Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

I wonder how reliable geekbench tests are. Afaik it's the most common benchmark run on apple devices, so apple has a great interest in making sure their newest chips perform great on the test.

I wouldn't be surprised to hear that the geekbench developers are heavily supported by apple's own performance engineers and that testing might not be as objective or indicative of real world perf as one would hope.



Wouldn't it be more surprising to you Apple has been selling 4 generations of the M series to great acclaim on the performance but it all turned out to be a smoke and mirrors show because the hardware is optimized for one particular benchmark they didn't even reference in their comparisons?

From the data side: the M4 hasn't made it to the charts yet but the M3 already holds the 4th place in PassMark's single thread chart as well https://www.cpubenchmark.net/singleThread.html as well as tops Cinebench 2024 https://www.cpu-monkey.com/en/cpu_benchmark-cinebench_2024_s...

The only areas the M series "lags" is in the high end workstation/server segment where they don't really have a 96+ core option or in spaces where you pop in beefy high end GPUs. Everything else the M4 tends to lead in right now.


Okay, the linked benchmarks proved me wrong. I trust cinebench numbers a whole lot more than a geekbench score.

My bad.

And just to be clear: I didn't speculate that Apple tune's its chip to Geekbench, I speculated that geekbench was overly optimized towards apple's latest chip.


> (...) but it all turned out to be a smoke and mirrors show because the hardware is optimized for one particular benchmark they didn't even reference in their comparisons?

I know iOS developers who recently upgraded their MacBooks and they claim they now feel more sluggish. I wouldn't be surprised if it was due to RAM constraints instead of CPU though.

So, take those artificial bencmarks with a grain of salt. They are so optimized that they are optimized out of the real world.


Always look the benches of the actual workload types you'll be using. Some of the above, e.g. cinebench, are actually production rendering programs. Doesn't mean dick if you aren't going to use it that way.




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: