Apple A19 Pro's single-core benchmarks beat the Snapdragon 8 Elite and Ryzen 9 9950X

I would say most workloads are short and bursty and lightly threaded. If you want numbers for a specific task rather than basic compute there are benchmarks for that.
Video manipulation, batch photo editing, playing a game, installing an application are long running tasks on a phone.
So this test is modeling very short workloads for which you can get by with most CPUs available today, but it still looks shady because:
- it has a very small memory footprint;
- it does not look to be affected by tasks running at the same time
- for multicore tests it does not load all the cores or it is too short to register
 
Video manipulation, batch photo editing, playing a game, installing an application are long running tasks on a phone.
So this test is modeling very short workloads for which you can get by with most CPUs available today, but it still looks shady because:
- it has a very small memory footprint;
- it does not look to be affected by tasks running at the same time
- for multicore tests it does not load all the cores or it is too short to register
What’s your point?
 
You guessed right. Cinebench is total trash. Developers said they don't support AVX512 because it causes CPU to lower clock speeds so much it yields lower performance.

Like we definitely can see for example here https://www.phoronix.com/review/amd-zen5-avx-512-9950x
You’ve literally linked an article showing that the performance is 1.5x better using 512, clock speed isn’t everything even though your source noticed no noticeable change in frequency.

geometric-mean-of-all-test-results-result-composite-ar97r99a5c.svgz


There was an issue on CPUs that are older, talking 2019 and earlier but your argument is just incorrect even using your own source
 
You’ve literally linked an article showing that the performance is 1.5x better using 512, clock speed isn’t everything even though your source noticed no noticeable change in frequency.

geometric-mean-of-all-test-results-result-composite-ar97r99a5c.svgz


There was an issue on CPUs that are older, talking 2019 and earlier but your argument is just incorrect even using your own source
Yep, Cinebench developers say they don't use AVX-512 because it lowers clock speeds and therefore lower performance.

That article does not mention much about clock speeds as neither Ryzen 7000 or Ryzen 9000 series have separate "AVX clocks" (lower clock speeds when using AVX-512). As you can also see, using AVX-512 gives much more performance too.
You guessed right. Cinebench is total trash. Developers said they don't support AVX512 because it causes CPU to lower clock speeds so much it yields lower performance.
Like I said, Cinebench developers are morons.
 
What’s your point?

two points:
1. disproving your affirmation that most tasks are short and bursty by giving examples of longer running tasks in daily usage.
2. pointing that the test does not load the system properly and giving the reasons that led me to that conclusion

geekbench
- does not model real life tasks properly
- it is skewed towards mobile usage patterns
- its score is not a good metric to compare systems that are designed for different use cases

It is like comparing a tractor to a motorbike by measuring top speed and concluding that in general the motorbike is better and more economical.
 
Back