I have recently build a new render node only to discover issues with the Vray render speeds. Per core number and clock speed it shuld render 5x faster. Also my two year old Intel workstation is only 10% slower and 5x cheaper.
New Node: Dual AMD EPYC 7401 24core/48threads 2.0GHz (96 threads x 2.8GHz when rendering temp 35c/95f)
Supermicro H11DSi-NT
64 GB RAM 8x8 DDR4 ECC 2666MHz
Samsung SSD /Win10 /Vray Next
VRay benchmark: 39-44sec
Cinebench score: 5900
There’s no way that your CPU temp is 35*C when rendering. I would expect 60*C at least. Unless what you’re seeing is delta temps meaning ambient temp of the room (~25*C?) + 35*C = ~60*C.
The Cinebench score of the render node seems to be in line with its power and capabilities, but the Vray benchmark time is not. My 1950X completes it in 42 sec. and it’s a single 16c/32t CPU. Your score should be ~25 sec. judging the score of this person which has tested a single EPYC 7401.
There’s no way that a i7-6950X, a 10-core CPU, is giving you 45 sec. in Vray benchmark, almost the same score as a 16-core 1950X, unless you’ve overclocked it to like 6-7 Ghz with liquid nitrogen. The average Vray score for a i7-6950X in the results page is almost double that at ~70 sec.
What OS are you using, Win 10 Pro? Did you do a fresh install of the OS or you just transferred the OS disk from an old machine to the new one? I remember someone saying that for best performance it’s best to do a fresh install of the OS whenever you get a new machine.
Can you post a screenshot of the Performance tab of your Windows Task Manager while Vray Benchmark is running? Make sure the graph in there is set to Logical processors not Overall utilization so we can see the load on each of the cores.
BTW: The CPU power in Ghz is not calculated by multiplying the logical cores (AKA threads) but the physical cores. So your math should be 48 x 2.8 Ghz.
Hi Alex_M,
Thanks for your response, you are right that my Intel workstation is slower than 45, just did another test and it is showing 68-70sec. However new AMD EPYC Node is definitely running way slower than what it should.
Vray benchmark is showing 41-44sec
Cinebench is showingscore around: 5200
Here are few benchmark screen caps showing temps at the end of the run. (~35C)
It is running on Win 10Pro fresh install nothing else on the machine.
Ok, try this. In start menu type “power plan” and click on “choose a power plan”. See if the power plan is set to “High performance”. If it’s not, set it to this one and see if helps (If it doesn’t help set it back to the old plan).
Strange. There are some Epycs benchmarked and all of them seem to run fine judging by the results, even the dual 32-core systems (128 threads) - V-Ray Benchmark | Chaos so there’s something going on with your setup IMHO. Maybe you already did that but make sure that you’re using the latest version of Vray Benchmark.
After Bios update AMD dual EPYC (48 cores x 2.8GHz) it is still slow, well over 43 seconds to render Vray benchmark. (1.6x faster than my Intel workstation and should be around 4X) . In Cinebench difference is 3x witch looks closer to the performance numbers.
I’ve tried to install the win Server2016 and my performance did not change at all, maybe lost 5% from Win 10 Pro. I also tried to compare the evermotion sceen render with my intel setup and the higher cpu count did not help much. For example, a 10core/20thread Intel did it in 10min and the 48core/96thread AMD did it in 7:32min. It reflects roughly to Vray benchmark stats.
The only thing that I could think of to try at this point would be to add more RAM, I have 64GB in 8 slots from 16 available. I do not know if that would make any difference and the ECC ram is pretty expensive …
Have you tried disabling SMT(hyperthreading)?
See how it performs with half the cores?
Also, changing mem speed and layout(2 dims, 4 dims) change anything?
OK, Got some progress here, when I turned off SMT (hyper threading) VRAY benchmark went from 45sc to 30 sec and Cinebench went from 5300-3000. Memory is set at 2666 as default. What does this mean right now? It shows 48 cores total and it is faster than before when it was showing 96 treads?
It tells you that there is a issue somewhere in the system with distributing tasks for certain software, if other users with the same setup get a different score its not in vray but in the setup of your system.
Have you installed all proper drivers/frameworks/updates?
Everything looks good as far as drivers go. When testing on Cinebench benchmark using hyper-tread option it shows big upgrade, score goes from 3000-5200. So it looks like Vray has issues with AMD hyper-treads?
Has there been any progress determining the rendering issues with EPYC?
We have built a new workstation based on EPYC, and it’s at least twice as fast as the previous workstation based on dual Intel XEON E5-2690 v3, 64GB RAM, Crucial 1TB SSD MX200 SSD in pretty much all cpu intensive tasks including Cinebench and Carona benchmarks, however Vray is not showing the same performance advantage.
Generally rendering tasks are similar or slower than the much older XEON, same with running the Vray benchmarks, both 1.08 & 4.10.07, so the issue isn’t fixed for Vray Next as far as we can tell.
To update our investigation we have just updated the bios for the H11DSi-NT motherboard to the latest 1.3 (H11DSI9.625) version and we have seen a 50% performance increase in Vray
We have the same experience here. New render nodes with Dual Epyc 7281 have rendering performances close to our previous / regular nodes (Dual 2630 V4).
We tried the CorePrio util (supposed to improve windows scheduler / NUMA performances for single CPU system) with no noticeable improvement.
I’ll try updating Bios to see if there is improvement and I’ll report here if there is improvement.
We just ordered a dual EPYC machine couple of days ago, it will take a bit of time to get it in the office, but then we will be able to do some profiling.
