Puget Systems print logo

https://www.pugetsystems.com

Read this article at https://www.pugetsystems.com/guides/1214
Article Thumbnail

V-Ray CPU Rendering: AMD Threadripper 2990WX Takes the Single CPU Performance Crown

Written on August 13, 2018 by William George
Share:

Introduction

V-Ray, from Chaos Group, is a widely used rendering engine for creating realistic 3D graphics. It includes two versions: one for rendering entirely on the CPU and the other for rendering on NVIDIA GPUs. The GPU version mostly depends on how many of which video cards are in a given workstation, with little regard for other system specs, but the CPU version is heavily impacted by both the processor's clock speed and core count.

Chaos Group also publishes a benchmark utility for V-Ray, which can test both CPU and GPU performance. This is a fantastic utility for comparing different processors and video cards, and we run it here at Puget Systems on every workstation we build. We also run it on new and upcoming hardware in Labs, to see how well various components should perform with V-Ray rendering. For this article, we used it to measure CPU-based rendering performance on AMD's new Threadripper 2 processor: the 32-core 2990WX.

AMD Ryzen Threadripper CPUs

Test Setup

This is AMD's second generation of processors in the Threadripper line, so to see how it performs in V-Ray we are also taking a look at the flagship of the first generation: the 16-core 1950X. For comparison we've also got several models from Intel's Core i7 and i9 series, as well as a few dual-CPU configurations using the Xeon Scalable line. Some of these results were pulled from previous V-Ray articles.

All of those CPUs were run through the free V-Ray Benchmark utility, in CPU mode. If you would like more details about the full hardware configurations involved in these tests, .

Benchmark Results

Here are the results for the various individual CPUs we tested in V-Ray Benchmark 1.0.8, with AMD models in red and Intel in blue:

V-Ray CPU Benchmark 1.0.8 AMD Threadripper vs Intel Core i7 and i9 Processors

To avoid too complex of a chart, we separated out the dual-CPU Xeon results. They are shown below, with the same single-CPU AMD Threadripper chips in red again - and, of course, Intel's Xeons in blue:

V-Ray CPU Benchmark 1.0.8 AMD Threadripper vs Dual Intel Xeon Scalable Processors

Analysis

AMD's new Threadripper 2990WX gives an amazingly good performance in V-Ray! That first chart shows it taking 46% less time to complete a render compared to Intel's Core i9 7980XE, even though the 2990WX costs 10% less ($1799 vs $1999) than Intel's flagship. It also beats the previous-generation Threadripper 1950X by about 72%.

Looking at the second chart, Intel can take back the absolute performance crown by ramping up to two physical CPUs with the Xeon Scalable series... but that doesn't tell the whole story. Yes, a pair of Xeon Gold 6148 chips will outpace a single 2990WX from AMD - but those Xeon processors costs over $3000 MSRP each! The less costly pair of Xeon Gold 6136 chips was actually slower than the 2990WX, even though they still cost over $2000 each - making them more than twice the price of AMD's faster solution, and that is before you factor in a larger and more expensive motherboard, case, power supply, etc.

Conclusion

As demonstrated above, the new AMD Threadripper 2 processor takes the crown for top single-socket performance in V-Ray's CPU benchmark. Not only is the 2990WX the fastest single-processor solution, it is also less expensive than many slower options from Intel (both on the single- and dual-CPU sides). This is a bit of a niche, where all 32 cores in this chip are fully utilized, but for CPU-based rendering, the only way to beat AMD's Threadripper 2990WX at this point in time is by spending many thousands of dollars more.

You can see the specific configurations we recommend for V-Ray (both CPU and GPU engines) below. The 1 CPU / 1-4 GPU variant uses AMD's Threadripper processors and should be updated to offer the 2990WX once it passes our hardware qualification process.

V-Ray Workstations and Render Nodes

1 CPU / 1-2 GPU
Compact

Configure


1 CPU / 1-4 GPU
Tower

Configure


2 CPU / 1-4 GPU
Tower

Configure


1 CPU / 1-4 GPU
1U Rackmount

Configure


Tags: CPU, Rendering, Chaos, Group, V-Ray, Performance, Processor, Intel, AMD, Core, i7, i9, Xeon, Scalable, Threadripper, Threadripper 2
glassworks

Thank you for Vray&Cinema benchmarks, any chance for maya/arnold updates? Cheers!

