Plot creation times on Apple M1

[AndyRPH]

Its come up in a few issues, so figured I'd make a discussion spot for plot creation times on Apple M1 CPUs.

The latest build is now giving me times around 17k seconds, so not great, actually, slower than the pechy-optimised plotter I was previously using. This is running it for 4 threads using tmp1 on a usb 3.1 NVME SSD and tmp2 on a TB3 NVME SSD. The MBPro has 16 GB ram and not much else going on during these times.

Two days ago, I was getting consistent plots of 3 hours or so, so it's slowed down a bit. Not sure if it was the blake3 commit or something else.

For the record, I even did one plot using my internal SSD and that too didn't get much faster than 3 hours for a plot. Plots using 2 threads are a bit slower than 4 threads but not by much.

Seems the M1 CPU machines are bottlenecked elsewhere than this plotter will help with.

[stingray072]

Its come up in a few issues, so figured I'd make a discussion spot for plot creation times on Apple M1 CPUs.

The latest build is now giving me times around 17k seconds, so not great, actually, slower than the pechy-optimised plotter I was previously using. This is running it for 4 threads using tmp1 on a usb 3.1 NVME SSD and tmp2 on a TB3 NVME SSD. The MBPro has 16 GB ram and not much else going on during these times.

Two days ago, I was getting consistent plots of 3 hours or so, so it's slowed down a bit. Not sure if it was the blake3 commit or something else.

For the record, I even did one plot using my internal SSD and that too didn't get much faster than 3 hours for a plot. Plots using 2 threads are a bit slower than 4 threads but not by much.

Seems the M1 CPU machines are bottlenecked elsewhere than this plotter will help with.

Try setting threads to 200 and see what happens. Seemed to give some good results in harnessing the M1 on prior plotters.

[AndyRPH]

It's a four core processor...(plus four "Efficency" cores). Why would you set the threads to 200?

[dmorrison91]

Before Max I also saw a person do 200 threads, and so I did and it sped up my plots incredibly and only used 30% of the CPU roughly. I don't know maybe arm process timer runs very different, also saw that you can go up to 2500 threads in the activity monitor so until you hit that limit... run multi-instances.

[stingray072]

Yup, something to do with the way M1 translates things. Don’t have to try if you don’t want to, just a suggestion. Would be curious what the time comes down to

[AndyRPH]

Okay, for fun, I ran it with it set to 16 threads. Slowed to a crawl. After 3 hours it hadn't even finished phase 1 and I killed it.

[dmorrison91]

Also how'd you compile, I'm getting an error message on the compile on my m1 mini.

[dmorrison91]

Yea I deleted everything and restarted the entire compile, but still getting this error now

/Users/derekmorrison/chia-plotter/src/chia_plot.cpp:16:10: fatal error: 'sodium.h' file not found #include <sodium.h> ^~~~~~~~~~ 1 error generated. make[2]: *** [CMakeFiles/chia_plot.dir/src/chia_plot.cpp.o] Error 1 make[1]: *** [CMakeFiles/chia_plot.dir/all] Error 2 make: *** [all] Error 2

[dmorrison91]

I've tried brew install libsodium-dev but get a no similarly named formula error

[leecarney]

On m1 I've found the libraries are stored in /opt/homebrew/Cellar/libsodium/1.0.18_1/include so you need to use these in your ln -s commands.
That being said I still have an ld -lsodium error (used on the last line of Cmakefile) that I just can't get past no matter what LDFLAGS options etc I use.

[dmorrison91]

Okay made the changes now I'm getting new errors:

[100%] Linking CXX executable chia_plot clang: warning: argument unused during compilation: '-fmax-errors=1' [-Wunused-command-line-argument] ld: library not found for -lsodium clang: error: linker command failed with exit code 1 (use -v to see invocation) make[2]: *** [chia_plot] Error 1 make[1]: *** [CMakeFiles/chia_plot.dir/all] Error 2 make: *** [all] Error 2

[moertel187]

have also these errors

[xxxxxxPETExxxxx]

Just as comparison I'm using Plotman with regular chia plotter on M1 MBP with 4 parallel and getting 18-19 plots per day on a 1TB Samsung 980 pro in TB3. So If you're getting 3 hours using MadMax I think you should switch to Plotman since my average time per plot is much lower than yours.

[xxxxxxPETExxxxx]

no problems with 8 threads each. If you look in Activity monitor your Mac has more than 1500 threads going at any given time. I'm using plotman with the official plotter. My single plot time is just over 3 hours. with 4 parallel it's 4.5 hours. Can't imagine what your bottleneck is for 8-9 hours or 20 hours. Must be really slow temp drive.

[AndyRPH]

Oh on the pechy optimized plotter I can do 3hr/plot with 4 threads, if I run two staggered I can get it to 4.5 hr/s per plot x2. Interesting on the threads, it did NOT work at all that way with max's, plotter, 16 threads dropped speed dramatically, but I'll try that in plotman and see what happens.

[xxxxxxPETExxxxx]

Post back your results please. I had never heard of this optimized plotter.

[AndyRPH]

I linked to it above, it's about 30% faster than the official plotter, but works differently than max's plotter I believe. We'll see tomorrow if giving plotman using that optimized plotter 16 threads makes a difference over 2/4 threads.

[xxxxxxPETExxxxx]

Thanks. Another program I use is "TG Pro" to monitor CPU temps and "auto boost" the fans. you can try it free for two weeks and after that it's only $12. Keeps my temps much lower than without which I assume helps to avoid CPU throttling.