OTOY Forums

Wow, I didn't know Octane needs CPU power that much. I have a rig with 6xGTX 980Ti and I was getting a score of 568.60 with bench version 3030004.

That was with a shitty Celeron CPU inside.

Then I swapped it with i7-4790 @ 3.60Ghz and BOOM ! score 688.

That is a whopping 119.4 score point increase just on the account of a CPU, that is a score of another extra GPU !

I mean, I thought GPUs are doing all the work and CPU is just there for decoration, but apparently I was wrong.

I am getting a CPU upgrade for the second slave too.

Yes - CPU is for voxels and obj loading times that is why you get a faster score. But CPU will not speed up rendering speed itself.

I know about that, but I wasn't talking about preprocessing, I understand this takes CPU resources.

What puzzles me is why is the score for 1 GPU 109 and for 6 GPUs of the same type 580 when it should be 654 (6x109)!
Preprocessing is done only once and not for each GPU so why is there a slowdown with more GPUs ?
Are you sure OctaneBench takes into account CPU work for preprocessing ? This would be silly since you can't properly bench the GPU regardless of what your CPU is.

And heat is not the issue here, GPUs are in an open rig, properly spaced out with nice STRIX coolers. Maybe this could lower it by a few points but not that much.

This was with bench version 3030004

But with the bench version 2170000 the situation is very different

Score for the single GPU is 132 and for 6 GPUs 790 which is exactly 6 times more !! So there is no loss with muiltiple GPUs.



So, this makes me believe that from version 2->3 something changed and now CPU is also used during rendering for household tasks, not actual rendering. Am I correct ?

You are right Boris - from my experience while I could get 155 from a single 1080, from 7 I never get 7x155=1085 or so but much lower. What I noticed in GPUs activity that geometry is not loaded at the same time to all gpus at once but with slight delay (2x xeons) so this is probably related to PCIe architecture and their speed.

This is what happens with increasing GPU count from 1 to 6:
(bench 3030004)

CPU Celeron:

1x980Ti score 109
6x980Ti score 568


CPU i7:

1x980Ti score 112
6x980Ti score 695


so with Celeron you get less then 6 x single score and with i7 you get more then 6 x single score.

Oh i see. This is interesting.

Maybe this user community hacked bench V3 doesn't work correctly ?

So should I upgrade the other Celeron or all these results are fake ?

If you used hacked bench I would not relate any of these results to the official bench.

Well I didn't relate it to the official online benchmarks, I just compared results from my same machine with Celeron and then swapped it with i7 and got a 128 point boost. So I guess it is worth upgrading. I would just love devs to explain this a bit.