Speed test 2 x gtx 1080ti vs 3 x gtx 1080ti V-Ray Next GPU

[dimension3plus]

Hey guys,
Currently I'm working with a 2x1080ti system (core i7 5960). I try to decide if a third one is worth the money and I really like to know how the performance behaves when putting in a third 1080ti.
Can you give me your result when rendering (vray next gpu) first with only 2 and then with 3 GPUs?

Thanks in advance!
Best regards
Olaf

[dimension3plus]

When looking at the vray GPU benchmark results I found out the following:
AMD 1950x. GTX 1080 ti x 1 = 26,3 s
AMD 1950x. GTX 1080 ti x 2 = 24,3 s
AMD 1950x. GTX 1080 ti x 3 = 22,3 s

So every additional GPU improves the speed about 10 %. It seems pretty low in my opinion. What is your experience?

[JezUK]

Why not take a look at the VRay Benchmarks ?

https://benchmark.chaosgroup.com/gpu

You can then look up 2 x 1080ti and 3 x 1080ti

[dimension3plus]

Thanks. Yes, I did - see my second post, but the results have limited informative value.

[Micha]

When looking at the vray GPU benchmark results I found out the following:
AMD 1950x. GTX 1080 ti x 1 = 26,3 s
AMD 1950x. GTX 1080 ti x 2 = 24,3 s
AMD 1950x. GTX 1080 ti x 3 = 22,3 s

So every additional GPU improves the speed about 10 %. It seems pretty low in my opinion. What is your experience?

I can't believe it. I hope someone says that this isn't true. So, GPU rendering would be useless.

[Morne]

The 1st versions of the benchmarks I view more of a proof of concept for quick reference, but not as an actual FACT guide. They did a pretty good job with the benchmark, but it is mostly useless with such simple scenes that take few seconds to render. The scene preperation probably takes longer than the actual rendering. Some of us hope in an updated version, they will add more real world scenes. Something that takes at least few minutes with powerfull hardware.

[Muhammed_Hamed]

I can't believe it. I hope someone says that this isn't true. So, GPU rendering would be useless.

This is not true at all
Vray Next GPU scales linearly with more cards you add.
Say if a scene renders in 30 minutes with one card, it will render in 15 minutes with 2 cards and like 8 minutes or so with 4 cards
Thing with Vray benchmark is that it takes few seconds to finish and GPUs get barely engaged in the test. So you will need a benchmark that runs for much longer. Octane benchmark is a perfect example, it takes like 5 minutes to finish regardless of the hardware you have and the benchmark score is very accurate.
I have 7 1080tis in one machine in office, 3 1080tis in my personal machine home and 2 2080s in a third machine. Render times are very different between all 3 systems.
The 7 GPU machine is more than twice as fast in rendering for every scene. I have been using both in production for over a year now
One more thing is that there is no way a single 1080Ti finishes the benchmark in 26 seconds. This is not possible. Even the Titan V and Quadro GV100 take over 40 seconds to finish this test.
Vray benchmark is only accurate with one or 2 GPUs.. With 3 GPUs and up you get weird scaling issues because the benchmark scene is very simple.

[Muhammed_Hamed]

savage309 Morning Blago!
This is why I talked about the Vray benchmark before. It should have a more complex scene or probably an approach similar to Octane benchmark and Cinebech(CPU), where it doesn't rely on render times
Or it should probably mention that it is only for testing one or 2 GPUs at most. Right now from (the benchmark results page) many people would think that Vray GPU doesn't scale linearly with multiple GPUs which is not the case.
From my testing, it can scale 100% linearly with 2 cards, 4 cards and 7 cards without issues. Even scaling better than Redshift(which they mention on their website that 2-4 GPUs are best for optimal performance) and it is also limited to 8 GPUs per machine unlike Vray GPU
So with the Vray benchmark tool, I get 21 seconds render time for 4 1080tis and 20 seconds for 5 1080tis.. with 6 cards then 7 cards I still get 20 seconds render time.
Unlike Octane benchmark which does scale linearly with more cards added. 400 points with 2 cards, 800 points with 4 cards, 1000 points with 5 cards and so on
And sorry for constant spamming your name.

[Micha]

Thank you very much Muhammed_Hamed ! I hoped to hear something like this. Helps a lot.

[savage309]

New version of the benchmark tool - V-Ray Bench Next, comes soon and it should show more correctly how V-Ray and V-Ray GPU scale with more hardware.

Best,
Blago.

[Muhammed_Hamed]

New version of the benchmark tool - V-Ray Bench Next, comes soon

Amazing news!
Looking forward to that and Thanks Blago!

[Muhammed_Hamed]

savage309 a7az0th Testing the new Vray Next benchmark now
It does scale nicely and the scores are very accurate across multiple GPUs
The new system for GPU scores makes a lot of sense and will be very helpful to the community
Every single point I have mentioned above has been tackled basically.
Thanks for considering your users feedback like always.

[francomanko]

does the new benchmark allow to exclude search terms, or say only show gtx 1080 x2 and not x7 etc?

[savage309]

Yes, just type "x2" in the search.

[francomanko]

if you do that, you dont just get 1080ti x2 setup, you need to be able to exclude setups too.. when i searched 1080 ti x 2 i got this : GeForce GTX 1080 Ti 11264MB, Quadro GV100 32768MB x4, TITAN X (Pascal) 12288MB x2

which isnt useful