[LOCKED] cpuminer-opt v3.12.3, open source optimized multi-algo CPU miner

[joblo]

cpuminer-opt maximum support only 64 threads CPU? My CPU threads more than 64, how to use it to be able to fully play

There is nothing in cpuminer-opt to limit threads, maybe resource limitations or a system issue.

[double-mint]

Are we likely to see gains with 4-way in lyra2z in the future, or the algo isn't affected by this optimization?

[bl0atw4re]

Hello joblo can you add AES to yescryptr16 algo .. it only mines with avx and sse2 on my sandy bridge machine

[joblo]

Are we likely to see gains with 4-way in lyra2z in the future, or the algo isn't affected by this optimization?

The simple answer is no, but it's not so simple.

The lyra2 code is already AVX2 optimimized. Converting it to 4way would not improve it and would be
double the work because the existing optimizations would have to be undone first. This applies to all
lyra2 based algos.

A small part of lyra2z could be improved with 4way but it wouldn't have a big impact on the overall
algo performance.

Finally the lyra2z algo is on life support. It's only used by Zcoin and only until their new algo is ready.

[joblo]

Hello joblo can you add AES to yescryptr16 algo .. it only mines with avx and sse2 on my sandy bridge machine

That will be difficult because yescrypt doesn't use any AES.

[joblo]

I stumbled upon this today, well done with a lot of detail. I'll add the link
to the compile instructions in RELEASE_NOTES.

https://lxadm.com/Static_compilation_of_cpuminer

[Sam67]

I want to say a huge thank you to the developer for this minor and for his timely update - my 6950x produces excellent results on it - I often fly into solo at the start of new coins!

[Sam67]

I have a question - why in Monero cryptonight on 6950x on cpuminer-opt-3.7.3 the speed is lower than on xmrig-2.4.2 -? In xmrig with 10 streams, the speed is higher than cpuminer-opt at 20. Is it possible to somehow configure the config file to give more speed at 6950x than in xmrig-2.4.2?

[4ward]

I have a question - why in Monero cryptonight on 6950x on cpuminer-opt-3.7.3 the speed is lower than on xmrig-2.4.2 -? In xmrig with 10 streams, the speed is higher than cpuminer-opt at 20. Is it possible to somehow configure the config file to give more speed at 6950x than in xmrig-2.4.2?

From what I know, cryptonight requires 2mb of cache per thread

6950x has 25mb, which brings us to 12 threads
Try benchmarking with 12 threads and see the performance (use -t 12)

[joblo]

I have a question - why in Monero cryptonight on 6950x on cpuminer-opt-3.7.3 the speed is lower than on xmrig-2.4.2 -? In xmrig with 10 streams, the speed is higher than cpuminer-opt at 20. Is it possible to somehow configure the config file to give more speed at 6950x than in xmrig-2.4.2?

From what I know, cryptonight requires 2mb of cache per thread

6950x has 25mb, which brings us to 12 threads
Try benchmarking with 12 threads and see the performance (use -t 12)

Agreed. Test both with 12 threads and post the results.

[Sam67]

use -t 12 - xmrig-2.4.2 - 752 -  cpuminer-opt-3.7.3 - 550

[joblo]

use -t 12 - xmrig-2.4.2 - 752 -  cpuminer-opt-3.7.3 - 550

Thanks. I believe xmrig uses large pages, a OS feature that can improve memory performance when
accessing large buffers. It has some side effects such as increased memory fragmentation and it
requires admin user on Windows. It's also rarely enabled by default in the OS and needs to be
manually enabled.

I haven't implimented it yet in cpuminer-opt but it's on my list.

[felixbrucker]

I haven't implimented it yet in cpuminer-opt but it's on my list.

looking forward to that

[joblo]

I haven't implimented it yet in cpuminer-opt but it's on my list.

looking forward to that

Large pages has already been done for cryptonight. I'm doing something that hasn't been done yet.
Large pages for cpuminer-opt will have to wait, though it could benefit a couple of memory hard algos.

[felixbrucker]

Large pages has already been done for cryptonight. I'm doing something that hasn't been done yet.
Large pages for cpuminer-opt will have to wait, though it could benefit a couple of memory hard algos.

looking forward to that as well

[nizzuu]

Large pages has already been done for cryptonight. I'm doing something that hasn't been done yet.
Large pages for cpuminer-opt will have to wait, though it could benefit a couple of memory hard algos.

Hi, this may be useful as well: https://software.intel.com/sites/landingpage/IntrinsicsGuide/#othertechs=BMI2&expand=3773

AVX-512F section as well, but I have no supporting cpu :-( So I can't test any benefints as compared to AVX-2. They should be, but...

[joblo]

Hi, this may be useful as well: https://software.intel.com/sites/landingpage/IntrinsicsGuide/#othertechs=BMI2&expand=3773

AVX-512F section as well, but I have no supporting cpu :-( So I can't test any benefints as compared to AVX-2. They should be, but...

LOL. That page is permanently open in my browser.

I was wondering when someone would mention AVX-512.
I'm already dreaming about 8-way. It should be easier than going from 1-way to 4-way. That makes my next
CPU a difficult choice. Do I go with Ryzen and just SHA or wait for Cannonlake with SHA and AVX-512?

[felixbrucker]

why not both

[joblo]

why not both

Timing. Cannonlake is delayed until end of 2018 now, still possible for more delays. A Ryzen purchase
could be done in spring when the next Ubuntu LTS is released.

Both does have some advantages. I could get the Ryzen earlier before next LTS and then Cannonlake
delays don't matter.

[nizzuu]

Do I go with Ryzen and just SHA or wait for Cannonlake with SHA and AVX-512?

Well, Ryzens have 2x128-bit wide AVX units instead of 256, don't forget about it I think this is not a good implementation to target to.

As for AVX-512, the only adequate choice for now is i7-7800X, it's not so expensive but has 140W TDP (and liquit ship inside instead of solder).