CPU Performance: Rendering and Synthetics

For the rest of our CPU tests, we’re using a mix of rendering and synthetic workloads. This is slightly different to our previous server reviews, due to some adjustments, and we hope to be running something similar to our standard server workloads in the near future.

All CPUs are run with SMT/HT enabled.

Corona 1.3 Benchmark

 

Blender 2.79b bmw27_cpu Benchmark

VRay Benchmark

AES Encoding

3DPM v2.1 non-AVX (64T Max)3DPM v2.1 AVX2/AVX512 (64T Max)

NAMD 2.31 Molecular Dynamics (ApoA1)

CineBench R20 MT

Even with the addition of a socket-to-socket in the mix, the dual 7F52 setup scores up to +100% in some benchmarks over the previous generation EPYC 7601. Against Intel’s latest 16-core Cascade Lake Refresh hardware, the AMD takes a sizeable lead in most benchmarks (except notably AVX512), which is perhaps to be expected given the price difference and power difference. What is interesting is how in certain workloads, the 2P 7F52 setup can make a reach up for the Xeon 8280s, despite the 8280s being 3x the cost each.

SPEC2006 and SPEC2017 (Single Thread) F is for Fast
Comments Locked

97 Comments

View All Comments

  • 8lec - Tuesday, April 14, 2020 - link

    Defenitely an interesting CPU... Great review you guys. Keep up the great work
  • Gondalf - Tuesday, April 14, 2020 - link

    Is It intersting?? This silicon is absolutely leaky, 240W is a madness for a 16 cores a on 7nm.
    The Intel counterpart is only 205W (6246R) on a crap 14nm.
    Definitively not good at all
  • Fataliity - Tuesday, April 14, 2020 - link

    Thats because, to get that much cache, they are only using 2 cores per chip. So there's alot of redundancy that isn't needed to achieve that level of cache.

    For the workloads this is made for, the power consumption won't matter much. This is more of a part for RTL, silicon design, financial uses, etc. In those businesses, time is money. Much more money than the power consumption.
  • Qasar - Tuesday, April 14, 2020 - link

    i find it interesting that now gondalf is crying about power usage. where was his crying when intel was the power hog ? when intels cpus are listed as being 95 watts, but they use up to 200 watts ? seems he has the : its ok when intel does it, but when amd does it, its outragous. mindset
  • StevoLincolnite - Tuesday, April 14, 2020 - link

    In other words... Just your usual hypocritical fanboy.
  • Qasar - Tuesday, April 14, 2020 - link

    how so ? i kept asking those that were defending intel about its power usage, compared to what amd currently uses. maybe you need to reread what gondalf said, and then what i said
  • ballsystemlord - Tuesday, April 14, 2020 - link

    I think he was referring to Gondalf as the fanboy @Qasar.
  • Qasar - Tuesday, April 14, 2020 - link

    ahh :-)
  • bananaforscale - Wednesday, April 15, 2020 - link

    10900F, TDP 45W, PL2 224W...
  • Gondalf - Wednesday, April 15, 2020 - link

    To me this look like a kamikaze strategy. First of all the wattage matters even in this segment, second one this is a waste of 7nm silicon to match Intel on 14nm, last thing this approach is useless because Intel is shipping server SKUs on demand up to 5Ghz turbo for customers that ask for performance. This cpu line is low margin and unable to seriously beat Intel big superiority in raw core performance.
    In fact right now AMD is below the long awaited 5% of global x86 server market share, they hope to reach this in the middle of this year but they are late a lot in their adventure.
    The manufacturing process is not enough to have a winning horse

Log in

Don't have an account? Sign up now