I’ve been looking at some used older EPYC chips (Zen and Zen2) and those price to core count ratios sure seem interesting. However, there seems to be a complete lack of any benchmark information on EPYC chips in Unreal Engine related workloads such as compiling the engine itself, shader compiling and light builds. Based on the benchmarks I’ve seen, core count seems to be what you want. Has anyone tried running the Unreal Editor on EPYC and would any of them even be faster than a modern desktop chip, say a 5800X?
1 Like
I cannot speak for Epyc CPUs, but I do use it’s cousin, a Threadripper 3970X. I can say having 32 cores available has significantly improved engine & shader compilation. I have not tested any lighting builds yet, but compared to my previous setup which was an Intel i9-9900X, compile times dropped significantly. Generally I can compile the engine in a fraction of the time it takes some of my coworkers.
1 Like
Threadripper would be my first choice also for the clockspeeds but they are significantly more expensive for me.
Something that AMD doesn’t necessarily advertise is that you can disable cores or CCDs in your BIOS to get higher single-thread performance from the remaining cores. This is particularly helpful if you’re building a mixed use system, like for real-time & offline rendering.
For example, on my 7713P machine, by disabling 32 cores the remaining cores get 2X the power and run at a higher clock speed. With 64 cores enabled, each core gets about 2.15 watts. With 16 cores enabled, each core gets about 6.75 watts and all cores run at the full ~3.7 GHz.
I personally wouldn’t buy any Zen1 or Zen+ processors (especially something like the 2990WX) as the CCD / CCX / MCM architecture hadn’t matured at that point. Zen3 would be my first choice for Unreal due to the larger unified L3 cache, but something like a 7742 will clock high enough (and indeed, I have used HCC Zen2 and Zen3 with Unreal, Blender, Substance, and various offline renderers for the past 2 years).
For code compilation, I find that 32 cores works best. For rendering, 64 cores. For pure viewport/game performance, I drop down to 16 cores. You’re not going to match a 5800X in entirely single threaded tasks, but very few content creation tasks are entirely single threaded.
2 Likes
there are other benches to be found on pugets site also. that contain TR numbers in the comparisons. i just couldnt find one with unreal 5.
seemingly they kick ass.
although the gpu performance on amd cards leaves a little to be desired the engine works well enough on amd cpu’s to show up intel.
Interesting, I’ll try this on my 7443p. Thanks for the information.
This topic was automatically closed 273 days after the last reply. New replies are no longer allowed.