Don't understand the benchmark output.

TimG · 2022-01-11, 17:43

I would like to know the optimal cores and workers windows to use, so I ran the benchmark options.

https://i.imgur.com/SoxvZvq.png

These are the options I am running the benchmark with. The problem is, it's been running for a little while now, and the output looks the same to me (as if it is retesting the same options).

See the attached log file as of this moment for what I mean.

kruoli · 2022-01-11, 19:25

What kind of work are you going to do?

With your settings, it will benchmark different parameter settings for each FFT. Each FFT will run a few to some more of them, only after which that the next FFT size will be benchmarked. This can indeed take very long.

Mark Rose · 2022-01-11, 23:45

You can probably start with an FFT of 5632 if you're interested in working exponents at the wavefront.

I wouldn't benchmark hyperthreading. Hyperthreads may give one or two percent increase, but will make your system sluggish. Hyperthreads do seem to benefit on cloud instances though.

For your CPU, either 1 or 8 workers is likely to be best. I'd switch 1-8 with 1,8.

Benchmarking all FFT implementations is useful for Prime95 to find the fastest tweaks. If you're just curious, I'd uncheck the box.

In the text file, look at the Throughput iterations/second to see what gives you the most. The important bit is should you run 1 worker with 8 threads, or 8 workers with 1 thread. Most likely 1 worker with 8 threads will be faster. Working on a single exponent at a time also produces results sooner, which is more fun.

2022-01-11, 19:25	#2
kruoli "Oliver" Sep 2017 Porta Westfalica, DE 2³·3·43 Posts	What kind of work are you going to do? With your settings, it will benchmark different parameter settings for each FFT. Each FFT will run a few to some more of them, only after which that the next FFT size will be benchmarked. This can indeed take very long. Last fiddled with by kruoli on 2022-01-11 at 19:27 Reason: Did not read correctly.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Can someone help me understand this...	petrw1	Puzzles	9	2020-10-28 05:42
Help me understand this result?	LOBES	Information & Answers	4	2020-07-23 14:37
Benchmark output Prime95	lukerichards	Software	7	2019-05-19 19:46
Finally I understand EVERYTHING!	Flatlander	Lounge	2	2010-10-10 14:30
I just don't understand this	kuratkull	Sierpinski/Riesel Base 5	2	2007-03-12 21:02

2022-01-11, 23:45	#3
Mark Rose "/X\(‘-‘)/X\" Jan 2013 2,953 Posts	You can probably start with an FFT of 5632 if you're interested in working exponents at the wavefront. I wouldn't benchmark hyperthreading. Hyperthreads may give one or two percent increase, but will make your system sluggish. Hyperthreads do seem to benefit on cloud instances though. For your CPU, either 1 or 8 workers is likely to be best. I'd switch 1-8 with 1,8. Benchmarking all FFT implementations is useful for Prime95 to find the fastest tweaks. If you're just curious, I'd uncheck the box. In the text file, look at the Throughput iterations/second to see what gives you the most. The important bit is should you run 1 worker with 8 threads, or 8 workers with 1 thread. Most likely 1 worker with 8 threads will be faster. Working on a single exponent at a time also produces results sooner, which is more fun.