GPU benchmarks

[savage309]

The sampler and the V-Ray has been changed a lot during the years. And this is kind of obsolete benchmark, V-Ray Bench (benchmark.chaosgroup.com) is the right one these days

It is okay for V-Ray to do a bit more passes. Your runs seems fine. I can imagine that these rather old and mid-tier GPUs are slower than a modern Ryzen CPU. Most of our users tend to use 1080ti or 1070 these days.

Best,
Blago.

[vitruvius maximus]

So, guys,
Vray 3,6 in machine, both cards too, quadro for viewport, gtx for RT

Im running on Vray RT 3.60.03, Max2015. In one slot PCIe 3.0x16 Quadro4000 an in PCIe 3.0x8 is GTX460, in case of use this two cards both slots are runnig PCIe x8. Quadro has drivers 376.33 and GTX has 382.33. I downloaded benchmark 1.0 scene from http://ftp.chaosgroup.com/support/benchmark_scene.zip. Just open it, select card for GPU rendering, and have a problem:
c++/cpu 3:50 min (AMD Ryzen 7 1700 3,6OC)
quadro4000 18:39,8
gtx460 14:15

What i a m doing wrong? Just only cpu has times i expected. Cards are totally out. When I see results in benchmark I chart, i should be elsewhere around 6 min. I remember years ago I run this benchmark with the same GTX 460 and everything worked fine (few yaers back)




What I have to change?
And one thing more - I guess RT should stop after reaching 512 path/pixel. But it doesnt, it continues to infinity and I have to stop it manually - it means to look on monitor and immediately to stop rendering after 512 is reached, i other way RT has 512, but time continues to run...
When I run VRAy benchmark application with quadro, it reaches approx 7:30 and GTX460 doesnt run because of just 1 GB ram... I wrote this in two threads, but this one is little bit dead , just few posts this year
roman

[allemyr]

Thank you very much for the standalone benchmark program haven't tried it myself. Would be nice with a single card option to even more easily compare GPU with each other.
Also interesting that it is up to 5-10 seconds difference for same GPU with different CPU RAM and SSD setups.

What more even more interesting is to actually se the different resluts on the GPU.

It seams for me like the Titan X Pascal is only 15-20% faster then Titan X Maxwell, which seams a bit odd. And that Titan X Maxwell is a tiny bit faster then the regular GTX 1080.

Maybe for a architecture student like me it's best to buy some more second hand Titan X Maxwell's instead.

Great work again, thanks for that!

[savage309]

Hi,
I think that it would be super nice with a benchmark scene that more people use to see how different single cards performs.

I agree and we are working on that for a while now.

Don't know if it's only me that really don't have a great clue how GTX 1070, GTX 1080, GTX Titan X, Maxwell or Titan X Pascal cards compares to each other.

It depends on scene to scene - more or less for V-Ray GPU the Quadro/GTX series sorted by performance - Quadro P6000 (24GB), Titan X Pascal (12GB), Quadro P5000 (16GB), GTX 1080 (8GB), Quadro M6000 (12 or 24 GB), Titan X Maxwell (12GB), GTX 980ti (6GB), GTX 1070 (8GB).

Most interesting for me would be to compare Titan X Maxwell and Titan X Pascal.

Titan X Pascal is faster, probably ~40%.

Best,
Blago.

[allemyr]

Hi,

It's so cool that Forest Color is now supported with the 3.50.02 beta, I think that finally Vray RT GPU is the king over Octane now with the 3.5 Beta but I might be wrong. I feel very happy that I didn't start to hassle around with another way of building materials and continued in with Vray instead of Octane or FStorm. Ofcourse still a new user to Vray but I'am learning would be terrible to waste my time on other.

I think that it would be super nice with a benchmark scene that more people use to see how different single cards performs. Something that would have been nice aswell should be a scene that is quite fast to transfer to the GPU but takes time to render so the individual parts like CPU and RAM and SSD doesn't matter that much on render time. I think the existing scenes is still good for that, ofcourse brute force/brute force would be most interesting to se the power of the cards. If the cards is overclocked in the benchmark, it could be to see how many percentages it is overclocked, or you maybe could calculate that by yourself compared to a stock card. Don't know if it's only me that really don't have a great clue how GTX 1070, GTX 1080, GTX Titan X, Maxwell or Titan X Pascal cards compares to each other. Most interesting for me would be to compare Titan X Maxwell and Titan X Pascal. Ofcourse it's quite fast to check the speeds for games etc but they are always something else. Take for example the GTX 580 series that was faster then the GTX 680 in Vray RT GPU.

A new benchmark scene and a page to compare them in some way could be terrific, maybe especially for new users to compare different NVIDIA cards. To compare existing and later upcoming models like the 1080ti and on and on.

[flat]

This scene is linked in google doc.



As for this scene, Mr. Stan is increasing resolution and a sample?
(1280px and 2048 ppp)

http://forums.chaosgroup.com/showthr...721#post648721

Therefore, the linked scene is required in order to get the same bench mark as GoogleDoc.

[tashko.zashev]

Would you re-upload someone the following bench mark 2.0 scenes?
(Plz ChaosMirror)

https://dl.dropboxusercontent.com/u/...mark_scene.zip

https://dl.dropboxusercontent.com/u/...scene_2012.rar

Maybe this is what you are looking for: http://ftp.chaosgroup.com/support/benchmark_scene.zip
or I'm wrong?

[flat]

Would you re-upload someone the following bench mark 2.0 scenes?
(Plz ChaosMirror)

https://dl.dropboxusercontent.com/u/...mark_scene.zip

https://dl.dropboxusercontent.com/u/...scene_2012.rar

[biochemical_animations]

On the google doc spreadsheet there are entries for multiple 970 OC'd...is there somewhere mentioned what these were OC'd to? Also, I was curious as to the system used for these entries (CPU speed, memory, MB).

[savage309]

This extra time is cause by the image sampler additivity settings - without them it is much closer (in the automated internal tests, all these are turned off from the .vrscene file). Still, the new beta renders longer the benchmark scene, but this is because the result is closer to the CPU as well (it does more calculations). I am not familiar with the image sampler and can't tell if that is normal.

Best,
Blago.

[super gnu]

ok, so, i just left benchmark 2 to finish on my titan x.. it scored 8 min something on vray 3.3.. with the new beta it completes in 15 minutes.. ive not done a quality comparison, but from memory, it doesnt look noticeably better given all that extra rendertime.. is such a large discrepancy to be expected?

[savage309]

surely for the benchmark scene those values are fixed?

Still I would do the comparing of results after the same amount of time .

Best,
Blago.

[super gnu]

The better way to compare is to let it render for the same amount of time and compare the noise in the results. The image sampler can take different decisions for when a certain noise is reached - depending on the dynamic noise threshold, max samples per pixel, etc.

Best,
Blago.

surely for the benchmark scene those values are fixed?

[savage309]

i got 8 minutes something on my oc titan x .. doh.

The better way to compare is to let it render for the same amount of time and compare the noise in the results. The image sampler can take different decisions for when a certain noise is reached - depending on the dynamic noise threshold, max samples per pixel, etc.

Best,
Blago.

[super gnu]

It only seems so; eventually it will complete when all pixels that need sampling have been finished. Sometimes pixels may be (re)activated later on in the rendering if something changes around them.

Best regards,
Vlado

hm, in that case im getting much longer benchmark times with the new beta.. i got 4m 11 secs on my oc titan x before.. but with the beta it was over 6 minutes when i cancelled it.


-EDIT.. im being an idiot.. im confusing benchmarks 1 and 2... i got 8 minutes something on my oc titan x .. doh.