Apple M5 Max crushes Ryzen AI Max+ 395 by over 25%, outpacing desktop CPUs from Intel and AMD in Geekbench 6

[Redaktion]

A leaked Geekbench 6 listing for the newly launched Apple M5 Max SoC reveals impressive performance gains in both single and multicore performance. The SoC handily outpaces every other laptop chip, while leaving the M3 Ultra in the dust as well.

https://www.notebookcheck.net/Apple-M5-Max-crushes-Ryzen-AI-Max-395-by-over-25-outpacing-desktop-CPUs-from-Intel-and-AMD-in-Geekbench-6.1243938.0.html

[davidm]

Talk about memory bandwidth. In depth.

[RobertJasiek]

Geekbench is known as Apple PR bench. Therefore, it is always Geekbench that leaks benchmarks before launches of Apple devices. Not to mention ignorance of dGPUs.

[Serhii]

Bad info.They began to compare their processors by fake key features. Halo series invented for GPU  abilities, not CPU. For multiCPU abilities there Treadripper from AMD and Xeon from Intel.

[Gta Raja]

Woohoo! Most powerful processor for web browsing.

[Mama]

So by these numbers, the 64 core Threadripper 9980X is 21,2 % faster than the 16 core 9950X.

Crap benchmark then.

[dada_dave]

No GB is not an Apple PR benchmark. The issue, at least for Threadrippers and other similarly high core count CPUs is that GB 6 MT tests are, by design, not a good test for them. For GB 6, Primate Labs made the conscious decision that only a few of the subtests should be embarrassingly parallel (e.g. code compilation and 3D rendering/ray tracing) while the rest would be work sharing algorithms that don't scale as well. The reason they did this was to mimic how most consumer software actually behaves and to give the average consumer a better sense of the "real" performance benefit they might see from a bigger CPU as opposed to being wowed by marketing misrepresenting how much benefit they'd get out that many cores. John Poole, head of Primate Labs, stated one of the reasons he directed the team to do this was because of how upset he was of companies using GB 5 to try to upsell the average customer into getting devices they not only didn't need, but might actually be worse for them.

If you are in the market for Threadrippers or the like, then you are ... not an average consumer and GB 6 MT, beyond the scores for certain subtests, is not meant to be representative of your workloads.  While Primate Labs could add a third category or a different test for "workstation MT", they seem mostly content with making a consumer-focused benchmark.