Primate Labs Launches Geekbench 6 Benchmarking Suite

walkin said:
Interesting. Can someone please explain how the new benchmarks can still serve as a valid comparison with older hardware that has only run the previous version(s), given that the tests have changed with the new version? Or can they? Doesn’t changing the test defeat the very purpose of a benchmark? Do we need to run the new benchmarks on old hardware before it can be compared with new hardware?
Click to expand...
Geekbench 6 benchmark scores should only be compared to other Geekbench 6 scores.

You’ll need to re-run on older hardware to get valid data.

Comparing different versions scores makes no sense. They run different tests.
 
MacBook Pro M2 Pro 12 CPU cores / 19 GPU cores:
CPU Single-core score: 2642
CPU Multi-core score: 14250
GPU Metal score: 81105
GPU OpenCL score: 49675
 
For reference. I ran it on an AMD 5900X with 64 GB RAM and Windows 11 and got 2060 single core and 12686 multi-core. That's not great compared to what are getting posted so far on Macs.

Edit: Second run. Single core: 2173, 12608 multi-core. RTX 3070 OpenCL: 131742; Vulkan: 130069
 
Last edited:
My results:

Mac Mini M2 Pro (base model, 512 GB, 16 GB): 2671, 12139. Open CL: 43617
iPhone 14 Pro Max (256 GB): 2527, 6469
iPad Pro 12.9 M2 (128 GB): 2500, 9409
Macbook Air M1 (256 GB): 2346, 8518
iPhone 13 Pro (256 GB): 2316, 5677
Ryzen 5900X, 32 GB 3200 mhz: 2142, 12135
 
Last edited:
M1 24" iMac 8/8/16GB/1T MacOS 13.2.1
Geekbench 6:
Single-Core 2369
Multi-Core 8562
OpenCL 20110
Metal 32765

M1 Max 16" MBP 10/32/32GB/1T MacOS 13.2.1
Geekbench 6:
Single-Core 2426
Multi-Core 12542
OpenCL 69489
Metal 115823

M1 Max 16" MBP 10/32/32GB/1T MacOS 12.1
Geekbench 5
Single-Core 1778
Multi-Core 12756
OpenCL 61429
Metal 68453
 
Last edited:
IMO the value of this benchmark for creative users has become even more dubious than it already was. I have a Mac Pro 2013 12 core that, in most tasks (especially 4K video editing & colour grading), is comparable with my Mac mini M1, and in some tasks outperforms it by a decent margin. The exception is for something like hardware accelerated H.265 encoding where the Mini does exceptionally well. But this benchmark skews the results drastically in favour of the M1. In fact, even my iPhone 12 mini gets multi-core results that are almost as good as the 12 core Mac Pro, which strikes me as absurd. The Metal scores have also skewed in favour of the new hardware. I think Geekbench 4 reflected reality for purposes like mine much better. Maybe not for machine learning or H.265 encoding, but for run of the mill creative work.

My scores for reference (single-core/multi-core/Metal):

iPhone 12 mini: 1999/4203/16149
MBP 15" 2017 3.1GHz w Radeon Pro 560: 1200/3996/20002
Mac Pro 12 core 2.7 w dual FirePro D700: 638/4682/29152 (single GPU)
Mac mini M1: 2320/8296/32313
 
Last edited:
walkin said:
Interesting. Can someone please explain how the new benchmarks can still serve as a valid comparison with older hardware that has only run the previous version(s), given that the tests have changed with the new version? Or can they? Doesn’t changing the test defeat the very purpose of a benchmark? Do we need to run the new benchmarks on old hardware before it can be compared with new hardware?
Click to expand...


These new benchmarks begin a new generation of testing… do not compare to old versions - not the same


Just continue using geekbench 5 if you want to compare new hardware with old.
 
melliflu said:
MacBook Pro (14-inch, 2021)Apple M1 Pro 3220 MHz (10 cores) 16Go OS Monterey
Geekbench 5:
Single-Core 1770
Multi-Core 12561
OpenCL 38159
Geekbench 6:
Single-Core 2408
Multi-Core 12401
OpenCL 39863
Metal 56238

"Geekbench 6 scores are calibrated against a baseline score of 2,500 (which is the score of an Intel Core i7-12700)."
Click to expand...
It’s rather interesting that Geekbench 6 shows the M1 series with a single-core score in the 2400 range vs 1700 range on Geekbench 5, which can simply be attributed to a difference in how the measurement scale is set up, but that the multi-core scores are about the same. Previously mutli-core scores scaled fairly linearly with the number of cores present, but this is no doubt a factor of Geekbench 6 being “designed to measure how cores share workloads in true-to-life workload examples, and there are a number of new tests that measure how people use devices.” This makes sense, since more cores doesn’t equal more power unless your workflow is designed to utilize as many cores as possible.
 
dgdosen said:
Interesting lack of consistency:

Ran it on an iPad Mini 6
SC: 2096
MC: 4849
Click to expand...

I'd check if both devices are charged to similar %. Lower charge usually means lower scores:

BWNddkirtqdY2bLT.medium


dgdosen said:
Ran it on an iPad Mini 6
SC: 2096
MC: 4849

That's saying an iPad Mini 6 is better at SC than a 5900X?
Click to expand...

That tracks. The iPad Mini 6's A15 uses a 3.25 GHz Avalanche core (2021), while the 5900x uses a 4.8 GHz Zen3 core (2020). Avalanche has always been faster than Zen3 in single-core performance. So even an iPhone 13 Mini should be faster in SC than a 5900X.

From AnandTech many moons ago, the A15 is just a hair behind the 5950X in integer and noticeably ahead in floating point:

A15: 7.28 int, 10.15 fp
5950X: 7.29 int, 9.79 fp

SPECint-energy.png


SPECfp-energy.png
 
Realityck said:
M1 24" iMac 8/8/16GB/1T MacOS 13.2.1
Geekbench 6:
Single-Core 2369
Multi-Core 8562
OpenCL 20110
Metal 32765

M1 Max 16" MBP 10/32/32GB/1T MacOS 13.2.1
Geekbench 6:
Single-Core 2426
Multi-Core 12542
OpenCL 69489
Metal 115823

M1 Max 16" MBP 10/32/32GB/1T MacOS 12.1
Geekbench 5
Single-Core 1778
Multi-Core 12756
OpenCL 61429
Metal 68453
Click to expand...


Yes, this is only a single test, but it seems they got the GPU scaling issue fixed with ASi. Thanks for posting.
 
You must log in or register to reply here.
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.
Back
Top