Posted on 2018-08-13 16:35:24

TR2 is going to be fantastic for Arnold, and any other purely CPU-based render engines. The downside is the low single-threaded performance (compared to Intel processors)... so those who mostly work in Maya (or 3ds Max, Cinema 4D, etc) and do *not* render very often would be better off with a higher clock speed processor instead. The C4D counterpart to this article talked about that a bit, and we will continue to test other software packages as well... though I'm not sure if a Maya / Arnold article will be coming any time soon.

Posted on 2018-08-13 16:43:01

Devastating performance in V-Ray. Better performance at $1800 over Intel's $5000 dual 6136 setup (as the author mentioned). And the dual setup with 6148's at $6000+ is only marginally faster. One could actually build three 2990X systems for the price of dual 6136 setup.

Posted on 2018-08-13 18:04:27

Yup! Threadripper 2 (the 2990WX, specifically) is amazing for CPU-based rendering. I suspect it will be great in almost any application where all 32 cores can be put to good use, as long as there are not long periods where the software is single-threaded. For rendering, that would only happen between frames in an animation - and that should be pretty brief. It gets trickier in other situations where you need fast single-core performance alongside lots of cores. We still have several software packages to test with this processor :)

Posted on 2018-08-13 20:44:33

As always, looking forward to your guys' reviews.

Posted on 2018-08-13 20:53:25
Niko Nikolov

Question about the single threaded performance.I will use the word "gaming" as to refer to single threaded vieport navigation in 3d software.
In the first gen threadripper there was a strange thing that was observed when combining a threadripper with a radeon gpu and nvidia one.If someone gamed or in our case 3d model and navigate using the tr4 and nvidia gpu,they got lower fps which was fixable just by xmp the ram higher.But if someone used a tr4 cpu with a amd gpu, than there was no performance loss.No need to bump the xmp.

Do you think the same thing still apply with the new threadripper?Using the 2990wx with a amd gpu just to keep the single thread core performance high?

Posted on 2018-08-13 23:38:41

I just swapped the GTX 1080 Ti we had been using on our 2990WX test build for an AMD Radeon RX Vega 64, and then re-ran Cinebench and V-Ray Benchmark. The results were within 1% of the previous data (shown in these two articles) - so I don't think the use of an AMD vs NVIDIA graphics card is having any impact on the performance of this new processor.

Posted on 2018-08-14 17:24:33
Niko Nikolov

Thank you so much for this first hand info ! I guess, no witch hunting this time

Posted on 2018-08-15 00:04:03
Hwgeek

Thanks for the great article!
What about enabling PBO(Precision Boost Overdrive)? it'n not like manual overclocking and stability issues, it's just increased the TDP limit of the CPU when using good cooling.
with PBO it would get to 2X Xeon Gold 6154 performance level!.
Thanks!

Posted on 2018-08-13 19:02:23

We generally don't test with any overclocking features, aside from the basic Turbo type features that are part of the processor itself (and enabled by default). I wonder, though... I may go back and check the MSI motherboard we used for testing, to see if that feature is already on by default. We did run into an oddity on another article (Cinema 4D single-threaded performance) when comparing the MSI board to an older Gigabyte board. I wonder if that could be related to this 'feature'...

Posted on 2018-08-13 20:47:59
Hwgeek

If I am not mistaken this PBO is new feature part of the new TR 2000 line that's just "unlocks" the CPU performance restricted by TDP so it can run like real 4X Ryzen 2700X :-), looks like the Stock 2990X is on "ECO" mode to be just 250W TDP and AMD embedded option to unlock the Beast with PBO and it only asks for upgraded cooling(Each motherboard will have different affect since they got different max settings for PBO), Since I know you build PC's like NASA - I think you can test this new feature and test if it's stable for your clients and useful, 20% boost on 32 core CPU is like adding another 6 cores - OMG it's just crazy :-).
https://www.youtube.com/wat...

Posted on 2018-08-14 05:13:35

I was just researching PBO, to see how to enable it on our test system, and I found this on HardOCP (usually a trustworthy site):

"Using PBO however voids your CPU's warranty if you use it. Think of PBO as PB2 on steroids, as PBO allows you to tell XFR2 to go outside of the manufacturer's suggested power and temperature guidelines."

(source: https://www.hardocp.com/art...

Based on that, I won't be testing it. For our purposes - selling highly reliable workstations for professional users, which we stand behind and support for the life of the system - something that voids the manufacturer warranty is a no-go. I'm sorry.

Posted on 2018-08-14 17:19:47
Hwgeek

Thank you.

Posted on 2018-08-14 19:46:59
Hwgeek

What about using faster memory?

Posted on 2018-08-17 08:58:09

We are working to find out what the officially supported memory speeds / configurations are for Threadripper 2. I know that a lot of folks use higher speed memory, but AMD themselves specify certain speeds based on a combination of factors - including the number of memory modules being used. We have that data for the more mainstream Ryzen 2 CPUs, but not yet for Threadripper 2: https://www.pugetsystems.co...

Plus, given how well the 2990WX is already doing in CPU-based rendering, faster memory wouldn't change our conclusions anyway. Okay, maybe it finishes the V-Ray benchmark one second faster... that still doesn't impact where it falls compared to the other CPUs we tested. Its a great chip for well-threaded applications, even with 2666MHz - faster memory might make it marginally better, but wouldn't have a massive impact.

Posted on 2018-08-17 16:57:03
AP

TR3 is going to be even better :) "7nm" (reality prob 10nm/compared to intel) tech etc. There seems to be very little push in the net for Microsoft to increase the core limits on their windows 10 operatiting system considering the pace that AMD is catering cpus their not so far of new cpu could easity be more that 64 threads that the MS 10 pro is limited to (that is my understanding of the MS windows 10). MS sure does not have the know how or money to implement support for more than 64 threads when you think of it :) . Thanks for this test.

Posted on 2018-08-18 17:19:59
maratropa

win 10 has a processor group limit of 64 threads, after which any further cores will be put in a new processor group. (which can be probematic for some software) They support more than 64 threads fine, (I have 72 threads, although i turned off 4 cores). Although efficient use of so much cores is another story. Phoronix 2990WX review showed that linux often performs better with this number of cores..

Posted on 2018-08-20 10:03:11
ComputahNerd

Hi, are you planning to test the 2950X as well, now that it is out? It would be really awesome to see how well that CPU does vs. previous generation. Particularly in single core programs like Adobe AE and Premiere, since that is were the 1950x did quite bad in your last year tests. You guys are the only company that does these real life tests which I know of :D

Posted on 2018-09-05 05:33:45

After Effects and Premiere Pro articles for Threadripper 2 are coming soon. We're just waiting on our 2950X to arrive so we can finish up that testing.

That said, based on what we have seen so far with the 2990WX and the older Threadripper CPUs, none of them will be particularly good for AE compared to similarly prices Intel CPUs. That program is simply too geared towards single core performance which is where Intel has a strong lead. Premiere Pro, on the other hand, I expect AMD and Intel to be pretty much neck-in-neck so you could happily go either way.

Hopefully we'll have the benchmarks and articles up sometime next week so we can definitely say how those CPUs perform in those applications.

Posted on 2018-09-05 05:40:41
ComputahNerd

Thanks! yeah but in your previous benchmarking of TR and i9, 1950x didn't do very well, even with C4D rendering in AE - and the 7980XE scored better overall both in Standard and C4D projects in ram preview. But the latest 2950X has a base clock of 3.5GhZ (compared to 7980XE which has 2.6GhZ) and both CPUs tuboboost to 4.4GhZ. So in theory the 2950X should perform better than the 7980XE and get a lot better single threaded performance than its predecessor. Unless AE and Premiere works really bad with AMD CPUs....which is a an interesting thought.

