High end PC/Rendering advice for Maya

[sherridge101]

Hi all

Apologies for the following long list of queries.
We’re looking to invest in a high spec pc and or renderfarm/VCA, so want to get as many things squared away as poss.

We primarily create Automotive imagery for print.
Most of our work is photographic but we have recently invested in Maya to create some simple visuals.

Having access to CAD data, we are now looking to create some of our print imagery in 3D.

The CAD data can be up around 20 million polygons and we’ll required to render at a resolution of around 10K.

The main question is, would one monster pc with multiple 24GB Quadro cards be the way to go? Visualising and then creating the final render in RT GPU?

Or would it be best to buy a reasonable spec pc and link this to a VCA?
The demo on V-Ray with a VCA a while back, looked impressive.

I’m thinking GPU rendering is the way to go for speed, as opposed to CPU rendering, but this could be a back up option (as we have a lot of mac pro’s here in the office).
My recent experience with Octane for Maya was very good, but as this is a GPU only render, I am looking at V-Ray for the GPU/CPU flexibility.

GPU production render queries:
What are the RT GPU production render limitations?
Are there setting for caustics, dispersion etc. and is there a firefly/hotspot reduction tool?
Does RT GPU work with latest HDR Light Studio?

Other machine spec queries:
Would it be best to use Windows 7 or 10?
What amount of RAM would be best? 64GB+
Which cpu setup would be best? Intel Xeon’s?
Hard drives: SSD for Windows and then a separate SSD for project files?

Any other advice would be great, or pointers from anybody working with largish data sets and print res.

Cheers

Steve

[glorybound]

I think that the consensus here is as many cores as you can get, which implies XEON. Also, most would say multiple machines apposed to one beefy one. Personally, I have 128GB RAM, however, I think it's an overkill and 64GB RAM should suffice. As far as GPU, others here would know far more than I do.

[yolov]

What are the RT GPU production render limitations?

There are some that you could look up here: http://docs.chaosgroup.com/display/V...orted+Features
We constantly work to improve and expand the GPU feature set, so more and more features are supported with each new release.

Are there setting for caustics, dispersion etc. and is there a firefly/hotspot reduction tool?

I may need to doublecheck the status on dispersion on the GPU. Caustics are available on the CPU only for the moment.

Does RT GPU work with latest HDR Light Studio?

HDR images are supported by the GPU. To my knowledge, HDR Light Studio just links a HDR image to VRay's environment. Haven't tested it recently on the GPU, but I could give it a try and get back to you on this, although I don't see a reason why it won't work.

Other machine spec queries:
Would it be best to use Windows 7 or 10?
What amount of RAM would be best? 64GB+
Which cpu setup would be best? Intel Xeon’s?
Hard drives: SSD for Windows and then a separate SSD for project files?

There's no specific limitation with V-Ray that we know of in working with either windows 7 or 10.
As gor RAM/SSDs - the more, the better

Have you already tested V-Ray for your projects? It would be appreaciated if you shared any impressions with us. Feel free to post any other questions here.

[snivlem]

As far as Renderfarms go, Google's Zync is pretty nice and the easiest one to use with Maya + VRay. They have monster rigs (64 cores, 128GB RAM) that they added recently for $2.40/$4.49 per hour hire.

Rebus Renderfarm is also very good, but it is strict in that you have to convert a couple of things in the scene before rendering. Recently they finally enabled distributed rendering which is good. Something to consider I guess.

If you go the workstation route, check out this list for CPUs;

https://www.cpubenchmark.net/high_end_cpus.html

I've found it to be accurate with higher points on that chart = faster rendering time. Except for the 5960X which is a beast when you overclock it (gets around 21,000) if that's an option for a cost effective CPU.

[sherridge101]

Hi chaps

Thanks for all of your replies and info.

I haven't tried running V-Ray with our CAD files just yet:
We have Mac’s in the studio (as the bulk of the work is retouching at present) and so I have been using mental ray in Maya for the initial visuals.

The idea is to be able to open, light, shade etc. 20 million poly CAD and finally render out at 10K.
I was initially thinking GPU, as I would like to be able to sit with the Art Director and tweak the lighting etc.
With the GPU now closer and closer to the full on production render, I was considering rendering out the final shot on the GPU also, and worse case opening up the Mac Pro’s in the studio, should we need CPU rendering (for dispersion or Caustics).

There’s a vid on the nvidia site showing V-Ray on a VCA.
Is this the best GPU route, of would a large pc with 4+ cards be the way to go.

The VCA has Quadro cards with 12GB RAM, but with a dedicated pc, we could put in the new 24GB cards?

Cheers again.

Steve

[sherridge101]

Just to update, we're leaning toward the Boxx Apex 5.
This will house up to 5 GPU's.

Any experience with using one of these machines?

[beenyweenies]

Personally, I don't think Quadros are the way to go unless you really need that 24GB vram. Before going down that road, I would suggest that you look into optimizing the CAD meshes, because you could probably drop the poly count quite a bit with no visual impact. Just as in CPU rendering, GPU rendering speed is dependent on the number of GPU cores (in this case CUDA cores) on the card. The Quadro cards usually have far fewer CUDA cores than the GTX consumer cards, and actual GPU rendering performance is way, way slower as a result. You could literally get twice the GPU render speed on a $600 consumer card than you would on a $2,000 Quadro. Nvidia stacks the VCAs with Quadros because they are positioning those cards as the "workstation" solution, and this marketing effort would be undermined by dropping GTX cards into their compute appliances.

I've been doing a fair bit of GPU rendering R&D lately and my personal conclusion is this. The Nvidia GTX 980 ti (6GB vram), and now the new GTX 1080 (8GB vram), is the best option out there in terms of raw bang for the buck. I have a pair of 980 ti cards and GPU rendering is extremely fast. If you dropped 4 of these into a machine, it would tear through renders like butter. If you need more vram the Titan X has 12GB.

The downside is that some things are still a little different on GPU vs CPU. A scene that works well on CPU might look different on GPU, and vice versa. Things like refraction fog aren't there yet, but as Yolov said upstream they are adding new RT features rapidly. I would suggest that you assume you will be doing CPU rendering for now, but position your equipment and R&D efforts for GPU, it's right around the corner.

[sherridge101]

Cool, cheers for the info, beenyweenies.

[FSGFX]

a good read...
https://labs.chaosgroup.com/#/guide-to-gpu-rendering