is GPU vray benchmark test reflective of latest Vray?

[seandunderdale]

Hey guys,

A question about the Vray GPU benchmark test. Ive got a dual GTX 1080ti system at home, but Ive had a remote Tesla Volta V100 machine made available to me also. We ran the GPU benchmark on it, and my dual 1080ti machine scored higher. The IT chap who setup the V100 machine said the Vray GPU benchmark isnt using the latest Vray and isnt optimised for RTX rendering, and therefore scored badly. Can anyone shed some light on this? Is the vray bechmark not optimised for RTX mode, and therefore its not ginving a fair test between my home rig and the remote one? That was the suggested reason behind the low score, but Im wary about using a system that scores low on a benchmark.

This is the single GPU in the remote machine
https://www.scan.co.uk/products/pny-...c-pci-e-30-x16

[daniel.taskoff]

Hi there, it's true, that the V-Ray Next (which V-Ray Next Benchmark uses) is not optimised for RTX rendering, but we'll soon release a benchmark with the latest V-Ray, which has support for RTX rendering.

[seandunderdale]

Hey Daniel. Cheers for the response. Are you able to say if we are talking days / weeks / months?

[daniel.taskoff]

It'll be more like months.
However, I now notice, that the cards you're talking about (Tesla V100 and GTX 1080 Ti) don't have RT cores, so I have to correct my initial answer a bit:

V-Ray Next Benchmark doesn't use latest V-Ray, and is not optimised for RTX rendering, but that shouldn't matter when comparing GPUs not having RT cores.

[seandunderdale]

Ok good to know.

Also, my mistake about rhe card. Its a Tesla T4...but on the specs I still cant see any RT cores listed. Would this card also not see any benefit from RTX mode?

https://www.scan.co.uk/products/16gb...ecc-gddr6-81tf

[a7az0th]

Hi there

The Tesla T4 is a Turing generation GPU and it does include RT Cores in it. Meaning it will benefit from RTX rendering.
Since V-Ray Bench currently does not operate in RTX mode it only stresses the CUDA cores.
When you take a look at the 1080ti specs, each of your 1080tis has 3584 CUDA cores
The Tesla T4 has only 2560 of them, now, to be fair, the Tesla has Turing generation CUDA Cores and they might be better than the Pascal generation in the 1080ti, so it's not a 1:1 comparison, but still.
If you are comparing 2x1080ti to 1xTesla T4, given the specs of both cards i would expect the 2x1080ti to score better.

[seandunderdale]

Yeh youre right Alexander. Initially I went with the IT suggestion and used the T4, but things just werent as snappy in RT. Especially when using denoiser to see a cleaned up render, my 2x1080tis get there faster. Thanks for the confirmation!

[seandunderdale]

Any update on the RTX benchmark test? Would be nice to be able to really test the 30 series cards. Also, is there any plan to bring hybrid rendering to RTX mode?