !!

[xsipower]

Hello,

I downloaded the latest demo 1.022 beta 2.2. I ran the sample ocs benchmark with pathtracing and using only one of my 280 GTX and I get 1.39 M/s. When I run the same ocs with pathtracing with both 280 GTX enable in the Device Manager I get 1.37 M/s.

I also ran the other sample ocs and the results are always the same. Having one GPU is the same speed as having two GPUs. That's not right is it? What's wrong?

I am using Nvidia driver 258.96 with both SLI and PhysX turned off.

My Hardware:
Windows 7 x64 Ultimate
EVGA 680i SLI - 8 GB memory
Core2 Extreme Quad core Q6850 @ 3.00Ghz
EVGA GTX 280 (615Mhz, 1350Mhx) - PCI Expess slot x16
EVGA GTX 280 (615Mhz, 1350Mhx) - PCI Expess slot x16
EVGA 980 GTX+ (756Mhz, 1836Mhz) - PCI Expree slot x8

Thanks for the help

xsipower

[radiance]

Hi,

There must be something wrong as you should get much better speed.
To really see the difference you should bump up the resolution, but you can't do that with the demo version at this time...

Radiance

[tomas_p]

Hi.
Trie this. Switch the SLI off.

[xsipower]

Hello Tomas_p,

I mentioned in my first post:

"I am using Nvidia driver 258.96 with both SLI and PhysX turned off."

SLI is off.

Thanks anyway.

[radiance]

Can you try with a complex scene ?
like an interior with pathtracing, and set the pathtracing kernel to maxdepth = 1024 and the rrprob parameter to 0.65 or so.
that will make both cards work hard and if that does'nt offer more than just one, there's a configuration/driver issue on your system.

Radiance

[xsipower]

Hello Radiance,

I loaded a very heavy interior scene 370MB obj that has only one material shader for all objects and using a 22MB HDRI environment to light through the windows of the room. The render has 3021471 traingles and uses 602MB of the 997MB available on the 280 GTX.

I ran the same render with the settings you suggested with one 280GTX (octane showing one device below the render window) and then with two (two devices used showing on the bottom of the render window). I get almost the same 0.63 M/s with both renders. I have SLI off and the Device Manager in Octane corectly shows the three cards I have.

I am using the latest nvidia driver 258.96. Could the driver be the problem? The Device Manager says I have Cuda Driver 3.01 and Cuda Runtime Version 3.00.

Thanks,

xsipower

[xsipower]

Hello Radiance,

I ran TechPowerUp GPU-Z to verify that my SLI is disabled. I have attached the screen capture of it.

xsipower

[xsipower]

I ran the Octane Benchmark.ocs and set the pathtracing kernel to maxdepth = 1024 and the rrprob parameter to 0.65.

I used three instances of GPU-Z to see what the GPU Loading is on all three. This may not be a true test and I might be fooling myself.

I then ran the same benchmark with with 1x280GTX then 2x280GTX and then with 2x280GTX and 9800+. I attached screen captures of all three tests of GPU-Z.

What I found is that the GPUs that are selected do get used!! (yes) and here are the results with M/s:

1 GPU - 280 GTX(1) =~ 97%, 280 GTX(2)=0%, 9800+=0% M/s = 1.15
1 GPU - 280 GTX(1) =~ 96%, 280 GTX(2)=93%, 9800+=0% M/s = 1.12
1 GPU - 280 GTX(1) =~ 45%, 280 GTX(2)=43%, 9800+=97% M/s = 0.95

Its still a puzzle to me why my M/s gets less as I add GPUs? What am I doing wrong?

xsipower

[peterd]

Did you tried with older nvidia drivers 1xx?
Or wait for 2.3, which will do multiGPU more optimized.

Regards, Peter

[radiance]

HI,

I don't know, but you're result is definately not normal.
There must be a driver version issue or something.
Can you try with 197.xx WHQL drivers ?

Radiance