Hi everyone, I am currently investigating a scaling bottleneck in one of my codes which seems a bit elusive to me. In my investigation to get a MWE, I have tried to produce a somewhat representative example, which should model an ideal scenario, where I expect a perfect speedup with the number of …

[image] Salmon: (I sincerely hope that they also do not do any shenanigans with changing the CPU clockspeed Sorry if I missed it earlier in the thread, but have you actually measured the clock speeds across cores throughout your benchmarks? It should be easy to confirm. I thought most modern…

[image] Salmon: Currently I could not find the number of NUMA Domains for this, but I’ll have a look and write it here. Since you already seem to have ThreadPinning.jl installed: using ThreadPinning nnuma() nsockets() ncores() ncputhreads() # etc... :wink: [image] Sukera: Well, there m…

Yes exactly - the scaling is the same :slight_smile: Since the scaling is usually desired to achieve better performance though, I just thought it was interesting to see what actually optimizing the serial code can still achieve (the naive version still can’t beat the serially-optimized @fastmath ver…

[image] carstenbauer: Since you already seem to have ThreadPinning.jl installed: using ThreadPinning nnuma() nsockets() ncores() ncputhreads() # etc... Thanks for the hint. I didnt have access to the machine before but here it is. A few more infos since it might be important: I am running …

Here is the script I used for benchmarking threads: using ProgressMeter bench_threads(prog::String, nthread::Int = Threads.nthreads()) = bench_threads(prog, 1:nthread) function bench_threads(prog::String, threads::AbstractVector{Int}) script = prog * "\n" * """ using BenchmarkTools t = @belap…

[image] Salmon: Actually, seeing this I am slightly puzzled that it tells me ncputhreads() = 96. I did start Julia with nthreads() = 48 though. All of the ThreadPinning commands give you static information about the system itself and nothing about how your started Julia. ncputhreads specifica…

[image] Salmon: Actually, seeing this I am slightly puzzled that it tells me ncputhreads() = 96. I did start Julia with nthreads() = 48 though. Can it be that the existence of virtual threads that I never use somehow causes issues? I dont really see why though, but it seems noteworthy. Yes. I…

While it is commented out in his OP, he already has pinthreads(:cores) there. Given that Threads.nthreads() <= ncores() this ensures that Julia threads are pinned to different cores.

Hey again, I have had a look into many of your suggestions and would like to follow up. First, instead of using exp, I now perform some simpler floating point calculations instead. It might be that there was indeed a bit of thread scaling left on the table due to the call to exp. The scaling is no…

There are just so many things that can affect multithreaded performance scaling. There’s a heavy interplay between your code and your hardware — and some effects are purely hardware-dependent. Just off the top of my head: Modern processors might up- or down-clock based upon the workloads. For ex…

Right, this is exactly the reason why I want to start from a scenario which is as ideal as it can be. Maybe I should be explicit of the things I think am covering by now with my example. The “ideal” program, I have come up with has a number of task which is a mutliple of the number of cores, al…

How to achieve perfect scaling with Threads (Julia 1.7.1)

General Usage Performance

carstenbauer January 13, 2023, 11:07am 34

It isn’t required anymore for Julia >= 1.9 (see e.g. Using the Intel VTune Profiler with julia - #26 by vchuravy). I’ve updated the README appropriately.

Topic		Replies	Views
Help wanted: benchmarking multi-threaded CPU performance Offtopic hardware	20	1164	May 13, 2024
Help me understand multi-threaded scaling for matrix multiplication Performance question	22	872	April 16, 2024
Huge performance fluctuations in parallel benchmark: insights? Performance parallel , multithreading , benchmarktools	52	2931	December 1, 2021
AMD Rome vs Intel Xeon shows bad scaling with threads for AMD Performance multithreading , loopvectorization	20	2002	March 13, 2022
Same multi-threaded code, scaling observed only on some machines Performance	2	102	August 14, 2024

How to achieve perfect scaling with Threads (Julia 1.7.1)

Related topics