Do Tesla cards deliver performace boosts for V-Ray that justify the high price?

[Booska]

I'm in the market for a new system and have a budget of around $10k-$12k. I'm considering a dual processor motherboard (Xeons?) and a Quadro K6000 for running Max, V-Ray, Octane, Krakatoa, Real Flow, etc. What kind of performance enhancements could I expect if I added a Kepler card to the pile, say a C2075 or a K20? And would it be worth the extra cost?

Also, I have a Quadro K5000 right now. Could I use it as a render slave in another machine for V-Ray?

Navigating all the complexities of figuring out the best system setup is very tricky and time consuming. Any advice would be much appreciated!

[savage309]

Also, I have a Quadro K5000 right now. Could I use it as a render slave in another machine for V-Ray?

Yes (you will need render node license for the second machine, however).

For the performance of the GPUs, we maintain a benchmark scores here https://docs.google.com/spreadsheets...A0k/edit#gid=0
We have Benchmark #1, which is done using V-Ray 3.1, and Benchmark # 2, which is dones using V-Ray 3.2 and much higher settings (to make it render longer).
Just keep an eye on the driver versions used as well.
Other than that, V-Ray scales almost perfectly linear. And has no problems in using GPUs from different architectures (as long as the driver can see them, but this usually is not a problem).
Also we don't use double precision (which is the main advantage of Tesla at the moment), so you will have no benefit at all if your GPU has it (so Maxwell GPUs should be the best, I think).

[Booska]

Thanks, Blagovest. I just ran it on my machine

"Also we don't use double precision (which is the main advantage of Tesla at the moment), so you will have no benefit at all if your GPU has it (so Maxwell GPUs should be the best, I think)."

I couldn't quite follow what you said here. Could you please clarify?

[super gnu]

means dont bother with a tesla, as the main reason for the extra expense, is to have a feature that vray doesnt even use.

imho get geforce cards.. much better bang for the buck.

[savage309]

I couldn't quite follow what you said here. Could you please clarify?

When doing calculations, you can do them in half (16bit), single (32bit) or double (64bit) precision. V-Ray uses only 32bit precision for the calculations (and we have no plans to change that ever).

We support 3 nVidia architectures - Fermi (GTX 4XX, 5XX), Kepler (6XX, 7XX, K40, K80, K6000 and most of the GPUs starting with a K) and Maxwell (750, 750ti, 9XX, Titan X, M6000). In Maxwell, they have dropped the support for double precision, they have used that space for more single precision units and they have added other improvements too.

Since Tesla cards are used in domains that require double precision (for example, medical computing), nVidia will keep them with the older Kepler architecture. There are also other Tesla specific features (like error checking and correction of the calculations), but V-Ray do not any of these too, since they do not make sense in a raytrace engine.

Just check the render speed of the GPUs in the benchmark page, it can give you pretty accurate feel about the render speed of the different GPUs.

[Booska]

Thanks for the info, guys! So if I were to get something like a GTX Titan X for my new machine, there wouldn't be any compatibility issues using the Quadro K5000 in a render slave machine? Provided I purchase an extra render node license, that is. By the way, how much is the extra render node license?

[savage309]

Thanks for the info, guys! So if I were to get something like a GTX Titan X for my new machine, there wouldn't be any compatibility issues using the Quadro K5000 in a render slave machine? Provided I purchase an extra render node license, that is. By the way, how much is the extra render node license?

V-Ray can render with mixed architectures just fine - they even don't have to be in different machines. So if you put both the GPUs in one machine V-Ray will render just fine as well. This way you can go without extra render node license, too. You have to check if the driver would work with those two GPUs in one machine, but from my experiance usuallly there are not issues with that.
If you prefer two separate machines, here are the render node prices http://www.chaosgroup.com/en/2/vray_licensing.html (tab UNIVERSAL RENDER NODES). If you have older V-Ray, you can contact the sales guys and ask for render-node update price (should be lower).

[Booska]

Thanks again, Blagovest! And thanks for your patience. I'm sure you must get asked these kinds of questions a lot.

[savage309]

No worries, we are here to help. Should you have any questions please do not hesitate to ask (both in the forum and at support@chaosgroup.com).

[werticus]

dual socket cpu's are not really worth the money unless your running out of space for computers, more single cpu computers are better price vs performance.

quadro cards are definitely not worth the money offering zero benefits what so ever to a 3dsmax user for either rendering or the very very marginal view port speed increases vs 4x the price!

Sell your k5000 and get a couple top end consumer level gamer cards.

I've done extensive real world testing with k5000 vs 290x and 780 ti video cards. Look up some posts I've made on these forums regarding this. (RT and view port testing)
Conclusion: nearly zero performance difference vs 4x price difference

and I've done the numbers for xeon 8 - 12 core cpu vs 5960x 8 core i7 extreme based on other peoples posted benchmarks from these forums and you are just better off with more single cpu systems than multi cpu systems every time.
Conclusion: 1.5x better cost to performance ratio for single cpu system when overclocked.

[Booska]

Thanks, Werticus! That sounds like some good advice. I've been planning to pick up a Titan X or Z for my new machine. Haven't decided which one, though. One question, though. How about noise? I've heard the fans in the Titans can be pretty noisy when under load. My K5000 just quietly hums along no matter what I throw at it. But I do like to keep a temp app open on the desktop while I'm rendering a heavy scene because it runs a bit hot sometimes.

I couldn't quite follow what you were saying about the Xeons, though. Are you saying that they aren't worth the extra money either?

[savage309]

Keep in mind that the Z is actually 2 GPUs with 6 gigs each (you can load scenes that fit in 6 gigs on it, not 12 like the X).

[Garryclarke]

Regarding noise.

I've just got a new machine, a Dell Precision Tower 7910 (dual proc) with a 12gb Titan X graphics card. It's probably the quietest PC I've ever owned. Even rendering for 12hrs straight I can hardly tell that it's on. I work at home in a quiet area and I'm quite sensitive to the noise a PC makes. I've not had time to put the card through it paces yet.

[Booska]

That's great about the low noise. My K5000 is very quiet, but I just picked up two 980 Tis for my new build and I've heard they can get pretty shrill when maxed out.

But it will be a while until I can actually put my new machine together because my wife and I are currently without a home. We just sold our house and will move to Ireland in Sept. I wanted to buy my components here in the U.S. because they are cheaper than in Europe. I'm dying to jump in and build it, but there is still so many other things that need to get done before we leave. So I'll have to settle for delayed gratification.

[joconnell]

Good call - everything's bloody expensive in ireland - which part you moving to?