OTOY Forums

Just to know: I've lost the reason why there is not yet an official Octanebench release for Pascal Cards.
Anyone can remind me why?

Thanks a lot.

Hi,
just finish my 7 GPUs build and check for performance scale.
I noticed that it didn't scale well as I can see on my 4 GPU rig (with titan and 780ti ==> bench of 4 ~ sum of each individual bench)

1080ti 980ti Titan Black All 7 GPUs nominally, it should be about 1140, if it scales well (220+220+160+160+160+110+110).

So what do you think, the PMC perhaps, and memory ?
I have an xeon e5-2696V3, I think enough for that. I will check with a real scene.

itou31 wrote:Hi,
just finish my 7 GPUs build and check for performance scale.
I noticed that it didn't scale well as I can see on my 4 GPU rig (with titan and 780ti ==> bench of 4 ~ sum of each individual bench)

...

nominally, it should be about 1140, if it scales well (220+220+160+160+160+110+110).

So what do you think, the PMC perhaps, and memory ?
I have an xeon e5-2696V3, I think enough for that. I will check with a real scene.

here is my take on that, but I might be wrong..

You CPU seems to be not the fastest one & since it has a lot to do with 7X GPU what You end up is those cards waiting for information without being able to render. Again amount of cores doesn't matter so much here as Octane could not take advantage of more than 4cores unless smth has been changed lately.

In order to see what's exactly happening & where performance is lost it would be cool to have a printscreen of OctaneBench in action with some CPU monitoring app that would should load of CPU + Frequency it is running + like 7 instances of GPUZ with sensor tab opened for different cards or MSI afterburner with custom layout & graphs.

Seems that in V3 we start seeing more of these situation where scaling is no longer linear..however to find exact reason might take some time. Let us know if You would have more information to look at =)

Thanks Tom,
I have the same though about the CPU on the PMC Kernel, I can see the GPU load that fluctuate very much during PMC test bench. I will check that. I know that with my first rig with 4 GPU and a [email protected], the CPU load is about 75% VS 15% in PT kernel.
I will check that on this new rig

Just check with a real scene and calculate with rendering time in PT kernel and it scale perfectly.
Only PMC use much CPU. I don't know that octane only "use" 4 cores. With PMC kernel, I can see all cores in activity but only about 20% load on the xeon.
So, that's fine. With deactivating HT, the score is 1045.

So, I just tried a hacked OB in macOS Sierra, and my results are not as impressive as I hoped they'd be.
I'm aware it's not optimized for Pascal cards yet, but I was hoping for some kind of speed boost. My current setup is a GTX 1080 Hybrid connected via powered riser to Akitio Thunder 2 to late 2013 Late Mac Pro; the 1080 Hybrid replaces the 980TI I have. I believe I've read at some point in the past that one can't get 100% performance in a eGPU setup. I should be receiving an Akitio Node later this week, so I could factor out the riser being a bottleneck if it is at all.

I hope to be able to connect both the Node and Thunder 2 to the nMP, and if successful, I'll try to remember to post my results here.

NVIDIA 1080 GTX Strix win 7