Page 4 of 11
Re: OctaneBench
Posted: Wed Mar 11, 2015 11:05 pm
by abstrax
nuno1980 wrote:GTX 980 is faster than my videocard at PMC kernel!?????? But wrong because attention: GTX 980 has non-reference speeds...
Yes, it's faster in PMC. The reference 980 results come from a regular clocked 980, but using a manual fan curve with EVGA Precision to keep the GPU cooler in trade of more noise. I used this fan curve:

- fan_curve.png (4.35 KiB) Viewed 11037 times
Re: OctaneBench
Posted: Wed Mar 11, 2015 11:37 pm
by nuno1980
abstrax wrote:Yes, it's faster in PMC. The reference 980 results come from a regular clocked 980, but using a manual fan curve with EVGA Precision to keep the GPU cooler in trade of more noise.
Ok. Weird because...
-->
octane benchmark v1.0 at PMC kernel running OR 2.17 demo:
My GTX 780 Ti Classified - 5.05 MS/s
GTX 980 ("DayDreamer" user) - 3.5~3.6 MS/s
My card is faster than GTX 980. But...
-->
New octanebench, My card is a bit slower than GTX 980 - wtf!?
Re: OctaneBench
Posted: Wed Mar 11, 2015 11:58 pm
by abstrax
nuno1980 wrote:abstrax wrote:Yes, it's faster in PMC. The reference 980 results come from a regular clocked 980, but using a manual fan curve with EVGA Precision to keep the GPU cooler in trade of more noise.
Ok. Weird because...
-->
octane benchmark v1.0 at PMC kernel running OR 2.17 demo:
My GTX 780 Ti Classified - 5.05 MS/s
GTX 980 ("DayDreamer" user) - 3.5~3.6 MS/s
My card is faster than GTX 980. But...
-->
New octanebench, My card is a bit slower than GTX 980 - wtf!?
I get higher Ms with the old scene (what were your kernel settings?).
Anyway, the old trench scene was never really well suited for benchmarking because it was super simple compared to any normal scene (only one mesh of a few thousand triangles and one material...). Even the benchmark scenes used by OctaneBench are still very simple compared to actual production scenes.
-> With OctaneBench we hope to have things a bit more comparable to real scenes.
Re: OctaneBench
Posted: Thu Mar 12, 2015 12:26 am
by nuno1980
abstrax wrote:I get higher Ms with the old scene (what were your kernel settings?).
Click a node "RenderTarget PMC" as default settings.
Anyway, the old trench scene was never really well suited for benchmarking because it was super simple compared to any normal scene (only one mesh of a few thousand triangles and one material...). Even the benchmark scenes used by OctaneBench are still very simple compared to actual production scenes.
-> With OctaneBench we hope to have things a bit more comparable to real scenes.
Ok.

Re: OctaneBench
Posted: Thu Mar 12, 2015 12:39 am
by abstrax
nuno1980 wrote:Click a node "RenderTarget PMC" as default settings.

I assume you were using the trench scenes of the 1.20 demo scenes. There I get 4.02 Ms/s with alpha shadows enabled and 4.62 Ms/s with alpha shadows disabled on a GTX 980 in the 2.17 demo.
Re: OctaneBench
Posted: Thu Mar 12, 2015 1:04 am
by nuno1980
abstrax wrote:I assume you were using the trench scenes of the 1.20 demo scenes. There I get 4.02 Ms/s with alpha shadows enabled and 4.62 Ms/s with alpha shadows disabled on a GTX 980 in the 2.17 demo.
Ok, I have scenes demo v1.20.
I got:
alpha shadows enabled 5.05 MS/s
alpha shadows disabled 5.81 MS/s
Has your GTX 980 reference or OC edition? Do you use stock or do overclock for GPU?
Re: OctaneBench
Posted: Thu Mar 12, 2015 1:10 am
by abstrax
nuno1980 wrote:Ok, I have scenes demo v1.20.
I got:
alpha shadows enabled 5.05 MS/s
alpha shadows disabled 5.81 MS/s
Has your GTX 980 reference or OC edition? Do you use stock or do overclock for GPU?
It's a reference build from EVGA I think with stock clock. We never overclock here in the office.
All that proves is how useless the old benchmark scene is.
Re: OctaneBench
Posted: Thu Mar 12, 2015 2:15 am
by riggles
abstrax wrote:Tim fixed a bunch of issues with the results page. Could you have a look and tell us if that's better now?
Yes it is improved. Getting rid of the pagination helps a lot. And now the "single GPU only" preference persists between filtering modes. Thank you. I'm sure the Titan Z thing will take a bit more time to settle and implement.
Re: OctaneBench
Posted: Thu Mar 12, 2015 2:36 am
by abstrax
riggles wrote:abstrax wrote:Tim fixed a bunch of issues with the results page. Could you have a look and tell us if that's better now?
Yes it is improved. Getting rid of the pagination helps a lot. And now the "single GPU only" preference persists between filtering modes. Thank you. I'm sure the Titan Z thing will take a bit more time to settle and implement.
Currently, we don't want to change it, because it's always not 100% clear to the user what "1x Titan Z" means. Octane internally deals with GPUs and doesn't care if they are on the same board or not. So, a user who works with Octane and sees "1x Titan Z", will still scratch his/her head and wonder if that means that it's the result of a Titan Z video card or Titan Z GPU.
What we've done is explicitly mentioning multi-GPU video cards in the text above the result list.
Re: OctaneBench
Posted: Thu Mar 12, 2015 7:06 am
by smicha
Marcus,
Could you please add for dual gpu cards a simple explanation in the form
4x GeForce GTX TITAN Z (2 cards) 1 result
3x GeForce GTX TITAN Z (1.5 card) 1 result
2x GeForce GTX TITAN Z (1 card) 1 result
1x GeForce GTX TITAN Z (0.5 card) 1 result
This would be very handy.