test scene

[3LP]

Yeah the Tesla and quadro aren't that great at RT GPU compared to GTX,

Thanks anyways, useful info

Cheers

[Morne]

OK so I'm getting just over 7 minutes with CUDA on a GTX 670 with 4GB. People were getting 2m something on ancient 580s. I would think the 670 is faster?
On opening, max told me the files is setup with gamma 1.0

Now for some strange issues. If I switch between CUDA and CPU, the camera views dont match. CPU looks like it's giving me a slighlty wider angle?!?

[savage309]

OK so I'm getting just over 7 minutes with CUDA on a GTX 670 with 4GB. People were getting 2m something on ancient 580s. I would think the 670 is faster?
On opening, max told me the files is setup with gamma 1.0

Now for some strange issues. If I switch between CUDA and CPU, the camera views dont match. CPU looks like it's giving me a slighlty wider angle?!?

You should compare the same V-Ray versions. Different versions have different noise level for the same amount of time for example, which makes the tests between different V-Ray versions not correct.
580 is reported in the benchmark sheet to be around 3.5 minutes (https://docs.google.com/spreadsheets...A0k/edit#gid=0). Also 670 is newer, but uses 170Watt compared to the 250Watt usage of 580 .

[Morne]

Is there a newer benchmark somewhere that's already setup for 2.2 Gamma?

Also, whats up with the different camera fov between CPU and CUDA?

[savage309]

Is there a newer benchmark somewhere that's already setup for 2.2 Gamma?

Also, whats up with the different camera fov between CPU and CUDA?

I am not sure about if there is another benchmark scene. I think the camera is different because of distortion not being supported in RT GPU. We will get to that.

[Alan Iglesias]

OK so I'm getting just over 7 minutes with CUDA on a GTX 670 with 4GB. People were getting 2m something on ancient 580s. I would think the 670 is faster

Although I haven't tested the 670 directly against the "ancient" 580s, please keep in mind that the overall RT/GPU rendering power of nVidea's GPUs, core for core, has been reduced greatly as they have changed and ..er..."upgraded" their GPU architecture.

At the time, I was told by their project manager that the GPUs just were not being designed to be optimal for Vray RT/GPU-style rendering, so I shouldn't expect better performance with the new architecture(s), core for core. And of course this is exactly what we have seen.

As proof of all this, your 670 has nearly 3 times the cores as a single 580 and as you have seen, the rendering times do not seem to reflect that. It's just the way they decided to go.

As was mentioned, the new boards do use a lot less power per core.

The upshot of all this is that my 2 580 boards are still installed and running RT/GPU very well after all these years.

-Alan

[vlado]

Let's hope that the Pascal architecture will be better in that regard

Best regards,
Vlado

[Alan Iglesias]

Let's hope that the Pascal architecture will be better in that regard

Best regards,
Vlado

Yes, let's!

-Alan

[GregM]

Let's hope that the Pascal architecture will be better in that regard

Best regards,
Vlado

Sadly, not so much.

Single Titan X Pascal. For some reason 3ds Max Design 2014 (3m 52.6) was faster than 3ds Max 2017 (4m 41.1).



On the bright side, I am very happy with the view port performance of the new Titan compared to my old Quadro K5000.

-Greg

[savage309]

Sadly, not so much.

Single Titan X Pascal. For some reason 3ds Max Design 2014 (3m 52.6) was faster than 3ds Max 2017 (4m 41.1).

[ATTACH=CONFIG]31871[/ATTACH] [ATTACH=CONFIG]31872[/ATTACH]

On the bright side, I am very happy with the view port performance of the new Titan compared to my old Quadro K5000.

-Greg

As the docs say, please always use CUDA for NVIDIA GPUs . It is much faster and has more features.

Best,
Blago.

[GregM]

As the docs say, please always use CUDA for NVIDIA GPUs . It is much faster and has more features.

Best,
Blago.

Yikes, I just opened the file and hit render - didn't notice it was set to OpenCL

Titan X Pascal using CUDA - 2m 20.3.



Thanks for catching my error Blago.

-Greg

[Sbrusse]

Ah ha, the new Titan X then?

I'm doing 3 min 05 with my 980 Ti so it means the new titan is 24% faster.

Good to know,
Thanks!

Stan

[AtlJimK]

GTX1070 W10x64

CUDA latest drivers, single card. 110.1 seconds

[fraggle]

GTX1070 W10x64

CUDA latest drivers, single card. 110.1 seconds

Hello, it will be interesting to know the difference between 3.4 and 3.5 version!
If somebody can post it that will be interesting.
Thanks

[AtlJimK]

Another test with the work desktop. GTX780ti, single card, W7x64, VRay 3.50.03 = 322.2 seconds.