VRay RT (CUDA) consisently slower than VRAY RT (CPU) and VRAY Advanced. Why?

[magarlick]

Dear all

This is driving me nuts. Here are my specs.

Windows 10 Pro, latest build.
3ds Max 2016 SP1
Vray 3.30.04.
Memory = 32 GB DDR4 RAM.
CPU = Intel Core i7-5820 (hexa core) running at 4GHz
Graphics = Palit GTX970 with 4GB RAM

I recently invested in the above graphics card because I wanted to take advantage of GPU rendering. But in every scene I have tested (and there have been dozens), VR Advanced is ALWAYS faster than VRay RT. Sometimes it's much faster. Sometimes it's comparable, but still faster. Moreover, Vray RT set to CPU is also faster than VRay RT set to CUDA or OpenCL.

I read from the documentation that GPU rendering may not be faster than CPU rendering all of the time, and that it depends on the scene. But I have NEVER ONCE seen RT out-perform Advanced. WHY!?

The latest scene I am rendering is entirely Vray. V-ray materials (car paint) and Vray lights, and set to stop at noise levels of 0.01. Same result.

Please, somebody shed some light on this for me.

Kind regards,

M.

[benb32]

There are many variables that will dictate the speed of your renders - I know it's not going to be the answer you're looking for...

To help a little, it would be useful to post an example scene that you feel should be rendering more quickly. More specifically the baseline comparisons (I'm sure you're doing this already) should be using BF/LC on Adv and RT.

Ultimately your biggest issue will be that a GTX 970 is not a powerful card when compared against your own processor or a Titan X. However, the 970 is great value for money as a viewport card. I would recommend running more than one card for RT. It will be about managing your expectations as even running 3 970s will probably leave you disappointed. But if you have space and a power supply that can handle it:
- 3 titan Xs will be approximately 6 times as quick as your 970 i.e. 1 hour render would take 10mins
- 3 gtx 1080s (the new pascal card from nVidia) would be approximately 10 times as quick as your 970 (a loose estimate based on the reported fp32 performance reported by nVidia)

There are some benchmarks on the Vray RT section of the forum. Unless you invest more on your GPUs, you won't find RT will be something you use for production rendering or quicker feedback than RT CPU.

[magarlick]

Thanks for your reply. Seems I seriously overestimated the strength of this new card. (Which incidentally isn't great in viewport either, struggling at around 1 million polygons) with FPS <10). I assumed, naively, that this card's 1500+ cores would outperform my hexacore Intel, which is not the case. Yes, a Titan X would be nice, given the budget!

[savage309]

Thanks for your reply. Seems I seriously overestimated the strength of this new card. (Which incidentally isn't great in viewport either, struggling at around 1 million polygons) with FPS <10). I assumed, naively, that this card's 1500+ cores would outperform my hexacore Intel, which is not the case. Yes, a Titan X would be nice, given the budget!

It is very important how do you compare. The default settings in 3.30 were very bad for RT GPU. This is fixed for 3.40.
In the meantime, if you want I can check one of your scenes to see what is going on.

Best,
Blago.