Comparing CPUs based on multiplying the Thread Count by the Single Thread Rating

[humbug]

Hi there,

let's assume we are in a truly multi-threaded workload, running multiple VMs or containers running on the same machine.


AMD EPYC 7401P
- https://www.cpubenchmark.net/cpu.php...+7401P&id=3118
- with 24 Cores / 48 Threads
- an average CPU mark of 19555
- and a Single Thread Rating of 1488

Intel Xeon E3-1271 v3
- https://www.cpubenchmark.net/cpu.php....60GHz&id=2334
- with 4 Cores / 8 Threads
- an average CPU mark of 10089
- and a Single Thread Rating of 2270


According to the CPU marks from above 1x AMD EPYC 7401P machine would perform roughly the same as 2x Intel Xeon E3-1271 v3 machines.

But doesn't it make more sense to compare the two by multiplying the Thread Count by the Single Thread Rating for the given workload?

AMD EPYC 7401P = 48*1488 = 71424
Intel Xeon E3-1271 v3 = 8* 2270 = 18160

In that case 1x AMD EPYC 7401P machine would perform roughly four times as fast as 1x Intel Xeon E3-1271 v3 machine.

To me this makes a whole lot more sense but is it correct?


Thanks,
humbug

[David (PassMark)]

No, by doing this you are assuming perfect scaling of a single thread to 48+ threads.
Very few algorithms get even close to perfect scaling. Thermal issues, system bottlenecks (memory bus & NUMA) and the algorithms themselves prevent perfect scaling.

[humbug]

I see, thanks David!