Posted on 2018-09-05 05:51:15

That may be what happens, but keep in mind that clock speed between AMD and Intel (or even between different generations within each brand) can give vastly different performance. So trying to compare based on clock speed, core count, or any other singles specification is often not going to be accurate. That is really one if the big reasons we do all this real-world testing. It simply is the only way to know exactly how different CPUs and other hardware components perform in reality.

Posted on 2018-09-05 05:55:22
ComputahNerd

Thanks, very appreciated - you guys should receive review products from all the vendors. The tests you guys are doing are so much more important than testing the CPUs through various benchmarking software - which all the youtubers do and tricks you into buying wrong products for what they are intended for.

Posted on 2018-09-05 05:57:26

We actually have great relationships with AMD, Intel, NVIDIA, etc. and they regularly send us hardware for our testing. Honestly, I don't know if we could do anywhere near the testing we do without those relationships since buying one (or more) of every CPU, GPU and other hardware would be prohibitively expensive. The 2950X just unfortunately didn't get here as quickly as normal.

Posted on 2018-09-05 06:03:46
ComputahNerd

ah ok - good to hear.

Posted on 2018-09-05 06:18:23
ComputahNerd

I can understand that the 2990WX is a really bad choice for AE and Premiere though, it's extremely high core count and a very niche product. But I was hoping 2950x was closer to an all-round CPU than the previous Gen of TR4.

Posted on 2018-09-05 05:54:18

I think the main question will be how much of an impact the higher clock speeds on the 2950X (compared to the 2990WX) have. The "max boost clock" is 0.2GHz higher, or about 5% faster, and the "base clock" is 0.5GHz higher - about 17% faster. If single-core performance is only 5% faster, that won't be enough to bring it up to parity with Intel's processors. We'll have to wait and see what the testing shows :)

Posted on 2018-09-05 16:47:17
ComputahNerd

wasnt the max boost clock of 2950x 4.4ghz, and 1950x 4.0ghz. So 0,4ghz more. I'm definately sure it won't be on par with 8700K, but it will be interesting to see if it matters at all in After Effects. Especially their new xfr algorythm. Because the biggest drawback on 1950x is that After Effects feels more sluggish and slow, compared to Intel Processors of the same clockspeeds - which is kinda weird. But hopefully 2950x slightly improves that.

Posted on 2018-09-05 17:05:36

Oh, I was comparing to the 2990WX (since we've already been able to benchmark it) when I was listing those clock speed differences. I don't recall the 1950X's clocks off the top of my head, but you are probably right :)

Posted on 2018-09-05 17:08:18
forumer

What model of motherboard did you use on the test? Can you make the benchmark of overclocked vs stock please?

Posted on 2018-09-11 13:54:36

The motherboard used for the Threadripper processors in this article was the MSI MEG X399 Creation. It, and the other hardware used, is listed under the Test Setup section above - just click on the link toward the end of that section that says "click here to expand the following section". Each component is not only listed, but also linked to detailed specifications.

As for overclocking, though, we avoid that here at Puget. Our focus is on highly reliable workstations, so that our users have a smooth and stable experience when using them, which means that we do not push components beyond manufacturer specifications. Doing so would increase failure rates and potentially void warranties, which would then impact our ability to quickly repair or replace systems if they did fail.

Posted on 2018-09-11 17:16:26
forumer

Thanks William!

Posted on 2018-09-12 11:12:15
Petar J Petrovic

Would it be possible to get V-Ray benchmark for 2990WX with ported Cuda code? If I remember correctly, in an older artice, you did just that with 1950x and it significantly boosted cpu rendering time clocking the speed on the level of one 1080 Ti card.

Posted on 2018-09-28 21:51:54

That was a look at the combined CPU + GPU functionality found in V-Ray RT 3.6 (and I believe also included in V-Ray Next GPU, the release after 3.6). It was a bit trickier to set up, though, and since Chaos Group has said that they are going to update their V-Ray benchmark to use the V-Ray Next engine I have sort of been holding off for that. We do still use the older V-Ray benchmark for some basic stuff, and if they keep dragging their heels I may go back and re-create that test... but I haven't done so yet, and it isn't super high on the priority list. It is testing a different version of V-Ray's engine anyway, so the CPU-only result will still be useful for folks running V-Ray Adv / V-Ray Next CPU (instead of the RT / GPU versions).

