search

TheRainDoodle / Phenom-II-Benchmark

Small set of low level benchmarks for testing hardware speed against a Phenom II 810 Quad Core
GNU General Public License v3.0
91 stars 20 forks source link

Ryzen 3600 test #1

Open MauroMombelli opened 4 years ago

MauroMombelli commented 4 years ago

hi, as attachment the result for a ryzen 3600, stock clock, both single and multicore. Did not recompile your binary (seems make clean does not work) but should not be an issue since those are hardcrafted assembly. Would be mayne interesting to compare with an compiler's optimized code.. 3600single.txt 3600multi.txt

Inrixia commented 4 years ago

Saw this and figured I would add mine. 3950x Results: 3950xMulti.txt Didn't bother with single-threaded.

TediusTimmy commented 4 years ago

Why not? Ryzen 9 3900 (not X: it's the slower version). Ryzen9_3900_notX.txt It's interesting to see that it doesn't really look like your benchmark benefits from SMT: unless I'm reading the data wrong, it looks as though I am only getting the performance of boost of 12 cores and not 24 threads.

Nuadh commented 4 years ago

Jumping on the statistics with a 3700x

3700x.txt

wickwire commented 4 years ago

3900X here - using Rampagy's fork in order to run all tests (both single and multi) and save to file

3900xSingleMulti.txt

EDIT: hmmm I think the final score on test 7 multi threaded is wrong....... I will try it again as soon as I can

EDIT2: used TheRainDoodle's original benchmark - included binary for linux... 3900X@4.3GHz 1.275V

3900x@4.3GHz_single_multi_results.txt

Rampagy commented 4 years ago

@wickwire Thanks for testing out my fork! Test 7 is a little curious, but based on the code it does have have a 1:1 scaling from instructions to Phenom II's. The scalings are unchanged from TheRainDoodle's original benchmark.

Anyways here are my results: 3970x_linux.txt 3970x_windows.txt

TediusTimmy commented 4 years ago

@Rampagy If you check the sources, the 1:1 is only in the Linux source, compare Windows, and I assume it's an omission.

What I think is most interesting about this series of benchmarks: ten years on, you have a 2x improvement based on generation of the processor, and that's it. This benchmark doesn't seem to lend itself to SMT, so the multithreaded improvement is linear with cores and not threads. Are CPUs getting still getting better? Yes. But, not nearly as much better as I would bet the Phenom II compares to the original Athlon, a similar 10 year gap.