cpuminer-opt-3.8.6
Fixed argon2 regression in v3.8.5.
Added x16s algo for Pigeoncoin.
Some code cleanup.
https://github.com/JayDDee/cpuminer-opt/releases/tag/v3.8.6
I plan to remove the AVX Windows binary in the next release.
Please report any concerns with supporting data.
Topic: [ANN]: cpuminer-opt v3.8.8.1, open source optimized multi-algo CPU miner - page 18. (Read 444067 times)
March 31, 2018, 11:57:06 AM
March 31, 2018, 11:56:12 AM
Hello, it seems that avx512 can make a big difference for argon2d: avx512 can reach very high hashrate according to https://docs.google.com/spreadsheets/d/1ZTbJWvvSUu-5ULL-FGCdCuDGwa4SJsj6lj4ns50lJYM/edit#gid=0
with a 7820x (6 cores) can surpass the amd 1950x.
For this processor : 36000 with avx512 versus 28500 with probably avx2.
With beastpool avx2, I can reach 38 000 with my xeon E5 2696V3 @2,7 ghz.
So, where did this magnificent AVX512 code come from?
https://github.com/BeastPool/cpuminer-argon2d/releases
the sources are provided, with some info in the file avxdefs.h.
What a farce!
I put the AVX512 code in avxdefs in early preparation for when I start converting the existing
AVX2 code to AVX512.
It's expemimental and ISN'T USED ANYWHERE.
So beastpool forks my code, assumes the AVX512 is active and claims it in his release.
That's either pretty stupid or deliberately deceptive.
Or they didn't release the sources for their avx512 code... Now I have some doubts.
The miner included in the wallet pretend to have avx 512 code too:
https://github.com/duality-solutions/Dynamic
March 31, 2018, 11:48:30 AM
Hello, it seems that avx512 can make a big difference for argon2d: avx512 can reach very high hashrate according to https://docs.google.com/spreadsheets/d/1ZTbJWvvSUu-5ULL-FGCdCuDGwa4SJsj6lj4ns50lJYM/edit#gid=0
with a 7820x (6 cores) can surpass the amd 1950x.
For this processor : 36000 with avx512 versus 28500 with probably avx2.
With beastpool avx2, I can reach 38 000 with my xeon E5 2696V3 @2,7 ghz.
So, where did this magnificent AVX512 code come from?
https://github.com/BeastPool/cpuminer-argon2d/releases
the sources are provided, with some info in the file avxdefs.h.
What a farce!
I put the AVX512 code in avxdefs in early preparation for when I start converting the existing
AVX2 code to AVX512.
It's expemimental and ISN'T USED ANYWHERE.
So beastpool forks my code, assumes the AVX512 is active and claims it in his release.
That's either pretty stupid or deliberately deceptive.
March 31, 2018, 11:13:01 AM
Hello, it seems that avx512 can make a big difference for argon2d: avx512 can reach very high hashrate according to https://docs.google.com/spreadsheets/d/1ZTbJWvvSUu-5ULL-FGCdCuDGwa4SJsj6lj4ns50lJYM/edit#gid=0
with a 7820x (6 cores) can surpass the amd 1950x.
For this processor : 36000 with avx512 versus 28500 with probably avx2.
With beastpool avx2, I can reach 38 000 with my xeon E5 2696V3 @2,7 ghz.
So, where did this magnificent AVX512 code come from?
https://github.com/BeastPool/cpuminer-argon2d/releases
the sources are provided, with some info in the file avxdefs.h.
March 31, 2018, 11:00:26 AM
Hello, it seems that avx512 can make a big difference for argon2d: avx512 can reach very high hashrate according to https://docs.google.com/spreadsheets/d/1ZTbJWvvSUu-5ULL-FGCdCuDGwa4SJsj6lj4ns50lJYM/edit#gid=0
with a 7820x (6 cores) can surpass the amd 1950x.
For this processor : 36000 with avx512 versus 28500 with probably avx2.
With beastpool avx2, I can reach 38 000 with my xeon E5 2696V3 @2,7 ghz.
So, where did this magnificent AVX512 code come from?
March 31, 2018, 10:55:55 AM
Argon2 SOLO no any accepted block with 3.8.5
No error, cpu usage 100%, but no accepted block.
same with 2x2680v1/Manjaro and 1700x/W10
3.8.4.1 accepting blocks on both config.
No error, cpu usage 100%, but no accepted block.
same with 2x2680v1/Manjaro and 1700x/W10
3.8.4.1 accepting blocks on both config.
I think I found the bug. Argon2 and Argon2d had many name conflicts so I renamed all the argon2
conflicting names, but missed one, so it was calling the wrong version of a function.
I haven't found a pool to test it but it should work in the next release.
March 31, 2018, 09:31:12 AM
Argon2 SOLO no any accepted block with 3.8.5
No error, cpu usage 100%, but no accepted block.
same with 2x2680v1/Manjaro and 1700x/W10
3.8.4.1 accepting blocks on both config.
No error, cpu usage 100%, but no accepted block.
same with 2x2680v1/Manjaro and 1700x/W10
3.8.4.1 accepting blocks on both config.
March 31, 2018, 08:36:31 AM
cpuminer-opt-3.8.5
Added argon2d-crds and argon2d-dyn algos.
sha256t 8 way AVX2 & 4 way SSE4.2 optimized.
CPUs with SSE4.2 get optimizations previously reserved for AVX.
https://github.com/JayDDee/cpuminer-opt/releases/tag/v3.8.5
Added argon2d-crds and argon2d-dyn algos.
sha256t 8 way AVX2 & 4 way SSE4.2 optimized.
