Hello mates,
I want to update my desktop computer, mainly the graphic card and the processor.
I would like to know what processor and graphics card (GTX/RTX/Quadro) can you recommend me to get the most out of KeyShot when I go to render.
I usually do my models in Rhinoceros or SolidWorks in Windows 10.
The budget I have is around 1200 to 1400 USD.
I remain attentive, thank you very much.
Reggards from Colombia.
Hey Maobros,
We have done some extensive testing on GPU rendering and found with your price range an RTX 2080ti would be the best bang for your buck.
on some levels the RTX 2080ti has performed as well as an RTX 5000, you are loosing a little on Vram with the RTX 2080ti running 11G and the RTX 5000 runs 16G. The RTX 2080ti has performed very well with rendering speeds.
I hope this helps a little with your PC build.
You may also be wise in waiting to see what price/spec the RTX 3000 series cards are going to be and when they are available . . . if you can wait? The new 3000 series should be based on the Ampere chipset ??? but no firm details on them yet, rumoured release is Q4 2020-ish?
I'm considering this as an upgrade myself
best wishes with your build
David
Interesting topic.
I have an extended question on this topic: how reliant is Keyshot on GPU to be able to work fluently on large scenes?
We regularly work with files that have around 250milj triangles - to 20gb+ ram filled when opened - 150000+ parts.
Mostly Keyshot behaves very well. Our PC's are 2990WX with 64gb ram.
However, when interacting in the viewport we sometimes get freezes that the viewport isn't refreshing (and reducing noise).
Would it help to have a better GPU than the GTX1060 I currently have?
Happy my question is a bit clear.
Tnx for help,
Per
I have 4 x RTX 2080 ti and Keyshot works great on them, very fast.
NVIDIA is presenting their new GPUs with Ampere architecture tomorrow (https://t.co/ItASMffNbV?amp=1).
If the rumors are to be true, then these new cards will offer radically improved raytracing performance. Let's wait and see, but I would certainly include the new line-up in any GPU purchasing decision at this point in time.
Dries
Quote from: DriesV on August 31, 2020, 05:31:01 AM
NVIDIA is presenting their new GPUs with Ampere architecture tomorrow (https://t.co/ItASMffNbV?amp=1).
Aaarghhh...just ordered a new custom workstation with the 2080ti...
Quote from: phos4 on August 31, 2020, 11:44:07 AM
Quote from: DriesV on August 31, 2020, 05:31:01 AM
NVIDIA is presenting their new GPUs with Ampere architecture tomorrow (https://t.co/ItASMffNbV?amp=1).
Aaarghhh...just ordered a new custom workstation with the 2080ti...
The RTX 2080 Ti will remain a great card for GPU rendering, but the increased raytracing performance and especially the larger memory (24 GB) is going to make the soon-to-be-released RTX 3090 a very attractive card for GPU rendering in KeyShot.
With RTX 2080 Ti, you can run into situations where a scene with all its geometry and textures simply doesn't fit into the memory. Tying together two cards with NVLink can solve this, but at a not insignificant performance penalty.
I think it is safe to say the 24 GB of the RTX 3090 will be able to fit 99% of KeyShot scenes.
But all of this remains speculation, until after today's NVIDIA event. :-)
Dries
Quote from: DriesV on September 01, 2020, 01:44:36 AM
The RTX 2080 Ti will remain a great card for GPU rendering, but...
That is not really helping to ease the pain... ;)
Well, luckily we work mainly on CE products, than that should be fine....till tonight. :o ;)
I am using an RTX 2080 Ti with KeyShot at home. For the typical product shot scene, it is a great card. You can fit a lot of geometry on that card. And if you are smart about textures (i.e. not 4K or 8K for every single texture), you can fit a whole lot of textures as well.
Dries
would be nice to see some "official" benchmarks for different Nvidia-Cards directly from Keyshot...
in the meantime:
it looks like one could expect already an improvement going from Quadro RTX 6000 to new RTX 3080
within the range of 1.7x up to 2.6x (average >2.0).
So with 3090 this should be typically more than 2x in all cases.
where are theses numbers from?
from iray dev blog, which is the "offical" nvidia render which leverage the same Nvidia OptiX Ray Tracing Engine like Keyshot does...
just have a look: https://blog.irayrender.com/post/628125542083854336/after-yesterdays-announcement-of-the-new-geforce
Yes, RTX 3080 looks like a screamer for raytracing.
Please note though that this card "only" has 10 GB memory, compared to 24 GB for RTX 6000. Also, for some reason NVIDIA has decided against supporting NVLink on RTX 3080. So you won't be able to double the memory pool like you could with RTX 2080 (Ti). Only RTX 3090 supports NVLink.
Dries
maybe there will be an 3080 with 20GB in the not to far future.. (e.g. when Big Navi from competitor AMD arrives or when Micron brings their up-coming double capacity DDR6X Chips to market)
when choosing 3090, you shouldn't be shocked about it's HUGE size, allocating 3-Slots. When thinking about using a pair of these beasts (connected via NV-Link) it becomes really interesting to find a good fitting MoBo and case...
And of course you also need to re-think your power supply - we are talking about 2x 350W 8-D
After seeing the performance improvements (Nvidias presentation) from the 2080ti to the new 3090, I think it is a no brainer to get the 3090 for about the same price. The only question will be, does it fit your case? ...;)
Quote from: phos4 on September 02, 2020, 12:31:28 PM
... The only question will be, does it for your case? ...;)
Of course we will have to test for compatibility with an actual card, but I see no reason why KeyShot wouldn't benefit to the same degree as other CUDA/OptiX applications.
Dries
Ha, sorry, meant to write, but does it FIT your case (will edit)... I am sure (or very positive) that KS will benefit from it.
Besides pure speed for ray-tracing, Nvidia's Ampere brings two new possibilities which may be interesting for rendering:
1. hardware acceleration for motion blur
2. RTX i/o
with this 2., one could stream data very fast from SSD to GPU Memory also in gamer-cards
see https://www.nvidia.com/en-us/geforce/news/rtx-io-gpu-accelerated-storage-technology/
(this will be build into Windows in 2021, see https://devblogs.microsoft.com/directx/directstorage-is-coming-to-pc/).
For Pro cards (Quadro) this was already available in 2019, called GPUDirect Storage see https://developer.nvidia.com/blog/gpudirect-storage/
If this could be used by Keyshot one day,
two (or three) things should become possible:
- reduce start-up times for GPU-Rendering
- store needed data in a compressed way on SSD, to save space and also stream it faster to GPU's VRAM where it is decompressed and used
- maybe solve also some limitation introduced by VRAM-size with streaming data "at the moment of need"