Nvidia driver version impact on fstorm performance?

[snakeboxmedia]

Can anyone clear up the debate that happened a while ago?

I still can't seem to find an answer to if the build of nvidia driver has an impact on fstorms rendering performance?
The debate I found, seemed to be more about the auto GI changes from 0.9 build to 1.0 build of fstorm, rather than driver versions? is that correct?

Thanks!

[Karba]

Latest nvidia driver gives bad performance.

[snakeboxmedia]

Alright, good to know. Do we know why? is there anything we can look for or do we have to guess every time nvidia releases a new driver? The latest drivers seem to perform better generally for everything else, that's why I am asking, if there is a sweet spot version or if we need to go back as far as possible.

thanks

[Lycan17]

Can anyone clear up the debate that happened a while ago?

I still can't seem to find an answer to if the build of nvidia driver has an impact on fstorms rendering performance?
The debate I found, seemed to be more about the auto GI changes from 0.9 build to 1.0 build of fstorm, rather than driver versions? is that correct?

Thanks!

Yes, the latest drivers don't give best performance. I've tried various drivers and for me 375.70 is considerably faster than the latest versions this year. I've posted some comparisons in the benchmarks thread here. I think it's probably due to the fact that almost all new driver updates are 'Game Ready Drivers' that are specifically tuned for the latest game that has come out and aim to just improve its performance and no general optimizations. Don't know what tweaks they make in these.

[Sylk]

The latest versions install ever more services and other heavy and useless pollution that we do not want!! With each new release the drivers take dozens of additional megabytes... and performances decrease...

[Iskánder ivp87]

Yes, the latest drivers don't give best performance. I've tried various drivers and for me 375.70 is considerably faster than the latest versions this year. I've posted some comparisons in the benchmarks thread here. I think it's probably due to the fact that almost all new driver updates are 'Game Ready Drivers' that are specifically tuned for the latest game that has come out and aim to just improve its performance and no general optimizations. Don't know what tweaks they make in these.

the faster for me is the 375.95

[Lycan17]

the faster for me is the 375.95

The 375.63, 70 and 95 gave me the same performance. The latest 382.xx versions are definitely slower. Atleast with v0.3.8a.

And Corona guys have actually raised this unreliable GPU driver issue as one of the problems with GPU rendering on their 'Proudly CPU-based' page. They're really trying hard to diss GPU renderers.

[Iskánder ivp87]

The 375.63, 70 and 95 gave me the same performance. The latest 382.xx versions are definitely slower. Atleast with v0.3.8a.

And Corona guys have actually raised this unreliable GPU driver issue as one of the problems with GPU rendering on their 'Proudly CPU-based' page. They're really trying hard to diss GPU renderers.

nah, I think cups just haven't enough power, maybe the thing will change with the amd competition, ryzens and i9s etc but for now I don't quite see it yet

[Syrom]

The 375.63, 70 and 95 gave me the same performance. The latest 382.xx versions are definitely slower. Atleast with v0.3.8a.

And Corona guys have actually raised this unreliable GPU driver issue as one of the problems with GPU rendering on their 'Proudly CPU-based' page. They're really trying hard to diss GPU renderers.

LOL! Can you link to that place?.... forget it.. found it.

[Iskánder ivp87]

The 375.63, 70 and 95 gave me the same performance. The latest 382.xx versions are definitely slower. Atleast with v0.3.8a.

And Corona guys have actually raised this unreliable GPU driver issue as one of the problems with GPU rendering on their 'Proudly CPU-based' page. They're really trying hard to diss GPU renderers.

well, I did an 375.70 clean install and the time drop 1 sec or so, with the same noise at the end, then I went to the oldest I have here, the 372.70, and the time dropped 6 full seconds more... !?

[snakeboxmedia]

My biggest issue is that using 1080TI's you are limited in how far back you can go with driver versions, also like I mentioned, for everything else the newer drivers run a lot better.

[Lycan17]

well, I did an 375.70 clean install and the time drop 1 sec or so, with the same noise at the end, then I went to the oldest I have here, the 372.70, and the time dropped 6 full seconds more... !?

I tried 372.70 as well but my render times only increased by 7 seconds in the new benchmark scene. It's important to note that Pascal's GPU Boost 3.0 causes dynamic overclocking and it's very hard to keep it fixed while my Maxwell 960 stays put. I have to monitor and track the clock speeds every time I bench.

Right now 375.70 has given me the best time of 7:53 with 1070 + headless 960 in v0.3.8a in the latest benchmark scene. I guess it's as fast as a stock 1080 Ti.

On a side note, the latest version of FStorm reaches final sample count faster but shows more undersampled noise levels at the end.

[Lycan17]

My biggest issue is that using 1080TI's you are limited in how far back you can go with driver versions, also like I mentioned, for everything else the newer drivers run a lot better.

NVIDIA would rather have us use the pricey Quadro cards since I don't think they'll do anything about this issue any time soon.

[snakeboxmedia]

I tried 372.70 as well but my render times only increased by 7 seconds in the new benchmark scene. It's important to note that Pascal's GPU Boost 3.0 causes dynamic overclocking and it's very hard to keep it fixed while my Maxwell 960 stays put. I have to monitor and track the clock speeds every time I bench.

Right now 375.70 has given me the best time of 7:53 with 1070 + headless 960 in v0.3.8a in the latest benchmark scene. I guess it's as fast as a stock 1080 Ti.

On a side note, the latest version of FStorm reaches final sample count faster but shows more undersampled noise levels at the end.

I'm not sure comparing sample count between versions makes much sense, because I think a lot has changed in terms of what happens per sample, and therefor also with noise. I remember everyone freaked out years back when vray updated their sampling to be much more accurate but also a little slower in simple scenes, but much faster and better in really complex scenes, overall faster but needed slightly different numbers or ways of thinking to use it. But people just compared straight up and said, mine is 5 sec slower this version is shit.

[Iskánder ivp87]

I'm not sure comparing sample count between versions makes much sense, because I think a lot has changed in terms of what happens per sample, and therefor also with noise. I remember everyone freaked out years back when vray updated their sampling to be much more accurate but also a little slower in simple scenes, but much faster and better in really complex scenes, overall faster but needed slightly different numbers or ways of thinking to use it. But people just compared straight up and said, mine is 5 sec slower this version is shit.

we are talking about the same Fstorm version here, the only thing we're swapping is the nvidia video driver, see the thread name!