CPUs with SSE4.2 get optimizations previously reserved for AVX.
https://github.com/JayDDee/cpuminer-opt/releases/tag/v3.8.5
Hello, it seems that avx512 can make a big difference for argon2d: avx512 can reach very high hashrate according to https://docs.google.com/spreadsheets/d/1ZTbJWvvSUu-5ULL-FGCdCuDGwa4SJsj6lj4ns50lJYM/edit#gid=0
with a 7820x (6 cores) can surpass the amd 1950x.
For this processor : 36000 with avx512 versus 28500 with probably avx2.
With beastpool avx2, I can reach 38 000 with my xeon E5 2696V3 @2,7 ghz.
March 30, 2018, 05:44:51 PM
March 30, 2018, 04:38:18 PM
cpuminer-opt-3.8.5
Added argon2d-crds and argon2d-dyn algos.
sha256t 8 way AVX2 & 4 way SSE4.2 optimized.
CPUs with SSE4.2 get optimizations previously reserved for AVX.
https://github.com/JayDDee/cpuminer-opt/releases/tag/v3.8.5
Added argon2d-crds and argon2d-dyn algos.
sha256t 8 way AVX2 & 4 way SSE4.2 optimized.
CPUs with SSE4.2 get optimizations previously reserved for AVX.
https://github.com/JayDDee/cpuminer-opt/releases/tag/v3.8.5
big move in cryptonight algo with Monero and other fork...
some move to V7, some not....
some be used antiAsic made by Monero Dev team, some(like Sumo Dev) go much deep with antiAsic and move to new algo cryptonight-heavy
can you make and add this algo to cpuminer-opt?
March 30, 2018, 07:31:52 AM
March 29, 2018, 01:20:59 PM
Hi, need an enlightenment here. 
Does CPUminer-Opt has built-in command line options to set CPU affinity?
Thanks before.
Does CPUminer-Opt has built-in command line options to set CPU affinity?
Thanks before.
start /affinity 55 cpuminer...... or --cpu-affinity 0x55
Got it. Thank you.
55 is 01010101 t 4
use calc for youre cpu https://yadi.sk/i/AhXE0h9D3Ttd79
March 29, 2018, 12:01:05 PM
Hi, need an enlightenment here.
Does CPUminer-Opt has built-in command line options to set CPU affinity?
Thanks before.
Does CPUminer-Opt has built-in command line options to set CPU affinity?
Thanks before.
start /affinity 55 cpuminer...... or --cpu-affinity 0x55
Got it. Thank you.
March 29, 2018, 11:55:09 AM
Hi, need an enlightenment here.
Does CPUminer-Opt has built-in command line options to set CPU affinity?
Thanks before.
Does CPUminer-Opt has built-in command line options to set CPU affinity?
Thanks before.
start /affinity 55 cpuminer...... or --cpu-affinity 0x55
March 29, 2018, 03:12:16 AM
Hi, need an enlightenment here.
Does CPUminer-Opt has built-in command line options to set CPU affinity?
Thanks before.
Does CPUminer-Opt has built-in command line options to set CPU affinity?
Thanks before.
March 28, 2018, 01:03:25 PM
argon2d algos seem somewhat slower than the reference on avx2 / windows
If by reference you mean the Beastpool fork, it was forked from cpuminer-opt and I just imported
all the argon2d code as-is. It should be identical.
How much slower is it? Are both algos similarly affected? Since you're using the binary my only guess
is a compile issue.
yes, the Beastpool fork
crds 26k vs 32k
dyn 8k fvs 10k
I can try compiling both from source to verify, but that will probably only happen on the weekend
Compiling beastpool requires a few extra steps. First you need to reverse the pthreadGC2 change for Linux
and chmod +x autogen.sh and build.sh.
I get virtually the same speed for both on crds: 29.8k opt vs 29.6k beast.
What compiler was used for the Beaspool bin?
March 28, 2018, 11:33:29 AM
argon2d algos seem somewhat slower than the reference on avx2 / windows
If by reference you mean the Beastpool fork, it was forked from cpuminer-opt and I just imported
all the argon2d code as-is. It should be identical.
How much slower is it? Are both algos similarly affected? Since you're using the binary my only guess
is a compile issue.
yes, the Beastpool fork
crds 26k vs 32k
dyn 8k fvs 10k
I can try compiling both from source to verify, but that will probably only happen on the weekend
Compiling beastpool requires a few extra steps. First you need to reverse the pthreadGC2 change for Linux
and chmod +x autogen.sh and build.sh.
I get virtually the same speed for both on crds: 29.8k opt vs 29.6k beast.
What compiler was used for the Beaspool bin?
March 28, 2018, 10:55:19 AM
argon2d algos seem somewhat slower than the reference on avx2 / windows
If by reference you mean the Beastpool fork, it was forked from cpuminer-opt and I just imported
all the argon2d code as-is. It should be identical.
How much slower is it? Are both algos similarly affected? Since you're using the binary my only guess
is a compile issue.
yes, the Beastpool fork
crds 26k vs 32k
dyn 8k fvs 10k
I can try compiling both from source to verify, but that will probably only happen on the weekend
March 28, 2018, 10:09:34 AM
argon2d algos seem somewhat slower than the reference on avx2 / windows
If by reference you mean the Beastpool fork, it was forked from cpuminer-opt and I just imported
all the argon2d code as-is. It should be identical.
How much slower is it? Are both algos similarly affected? Since you're using the binary my only guess
is a compile issue.
March 28, 2018, 03:59:12 AM
argon2d algos seem somewhat slower than the reference on avx2 / windows
yes, but more stable for me on 4 treads. Original gives me bsod...
Jump to: