Puget Systems print logo

https://www.pugetsystems.com

Read this article at https://www.pugetsystems.com/guides/1556
Article Thumbnail

V-Ray Next CPU Performance: Intel Xeon W-3200 Series Processors

Written on August 21, 2019 by William George
Share:

Introduction

Intel recently launched a new batch of Xeon W processors - the workstation side of their Xeon family. The previous generation of these CPUs was based heavily on Intel's enthusiast Core X series, with up to 18 cores and using the same socket 2066. This time around, Intel used their server-focused Xeon Scalable line as the basis for the new Xeon W processors, leading to models with up to 28 cores, six memory channels, and 64 PCI-Express lanes. We are testing these chips across a wide range of applications, but the focus of this article is on CPU-based rendering in V-Ray Next and how they compare up to other modern Intel and AMD processors.

For another look at this sort of use case, we also took a look at Xeon W performance in the native Cinema 4D renderer in another article.

V-Ray Next Logo

Looking for a Rendering Workstation?

Puget Systems offers a range of workstations that are tailor-made for your unique workflow. Our goal is to provide the most effective and reliable system possible so you can concentrate on your work and not worry about your computer.

Configure a System!

Test Hardware

Our focus with this round of testing is on the new Xeon W processors from Intel, of which we have three samples. Unfortunately we do not have the top-end model, the W-3275, but we should be able to estimate how well it will perform based on the behavior of the three we can test. For comparison, we have included recent results from a wide range of other Intel and AMD processors as well.

Intel Xeon W Test Platform
CPU Intel Xeon W-3265
Intel Xeon W-3245
Intel Xeon W-3225
CPU Cooler Noctua NH-U12S DX-3647
Motherboard Asus PRO WS C621-64L SAGE/10G
RAM 12x DDR4-2933 ECC Reg 16GB (196GB total)
Video Card NVIDIA GeForce RTX 2080 Ti 11GB
Hard Drive Samsung 970 Pro 1TB
Software Windows 10 Pro 64-bit (version 1903)
V-Ray Next Benchmark
AMD Ryzen Test Platform
CPU AMD Ryzen 9 3900X
AMD Ryzen 7 3800X
AMD Ryzen 7 3700X
AMD Ryzen 7 2700X
CPU Cooler AMD Wraith PRISM
Motherboard Gigabyte X570 Aorus Ultra
RAM 4x DDR4-2666 16GB (64GB total)
Video Card NVIDIA GeForce RTX 2080 Ti 11GB
Hard Drive Samsung 960 Pro 1TB
Software Windows 10 Pro 64-bit (version 1903)
V-Ray Next Benchmark
Intel Core Test Platform
CPU Intel Core i9 9900K
Intel Core i7 9700K
CPU Cooler Noctua NH-U12S
Motherboard Gigabyte Z390 Designare
RAM 4x DDR4-2666 16GB (64GB total)
Video Card NVIDIA GeForce RTX 2080 Ti 11GB
Hard Drive Samsung 960 Pro 1TB
Software Windows 10 Pro 64-bit (version 1903)
V-Ray Next Benchmark
AMD Threadripper Test Platform
CPU AMD TR 2990WX
AMD TR 2970WX
AMD TR 2950X
AMD TR 2920X
CPU Cooler Corsair Hydro Series H80i v2
Motherboard Gigabyte X399 AORUS Xtreme
RAM 8x DDR4-2666 16GB (128GB total)
Video Card NVIDIA GeForce RTX 2080 Ti 11GB
Hard Drive Samsung 960 Pro 1TB
Software Windows 10 Pro 64-bit (version 1903)
V-Ray Next Benchmark

Benchmark Details

We used the latest version of Chaos Group's V-Ray Next Benchmark for this comparison, and it includes tests for both CPU-only and GPU+CPU rendering. This is somewhat novel as most GPU rendering engines do not use the CPU at all - but in V-Ray Next, they have implemented CUDA emulation on the CPU to improve performance a bit. Even the fastest CPU doesn't add as much performance in this mode as a single high-end video card, but who would turn down additional performance during renders for no added cost?

With the information we have gathered, then, we put together three charts. The first is the CPU-only performance, the second is GPU+CPU, and the third is the CPU by itself but in GPU mode. The first and third charts, then, are really the most important when considering which CPU to pick for this application - depending on which rendering pipeline you plan to use. Please also note that the selection of CPU (and motherboard) will impact how many video cards you can use, so if you are going the GPU route you can fit more cards in systems which use higher-end processors.

Results

Here is a gallery of the results from V-Ray Next Benchmark 4.10.02, color-coded based on CPU family. The new Intel Xeon W models are purple, while Intel's various Core models are blue, and then AMD Threadripper chips are shown in orange and AMD Ryzen is red.

