GPU rendering speed - Q6000 vs K6000 vs M6000

[hm4]

Hi,

Does Flame rely on CUDA for GPU rendering/processing?

How significant is the speed difference between a HP Z800 with Quadro6000 and HP Z820 with K6000 and HP Z840 with M6000?

Obviously there is a difference, but is in ex. HP Z820/K6000 twice as fast as HP Z800/Q6000?

I'm aware there's no benchmarks or guarantees, but some experience, guidelines, opinions, tips etc. are most welcome 🙂

Many thanks!

Henrik

[joelosis]

There is a google doc on logik somewhere that has benchmarks of not just GPUs but also CPUs,
That said the latest architecture for GPU can make a massive difference. My Titan Pascal GPUs although unsupported outperform older gen Quadros, i have no doubt that a p5000 or p6000 would outperform my Titans
Remember the more vram the merrier and u should be fine

[hm4]

Hi Joel,

Thanks for the hint.

Unfortunately i didn\t seem to have luck in finding these benchmarks.

Have you experienced any issues with the Titan cards?

Or does everything perform as expected incl. background reactor?

Many thanks!

Cheers,

Henrik

[joelosis]

Everything works as expected including background reactor, if i were to get a card today i would go the 1080ti, best bang for buck in my opinion and a decent amount of vram.
Again if u go GeForce over quadro route remember its officially unsupported

[Peter_Tresize]

benchmarks:   https://docs.google.com/spreadsheets/d/1yJwSRl3JGInfyWulAVGNG-q_JgF3nRdjsSTAjAy2_AQ/edit?pli=1#gid=0

[hm4]

Hi,

Thanks for the benchmarks.

It obviously seems like you need lots of CPU power, memory, fast storage and top GPU to get the the best performance. I actually would have thought it was primarily tied to GPU, but that doesn't seem to be the case. Then again, i don't know what the test contains.

Is there a place where the test/projetc can be downloaded?

Cheers,

Henrik

[joelosis]

Alot is tied to GPU, my test on only a ssd before i did storage setup was only 3 seconds slower so there's that, also more CPU doesn't mean anything with flame, its a balance of clock speed and cores, if anything that sheet show u dont need a high core count over priced xeon to have a fast flame.
The benchmark lives on logik i believe

[joelosis]

To prove the point if u r on Linux and open a shell and type 'watch nvidia-smi' then open flame and render a heavy setup u will see your vram get eaten up and usage on the gpu, however if u run top or htop u will see that all those CPU cores rnt being utilized to their max potential.

[hm4]

Hi Joel,

Thanks for all your input!

I will for sure test out your suggestion.

Thank you 🙂

Cheers,

Henrik

[Peter_Tresize]

Hi Henrik,

The link from Logik: (or I can pm you the file)

https://lookaside.fbsbx.com/file/Speed_Check_March_2015.zip?token=AWxOC1XovWIrO0HIPjB6tWII3H4WEbfGXT...

Peter

[joelosis]

thanks Peter