Hacker News Reader

Comment by srcmap | Hacker News Reader

Two additional benchmarks I like to see from Anandtech:

1) some VM base tests for these kind of CPU - 32, 64, 128 VM all running some kind of web/db/redis benchmarks inside.

2) Some compilation testing - time the clean build of AOSP, BSD, some very complex linux app - use max jobs setting for parallel compile and time how long it take to finish the jobs. ( measure the over CPU usages at the same time. )

View on HN

Phoronix had it compiling the Linux kernel in 32 seconds, compared to 37.5 for the 7980XE.

Link https://www.phoronix.com/scan.php?page=article&item=amd-linu...

Wow. I wonder how fast it could do a full bootstrapped build of gcc...

Even more generally I would like to see some benchmarks being done of the VM extensions or other processor features. I don't know any benchmarks that helps you understand if the AES native instructions are any good or improved between models. There's a lot of chatter about the memory architecture but knowing that features like nested page table work well would be really useful for me personally.

For these kind of test I think Servethehome.com and https://www.phoronix.com does a better job.

There are Chromium compile tests. https://www.anandtech.com/show/13124/the-amd-threadripper-29...

They do a compilation of Chrome in the "Office Tests" section. The 32Core takes about an hour per compile, but the 16Core is actually a bit faster (my guess would be that linking takes longer on the 32 core machine as it's very memory intensive).

FWIW, I believe the version they use relies on a version of MSVC where the linker is predominently single-threaded, and linking the final binary ends up as a significant proportion of the time.

Does anyone do multithreaded LTO yet?

GCC and clang do. As a matter of fact, MSVC does too, to some extent (CGTHREADS option, iirc).

Yes, the issue is the LTCG build, which by default uses 4 threads and with Chromium the compiler hits some pathological issues with such a huge program and it ends up using mostly 1 core and swapping lots of memory.

Reply on news.ycombinator.com