Posted on 2018-09-28 22:09:08
Omar Bianchino Quattropani

Hi Everyone, I'm new here and I have a question I hope someone can answer.
I have this configuration:

AMD Ryzen Threadripper 2990WX 32-Core Processor, # of logical cores: 64
GeForce RTX 2080 Ti 11264MB
NVIDIA driver version: 416.34
ROG Zenith Extreme
gskill 2x 16 gb ram DDR4 3600
Microsoft Windows 10 Pro
V-Ray 3.57.01

I run the vray benchmark test and with the cpu i score 33 sec avarage. I don't know why everyone with my same cpu score an avarage of 22/25 sec.
I hope someone can help me to figure this out.
Thank you very much and sorry for my basic english skill.
Thank you again

Posted on 2018-11-09 03:28:31

Do you really only have two sticks of memory? You listed 2x 16GB, and if it's true that your only populating two memory slots on your motherboard then you're only getting half as much memory bandwidth as the Threadripper processor is designed for. These CPUs have quad-channel memory controllers, which means they're designed to run on 4 or 8 sticks of memory at a time. AMD processors tend to be very sensitive to memory bandwidth, so that would probably explain it right there.

Posted on 2018-11-09 03:32:33
Omar Bianchino Quattropani

Thank you so much for your fast response,
I'm gonna order right now 2 more 16 gb stick of ram and the remaning next month ( Because I run out of budget XD )
Thank you again, I'll repeat the test as soon as I add the new ram and I'll post here the result again.
Thank you so much

Posted on 2018-11-09 03:47:54
Omar Bianchino Quattropani

Hi William, I did the test again with 64 gb ram ( 4 x 16 ) and it went from 33 sec to 27 sec.
So I'm still a bit far from the 22/23 sec that I see online.
Do you think that with 128gb in total (8 x 16 ) I could low the seconds from 27 to 22?
Or I'm still doing something wrong?
thank you very much

Posted on 2018-11-12 05:05:09

Once you have filled all four memory channels (assuming you used the proper slots on the motherboard - double check the manual for your specific board) it won't get any faster if you double the number of memory modules in each channel.

27 seconds is very close to the result we got for this CPU (26 seconds, as shown in the charts above). I haven't looked around on other sites to see what might be different on systems where folks are saying they got 22-23 seconds, but it is possible that they have overclocked their CPUs or are using even faster memory. You might want to look at the places where you found those scores and see if you can identify what differences there are between their system specs and yours.

Posted on 2018-11-12 18:27:19
Joey Rodriguez

SketchUp + Vray: 1x TR2 2990wx VERSUS 3x Ryzen 1700x (in Swarm)?

Posted on 2019-03-13 15:20:40

We haven't done any cluster / swarm style testing, but I think there is a good chance the 2990WX would still be in the lead. The 1700X has 8 cores, so 24 between three of them, which is out-classed by the 32 cores in the 2990WX. The 2990WX is also a generation newer and has a higher single-core turbo speed than the 1700X. Moreover, when you consider that you only need to pay for one chassis, motherboard, power supply, video card, set of RAM, etc... I think the 2990WX might also end up costing less than three full 1700X systems. Not to mention the lowered complexity, the ability to run applications that don't have cluster / swarm functionality, reduced licensing costs (only one copy of Windows & other software instead of 3), etc.

Posted on 2019-03-13 16:51:02
NBNA

I spent a lot of time overclocking the 2990wx with every solution I could find on the internet. The number one thing that helped my rendering times was simply lowering the voltage of my ram to 1.32v. Huge difference for me and I dont know why? Almost everything else I tried either gave very little improvement or was slower. But I m happy as this new PC is a tad bit faster then my older dual xeon 2996v3 that costed $2000 more. Maybe 128gb ram filling all slots is a bottleneck for the 2990wx 32 core? One more thing I am using the Vengeance 2933 ram and Fatality x399 motherboard, but the 2666 setting seems to be the most stable right now. I have a feeling I did not get the best die in the tossup.

Posted on 2019-08-20 11:06:38