search

johguse / profanity

Vanity address generator for Ethereum
834 stars 324 forks source link

Apple M1 Performance #36

Open k06a opened 3 years ago

k06a commented 3 years ago

First of all, performance is pretty cool and MacBook Air M1 not started throttling and heating even after few minutes of bruteforcing private keys at speed 46 MH/s.

The only my concern is about 8 cores. It seems OpenCL discovered 8 CPU cores and not discovered another 8 GPU cores:

fluke9 commented 3 years ago

??? GPU != CPU

k06a commented 3 years ago

@fluke9 I see 16 cores on processor and only 8 are utilised by OpenCL

du5 commented 3 years ago

OpenCL has been deprecated, https://developer.apple.com/metal/

zlace0x commented 3 years ago

Could you sharing instructions to run this on M1 mac?

k06a commented 3 years ago

@zlace0x AFAIR make command compiled it successfully.

zlace0x commented 3 years ago

Thanks, re-downloaded a fresh copy and building under rosetta works.

obelisk commented 2 years ago

Running natively on an M1 Max 32-core GPU results in sustained 150MH

14" MBP, 64 GB Ram

k06a commented 2 years ago

@obelisk confirming same numbers for M1 Max 32-core 64GB (14")

heathdutton commented 2 years ago

I suspect the benchmark is somehow misleading in the case of the M1.

Even though it scores less than a tenth of my PC, the timing of the results my M1 mini finds are very close. I can also leave profanity running in the background on the Mac and forget it's running, while the PC would heat up the room... I'm not sure why the mac seems to match so well. Perhaps a difference in compilers. Getting this to compile in windows is a headache and I've likely messed something up.

I looked at the benchmarking code, but couldn't figure out a reason that it could be off... so maybe it's just me.

Either way, I've been just using the M1 instead, because it handles this very well. I suggest -I 20000 with the M1.