Controversial Geekbench 4 scores

[zxcvbad]

So Geekbench 4 is now released, I've been following Reddit posts closely while it was in beta state, seeing all those claims SD820 was outperformed by SD650, Kirin, Exynos and A57 being slightly lower:
SoC: Single-core / multi-core

• 820: 1694/3458
• 652: 1413/4012
• 8890: 1874/5342
• 950: 1709/5377
• A9: 2513/4235
  

Now I see it myself, mainly looking at single-core of the hardware that I have with me:
SD801: 1012
Exynos 7420: 1265
SD820: 1674

It looks absolutely weird to me, especially after all the profiling that I've personally made on all the SoCs mentioned, SD820 got miles ahead in sustained performance.

Any thoughts on this? Benches isn't something I completely trust but Geekbench was one of the better ones, at least pretty useful when having lots of hardware around that needs to be profiled.
GB4 vs GB3 are so crazy different

[Grey1213]

Well, the SD 652 has more cores, so it would do better for multicore. And for single core it does use a72 cores so it shouldn't be that far behind.

[zxcvbad]

Atm I'm feeling like they tend to mix all the base spec with every new gb release. It doesn't feel right when you get an old SD801 28nm scored 1012 vs A57 14nm 1265. I never used GFXBench CPU tests properly, so I guess I have to give it a try

[zxcvbad]

Just decided to profile Geekbench 4 renderscript in action.
SD801: first half of the test (exact 50%) it's switching between cores 0-4 with a maximum load for single-core bench measurement, then it does multi-core by loading all cores simultaneously, you can see CTemp gets high so it starts to reduce freq relatively. Interestingly on S820 cores works in clusters, so instead of switching between cores 0-4 for single-core bench it's pushing cores 3+4 (bigger cluster) simultaneously, then multi-core test starts (exact 50%) loading both clusters up to the max, just as on S801. Not 100% sure but it looks like it's Qualcomm's config to load cluster+cluster since I just tried Vellamo and it's a similar behavior. Was pretty cool to watch S801 gets up to 70c while S820 never reached above 53c

[zxcvbad]

I'm done with this. Likely they've changed entire spec with GB4 release, results are no longer relevant, especially if comparing them to any other CPU based renderscript benchmarks eg Vellamo, Antutu, Compubench etc.

Single-core
S801: 1012
Exynos 7420: 1407
S820: 1674

P.S. Noticed for single-core Geekbench 4 is always pushing all the available bigger cores on any SoC except S801, so with Exynos 7420 it's 4 A57s vs 2 Kryo 0x205 on S820