Analysis

In CPU mode, shown on the first chart, Intel's Xeon W-3265 took the top spot - and it isn't even the top of the line model in this series! That honor goes to the W-3275, which we don't have a sample of, and it would be even faster... probably scoring in the 35,000 to 37,000 range in this benchmark. Oddly, AMD's Threadripper chips don't seem to perform very well in this particular rendering pipeline.

GPU mode, however, flips the tables a bit. On the second and third charts the CPUs are running CUDA emulation code so that they can mimic the behavior of a video card, and in this workload the Threadripper processors excel. The new Xeon W chips performed on-par with Threadripper models of the same core count, so I would expect the 28-core W-3275 to score a little shy of the 32-core TR 2990WX.

Conclusion

Intel's New Xeon W Processors Get Top Performance in V-Ray Next CPU Rendering

The new Xeon W-3265 was the fastest processor we tested in this roundup, for V-Ray Next's CPU mode, and the W-3275 that we didn't have available would be faster yet. However, they may not be the best choice for everyone because of their high price and the scarcity of workstation-oriented motherboards for them. In terms of price:performance ratio, AMD's Threadripper 2990WX was faster than the W-3245 for less money - and in order to get the higher core count Xeon models that would beat the 2990WX you'd have to spend an additional $1500 or more. On the other hand, Xeons do support higher amounts of memory - so if your projects are complex enough to need more than 128GB of memory these new Xeon models might be a viable option.

AMD Threadripper Processors Provide the Best Performance for GPU Rendering Through CUDA Emulation in V-Ray Next GPU

If you are planning to use the GPU side of V-Ray Next, then the 64 PCI-Express lanes offered by these Xeons are tempting because of how many video cards that will support... but you can get similar multi-GPU capabilities with Threadripper for a lower price. Moreover, if you want to use the CPU alongside GPUs, Threadripper gives better CUDA emulation performance than any Intel CPU we tested.

Looking for a V-Ray Workstation?

Puget Systems offers a range of workstations that are tailor-made for your unique workflow. Our goal is to provide the most effective and reliable system possible so you can concentrate on your work and not worry about your computer.

Configure a System!

Tags: V-Ray, CPU, Rendering, Performance, AMD, AMD Ryzen 2nd Gen, AMD Ryzen 3rd Gen, AMD Threadripper 2nd Gen, Intel vs AMD, Intel, Intel 9th Gen, Intel X-series, Core X, Xeon W
Misha Engel

The Epyc Rome 7502P 32 core ($2400 street price) would be a nice addition for these kind of tests.
Upto 3200 MHz memory (8 channels).
128 PCIe-lanes + 1 PCIe-lane if you can find a motherboard that supports them all.
UMA memory (like the Xeon W).

Posted on 2019-08-22 22:42:27

We've wanted to test Epyc processors - both the original and the recent updated models - but finding a single-socket board in a normal form factor with workstation features (more than two USB ports on the I/O, good PCI-E layout, etc) has proven impossible for us. If you know of such a board, please point me in the right direction! Otherwise, I'm afraid we'll have to wait and see what the next generation of Threadripper processors hold.

Posted on 2019-08-23 16:23:39
Misha Engel

Interesting... I'll pass those along to our qualification team to look at, but only 2-3 USB and no audio is a tough pill to swallow for a workstation board. Thank you for the links, though! :)

Posted on 2019-08-23 18:56:32
Misha Engel

You have so many free PCIe-slots, that you can always put some extra I/O in the system for the $950 you save by using an (32 core) EPYC 7502P ($2.400) instead of the (24 core) W-3265 ($3.349).
We use the AsrockRack for our storage servers and for the pfsense firewall(yep it's overkill, we know).

EPYCD8-2T ~ $700 (with 2 integrated 10 Gbit ethernet controllers (intel X550))
+
StarTech.com PEXUSB314A2V USB 3.1 PCI-e Card – 4 Port – 4x USB-A with Two 10Gbps Dedicated Channels – Expansion Card – USB 3.1 card ~ $150 or equivalent.

+
EVGA soundcard ~ $300 or equivalent.

Posted on 2019-08-23 21:22:33

If all you are concerned about is CPU, then yeah PCI-E devices could solve the I/O. One of my big hopes for Epyc, though, was huge CPU plus multiple GPUs - and you'd have to drop at least one GPU in order to add some I/O via PCI-E slots :/

Posted on 2019-08-23 21:23:58
Misha Engel

That was also one of our concerns, till we found out that you can use PCIe-3 extensions cables upto 600 mm without loosing signal integrity, you are of course free to use the OCulink (2 of them) which are 4 PCIe lanes(each) with a different connector, or one of the M2 slots which are also nothing more than 4 PCIe lanes with a different connector. Those ad in cards are only 4 PCIe lanes each and you can use the full bandwidth.
(Linus tech tips goes even futher and makes the GPU's, "6 of them", water cooled 6xsingle slot).
And by the way, the 7702P (64 cores) is also something against the similar priced W-3275 (28 cores).

When you want to use more than 4 GPU's, go for a server rack

Posted on 2019-08-24 00:58:21