js-framework-benchmark icon indicating copy to clipboard operation
js-framework-benchmark copied to clipboard

Published 20 hours ago • verified publisher icon krausest

Intel testing?

Open tomByrer opened this issue 3 years ago • 2 comments
trafficstars

While Apple silicon is popular with devs, I doubt even 5% of the over all computer market & nearly 0% of servers uses that CPU. Could it be more accurate for the web world at large to keep running on Intel chips (even if a bit older)?

Big fan of your work; helped me find SolidJS & other devs who care about performance.

tomByrer avatar Aug 04 '22 23:08 tomByrer

I fired up my old razer blade and ran the benchmark with the build.zip: https://krausest.github.io/js-framework-benchmark/2022/table_chrome_104_windows.html

It wasn't quite a joy doing so, it takes much more time on windows (about 30 hours). It's really amazing that a laptop without a fan can beat that hot blowing gaming (though a bit old) laptop by such a large margin.

Other interesting findings:

  • Variance is higher on windows
  • Overall I find it surprising that the relative ranking is quite similar between OSX/arm64 and Windows/x64. At first glance I haven't seen a framework with differing performance characteristics.

Please let me know if you have further insights. Currently I'd say there's no reason to run windows regularly due to the similarity of the results (and absolute numbers don't matter with CPU slowdowns).

krausest avatar Aug 06 '22 18:08 krausest

it takes much more time on windows (about 30 hours).

Woh, thanks for your dedication!

relative ranking is quite similar between OSX/arm64 and Windows/x64

For the most part, seems true. react-zustand-v17.0.1 + 3.7.2 had a large shift: OSX 1.75 vs Win 1.63 & did fall in rankings. I'm sure there a few more we we really did a comparison. I'll have to build a diff to really compare.

I used to hand-optimize DSP in ASM back when AMD64 was king. When Intel Core2Duo came out, I was almost out of a job; x86 opcodes ran in execution order before, but Core CPUs were able to take some instructions out-of-order, so some of my poetization's became moot.

With M1/2 customized orders backloged & downtrend in economy, I don't think Apple Silcom will take over much of their small desktop marketshare soon. Though perhaps M1 chips reflect the A* mobile chipset better than x86 tests. & servers are still 95%+ x86.

I don't see such a drastic change in x96 vs M1/2, but there is enough for some IMHO.

tomByrer avatar Aug 10 '22 18:08 tomByrer