So I just pulled the trigger on a new system... Did I make a mistake?

Started by jayfade, June 26, 2019, 11:04:29 PM

Previous topic - Next topic

0 Members and 3 Guests are viewing this topic.

jayfade

After using Keyshot for a while and projects grew bigger the render time got way to much longer I finally pulled the trigger to buy a new system.
But the big question is if I've made a mistake?
Here's what I picked: https://pcpartpicker.com/list/9GxrTB
I've done some research about the Threadripper 2990wx but couldn't find much about it related to Keyshot.
From what i've seen Keyshot relies on core count so that why I've picked the Thredripper 2990wx.
Any thoughts about the build?
The system I've in this moment runs with a i7-6700HQ 2.6 GHz, 32gb RAM, NVIDIA Quadro M2000M

Nils Piirma

As a Threadripper owner and also running the second gen threadripper, then it will run fantastic no doubt about it.

Go ahead with the build, Highly recommended.

here's a little better list to go for.
https://pcpartpicker.com/list/BcFT8Y

jayfade

Quote from: Nils Piirma on June 27, 2019, 12:51:39 AM
As a Threadripper owner and also running the second gen threadripper, then it will run fantastic no doubt about it.

Go ahead with the build, Highly recommended.

here's a little better list to go for.
https://pcpartpicker.com/list/BcFT8Y

Thanks, this makes me more confident about the build :)
Do you know if Keyshot will add GPU support?

Nils Piirma

GPU is only used in the viewport to generate FX by using the image tab in project menu, some settings are used to generate artificial effects.
Other than that, GPU is not used for rendering, only CPU, and future plans lead me to belive that there wont be any GPU rendering coming from KeyShot (spoken with the Dev leader about this)

jayfade

Quote from: Nils Piirma on June 27, 2019, 03:18:52 AM
GPU is only used in the viewport to generate FX by using the image tab in project menu, some settings are used to generate artificial effects.
Other than that, GPU is not used for rendering, only CPU, and future plans lead me to belive that there wont be any GPU rendering coming from KeyShot (spoken with the DEV's about this)

Thanks for the quick answer! :)

DriesV

Regarding GPUs...

KeyShot 9 will in fact introduce raytracing on the GPU. Our implementation is based on NVIDIA's OptiX API and will thus require a NVIDIA GPU to run.
Any GPU from the Maxwell generation (e.g. GTX 980) and newer can be used. Using the newer Turing cards with RTX technology, significant speedups can be achieved, since we also leverage the RT (raytracing) cores on those cards.

Dries

Nils Piirma

Quote from: DriesV on June 27, 2019, 08:30:15 AM
Regarding GPUs...

KeyShot 9 will in fact introduce raytracing on the GPU. Our implementation is based on NVIDIA's OptiX API and will thus require a NVIDIA GPU to run.
Any GPU from the Maxwell generation (e.g. GTX 980) and newer can be used. Using the newer Turing cards with RTX technology, significant speedups can be achieved, since we also leverage the RT (raytracing) cores on those cards.

Dries


Thats good to know, last time I spoke with Niels there weren't any plans to support it, but now I stand corrected.

LM6

Quote from: DriesV on June 27, 2019, 08:30:15 AM
Regarding GPUs...

KeyShot 9 will in fact introduce raytracing on the GPU. Our implementation is based on NVIDIA's OptiX API and will thus require a NVIDIA GPU to run.
Any GPU from the Maxwell generation (e.g. GTX 980) and newer can be used. Using the newer Turing cards with RTX technology, significant speedups can be achieved, since we also leverage the RT (raytracing) cores on those cards.

Dries

Hi Dries..

Will version 9 support multiple GPU or just the one?

Cheers

Peter


DriesV

Hi Peter,

Multiple GPUs are supported.
We have been testing with a couple of RTX 2080 cards in a single system and that works fine. With the Turing cards you can also leverage NVLink, for the cards that support it, to double (not exactly double, but at least significantly expand) the pool of available GPU memory.

Dries

LM6

Quote from: DriesV on July 01, 2019, 07:32:31 AM
Hi Peter,

Multiple GPUs are supported.
We have been testing with a couple of RTX 2080 cards in a single system and that works fine. With the Turing cards you can also leverage NVLink, for the cards that support it, to double (not exactly double, but at least significantly expand) the pool of available GPU memory.

Dries

Sounds good, I have 4 x RTX 2080 Ti  plus a Threadripper 2990WX in my system now.
Looking forward to try KS9, when will the user testing start?

SK1107

Quote from: DriesV on July 01, 2019, 07:32:31 AM
Hi Peter,

Multiple GPUs are supported.
We have been testing with a couple of RTX 2080 cards in a single system and that works fine. With the Turing cards you can also leverage NVLink, for the cards that support it, to double (not exactly double, but at least significantly expand) the pool of available GPU memory.

Dries

Hey Dries,

Can you please clarify if  several GPUs will neutralize advantage of Network rendering?   

I just have a chance to get 256 cores for a good price. But now i doubt if i should proceed towards Network Rendering or better to wait for GPU support (and just put several GPUs)?

Thanks


DriesV

Hey SK1107,

A system like Peter's with four RTX 2080 Ti cards will definitely be a very fast machine for GPU rendering.
You shouldn't see GPU as making all CPU-based rendering (including Network Rendering) obsolete though. To the contrary. I think the CPU and GPU rendering are quite complimentary and they each have unique benefits.

For some GPU will make a lot of sense, for many others CPU will remain the best choice.
There are several reasons for that:

  • GPU performance is very scene dependent. The speedup of GPU over CPU is typically largest for relatively simple models with simple materials/shaders. I.e. where there is a lot of pure raytracing. Our tests have shown that a single RTX 2080 (non Ti) can be 7 times faster than a Threadripper 2990WX in certain scenarios, but that number doesn't hold for all types of scenes.
  • GPU still has the limitation that everything from models to textures etc. needs to be able to fit in the GPU memory. An RTX 2080 card has 8 GB of memory and that is basically what you're stuck with as far as scene complexity and render size goes. Of course, higher end GPUs will have more memory, but they do come at a premium price. CPU rendering is much more flexible in terms of memory usage, and new PCs come with much more system memory than 8 GB.
  • For now, GPU rendering will be a local rendering solution. The benefit of Network Rendering is that jobs can be offloaded to different machines than the one you are working on.

You will have to make your own evaluation to determine if CPU or GPU makes sense for you. Of course, you can always decide to combine both options and leverage the strength of each.  :)

Dries

DriesV

I can also add that KeyShot 9 will have denoising functionality, which, depending on the scene, can slash render times. It will be available for both CPU and GPU rendering. Network Rendering will also support denoising.

Dries

SK1107

Quote from: DriesV on July 02, 2019, 12:45:00 AM
Hey SK1107,

A system like Peter's with four RTX 2080 Ti cards will definitely be a very fast machine for GPU rendering.
You shouldn't see GPU as making all CPU-based rendering (including Network Rendering) obsolete though. To the contrary. I think the CPU and GPU rendering are quite complimentary and they each have unique benefits...

Dries

Thanks Dries!

I guess the only way to find out what's better is to compare CPU-based and GPU-based rendering side by side by processing the same 3d scene. Without direct comparison it is not possible to make any correct conclusion. So probably it would be better to wait.

Best Regards!

SK1107

Quote from: DriesV on July 02, 2019, 03:48:59 AM
I can also add that KeyShot 9 will have denoising functionality, which, depending on the scene, can slash render times. It will be available for both CPU and GPU rendering. Network Rendering will also support denoising.

Dries

When will we see KS 9?   (Approximately of course :), e.g. Fall 2019, Spring 2020 )