3DMark CPU Profile - new CPU benchmark

[porina]

https://benchmarks.ul.com/news/new-cpu-benchmarks-for-gamers-and-overclockers

 

This one apparently came out a couple of days ago but I only just heard of it. It is a CPU benchmark that reports a series of scores for 1, 2, 4, 8, 16, and maximum number of threads running. This gets around the eternal argument caused by focusing on single or multi-thread based scores when it is not best relevant. You have both single and multi-thread scores from here. We can then focus arguments on what this benchmark is actually doing.

 

I'm away from home so can only run it on my laptop with a Zen 3 5800H mobile CPU, with 8 cores 16 threads.

 

Link to web results: http://www.3dmark.com/cpu/16620

 

Well, I get some numbers. How those numbers compare will have to wait until I get home next week.

 

As is, the bench results may be slightly skewed by thermal and/or power limits. It starts off at higher thread counts and works down. At the start, the CPU may be cooler and thus have less thermal effects early on. Likewise time based power budgets will be consumed early on and may give a boost in that area.

1 to 2 threads: 1.93x

2 to 4 threads: 1.92x

4 to 8 threads: 1.69x

8 to 16 threads: 1.17x

 

Looking at score scaling with increasing threads. We seem to get near ideal scaling from 1 to 4 threads, with a little overhead somewhere. 4 to 8 threads is a smaller jump. It will take more work to try and determine if this is due to software scaling overheads or is a hardware resource limitation. For example, this might be determined by running an 8 core CPU at a fixed lower clock. This will reduce the loading on non-execution hardware resources. If this scales better than expected, it is hardware. If it doesn't change in scaling, it is software limiting. 8 to 16 threads on an 8 core system is indicative of how much benefit SMT gives. 17% is a pretty unremarkable value, assuming it is not hardware limiting.

 

Open questions: 

What is the benchmark effective peak IPC on Intel vs AMD CPUs?

Is it strongly affected by cache sizes, speeds, memory bandwidth/latency?

How co-dependant are the threads to each other? For example, 1 task using 8 threads is different from 8 independent tasks using 1 thread each.

[porina]

I've had a bit more time to run this on more systems. Following is a summary of my observations based on core-clock normalised results:

 

It seems quite stable, scores typically within 1% run to run.

It doesn't seem to scale at all with ram speed (tested 2400 to 4000 dual channel on 10600k).

It does seem to scale well with CPU clock - when running single thread there can be a small difference from halving the clock, which increases score/clock by possibly 2%. No difference observed running multi-threaded.

Rocket Lake scores about 20% more per clock than Comet Lake where not limited elsewhere

Core/thread scaling isn't linear with scaling dropping off as they increase. It isn't Cinebench in that respect. Running threads = cores is one point where there is a drop in relative perf scaling. Above that you're into HT/SMT so a drop is expected, but still below ideal scaling.

I'm seeing potentially ~17% contribution from HT on Comet Lake/Skylake-X with the caution of the non-ideal scaling described above. I don't feel able to give values for other architectures without running more (core reduced) tests, which I don't plan to do.

It probably doesn't use AVX-512. Skylake-X didn't perform any better than Comet Lake, and the Rocket Lake uplift is probably from general IPC increases with the new architecture.

The mobile Zen 3 I have is about 5% behind Rocket Lake. Desktop version should do better from having more cache.

 

Overall I think this is a more meaningful benchmark to the average gamer than Cinebench, but it is also more complicated to compare scores since it isn't a single number. As a weakness, it doesn't seem to show any ram bandwidth scaling which we know many games do have. It is also questionable how representative the thread scaling is compared to any particular game, and it will vary from game to game anyway.