GPU usages: v2.xx vs v3.00

Tue May 24, 2016 8:48 pm

Hi, guys.

v2.xx: constantly 99%
v3.00: up to 77%, avg 65% but v3.00 is ~1.7x faster than v2.25
in GPU usages running GTX 780 Ti Classy (364.96WHQL)

Why is the GPU usage not full using v3?

Tue May 24, 2016 9:25 pm

nuno1980 wrote: Why is the GPU usage not full using v3?

Octane V3 had an overhaul of the integration kernels, it doesn't work like V2 anymore.
see here:
viewtopic.php?f=33&t=51679

Mon May 30, 2016 8:12 pm

calus wrote:
nuno1980 wrote: Why is the GPU usage not full using v3?
Octane V3 had an overhaul of the integration kernels, it doesn't work like V2 anymore.
see here:
viewtopic.php?f=33&t=51679

Thank. But I get only 1020 MHz core at 80% GPU usage because GPU isn't stress in 3.00. Before I got 1150 MHz core at 99% GPU usage in 2.xx.

My GTX 780 Ti Classified speeds specs:
- core normal 1020 MHz
- core boost 1150 MHz

Tue May 31, 2016 10:35 am

In v3.00, I tried to change "render priority" from "high" to "low" or from "low" to "high" but no any differences.

But yes to have differences in v2.xx.

Therefore, is multi-GPU bad scale in v3.00???

Beppe OctaneRender™ Italia Blue Sky · Tue May 31, 2016 10:51 am

Hi nuno,
as calus already said, v3 is completely diffrent from v2 so you cannot compare some results, like for GPU Load. Please, try to compare the TDP value, you should see an improvement.
Octane v3 works well in Multi-GPU as always, but if you have only one card is difficult to check, I guess

ciao beppe

Tue May 31, 2016 11:29 am

bepeg4d wrote:Hi nuno,
as calus already said, v3 is completely diffrent from v2 so you cannot compare some results, like for GPU Load. Please, try to compare the TDP value, you should see an improvement.
Octane v3 works well in Multi-GPU as always, but if you have only one card is difficult to check, I guess
ciao beppe

Ok. I tested to increase +104 MHz core (GPU overclocked) but no performance gained because only real 1020MHz core at 80% GPU usage and at 70% TDP running new 368.22WHQL driver.
--> Look my previous message (Mon May 30, 2016; 8:12pm UTC).

Tue May 31, 2016 1:58 pm

Sorry but don't know how to say but you spend your time on wrong direction. Btw, programming is not a straightforward way and there is lot of factors. V3 is mostly faster than V2 and you can't compare or judge it by it that as doesn't use all power of gpu. Because workflow of a program is not linearly distributed on cpus/gpus and depend what does on thousands lines and sections. Also some cores cannot be fully used on some operations or while waiting the completation of other sections. Most important is that the total result. See that. Btw i see now you have a license, congratz.

Please do some good renders to share with us. I can't wait to see them.

Best,
Ahmet

Tue May 31, 2016 3:26 pm

Btw i see now you have a license, congratz.

Thank you!

aoktar wrote:Sorry but don't know how to say but you spend your time on wrong direction. Btw, programming is not a straightforward way and there is lot of factors. V3 is mostly faster than V2 and you can't compare or judge it by it that as doesn't use all power of gpu. Because workflow of a program is not linearly distributed on cpus/gpus and depend what does on thousands lines and sections. Also some cores cannot be fully used on some operations or while waiting the completation of other sections. Most important is that the total result. See that.

Please do some good renders to share with us. I can't wait to see them.

Best,
Ahmet

Ok. I got only real 1020MHz core (stock speeds are only equal to OC speeds due to not real boost core enabled as 1150MHz) at 80% GPU usage while I rendered "specular material" scene with ior 1.51 and dispersion 0.006 at PMC kernel with default sets except 60 depthes (diff+spec), parallelism 1, no "caustic blur" and 10k-reject.

But look 2 screenshots about Idea scene:
-- v2.25 final

-- v3.00 beta 4

Is that normal in v3.00beta4 (final?)?

Tue May 31, 2016 3:39 pm

V3=4.30 on %80
V2=2.50 on %100

Really don't know the what's your concerns or targets?
I have explained the some basics of programming. V2 is more near to brute force with less clever algorithms on distribution of works. V3 is different on estimation of core distributions on work packages.

Result is that. Maybe less usage but faster results. Saves the world a bit more with less power consumption. Don't worry for %20, use it on other tasks.

Best

Tue May 31, 2016 4:07 pm

aoktar wrote:I have explained the some basics of programming. V2 is more near to brute force with less clever algorithms on distribution of works. V3 is different on estimation of core distributions on work packages.

Result is that. Maybe less usage but faster results. Saves the world a bit more with less power consumption. Don't worry for %20, use it on other tasks.

Best

Thank you.

Really don't know the what's your concerns or targets?

"Parallel samples" PMC kernel - lower value: appear caustics eariler but lower GPU usage (may disable boost core and may not stress for GPU) while higher value: higher GPU usage but appear caustics later. Ok?