[ANN]: cpuminer-opt v3.8.8.1, open source optimized multi-algo CPU miner - page 60.

joblo

legendary

Activity: 1470

Merit: 1114

Quote from: 4ward on October 25, 2017, 08:25:07 AM

Have you tries cross-compiling in Linux to windows? That could solve the old mingw version issues, let you use a newer gcc and perhaps allow to add SHA support
I have tried cross-compiling, but the build configurations you have, need some changes and I'm yet to have success figuring it out, due to lack of experience. Although cross-compiling openssl was pretty simple

edit:
after some playing around, i have managed to cross-compile, but, unfortunately, executions fails with "workio thread dead, exiting" message
any suggestions on where to dig?

How do you cross-compile? Isn't that what mingw does? How did you make it portable?

That last question may be why the workio thread is dying, possibly a pthread issue. To run on Windows it needs to link to
pthreadGC2 instead of pthread. I have run in to this problem trying to compile on cygwin. I have discovered that I had
set up a "cygwin" compile environment instead of a "mingw" compile environment. The mingw toolchain includes pthreadGC2.

ZenFr

legendary

Activity: 1260

Merit: 1046

Quote from: 4ward on October 26, 2017, 10:58:37 AM

Quote from: ZenFr on October 26, 2017, 09:02:48 AM

Unfornutaly, this miner don't work with equihash algorithm (exept an error from my side ?) !
Do you know a good CPU miner for equihash ?

yes, it doesnt support equihash
try nheqminer: https://github.com/nicehash/nheqminer/

Ok, thank you for the reply.
NHEQMiner have no update since 10 months...
But I will try this last version (My last version was the first, a long time ago :-).).

4ward

member

Activity: 473

Merit: 18

Quote from: ZenFr on October 26, 2017, 09:02:48 AM

Unfornutaly, this miner don't work with equihash algorithm (exept an error from my side ?) !
Do you know a good CPU miner for equihash ?

yes, it doesnt support equihash
try nheqminer: https://github.com/nicehash/nheqminer/

ZenFr

legendary

Activity: 1260

Merit: 1046

Joblo CPUminer is the best CPU miner :-).
Unfornutaly, this miner don't work with equihash algorithm (exept an error from my side ?) !
Do you know a good CPU miner for equihash ?

4ward

member

Activity: 473

Merit: 18

Quote from: joblo on October 23, 2017, 05:38:35 PM

Quote from: nizzuu on October 23, 2017, 04:43:03 PM

Quote from: joblo on October 21, 2017, 03:47:32 PM

I haven't tried it yet so I don't know how performance compares.

I can tell. With software AES, xmrig gives no performance improvement as compared to cpuminer-opt. With hardware AES, it gives ~15-20% boost on Kaby Lake pentiums, and around 40-45% on i5/i7 from same gen (Kaby Lake), as compared to cpuminer-opt. Thread (-t) number is the same.

Partly it's because of msvc compiler (your binaries are build with gcc), around 3-4% are there. I've personally compiled xmrig with three compilers - latest gcc, msvc 2017 and Intel compiler (trial parallel studio xe 2017 update 4) with different options. Intel and gcc give equal performance while msvc is slightly faster. So, for Windows and cryptonote part of cpuminer-opt it's probably no reason to purchase (or try) Intel compiler.

Wanted to compile cpuminer-opt using these three and test for performance (got different CPU's - G4560/G4600/i5-4570/G1840/G620/i3-4000M/E5-2609v3/E5-2630v4) , but the project structure is a bit complex for me to rework it for direct VS2017 usage. Will give it a try later if joblo won't release instruction (or rework the project)

The Windows binaries I provide are definitely slower than compiled on Linux. Both use gcc 4.8 so I do't think it's the compiler. My guess is
the libraries in my mingw environment are old. I've tried a couple of newer versions of mingw (mostly for SHA) but so far have not been successful
getting cpuminer-opt to compile.

I tried getting MSVC to work early on but gave up. At this point the project files would have to rebuilt from scratch to have any hope of
getting it to work.

Have you tries cross-compiling in Linux to windows? That could solve the old mingw version issues, let you use a newer gcc and perhaps allow to add SHA support
I have tried cross-compiling, but the build configurations you have, need some changes and I'm yet to have success figuring it out, due to lack of experience. Although cross-compiling openssl was pretty simple

edit:
after some playing around, i have managed to cross-compile, but, unfortunately, executions fails with "workio thread dead, exiting" message
any suggestions on where to dig?

nizzuu

full member

Activity: 187

Merit: 100

Cryptocurrency enthusiast

Quote from: joblo on October 23, 2017, 05:38:35 PM

Both use gcc 4.8 so I do't think it's the compiler.

I tried getting MSVC to work early on but gave up. At this point the project files would have to rebuilt from scratch to have any hope of
getting it to work.

Yep, the compiler side has a little affect. My main machine is i5-7600 non-K with 32Gb DDR4-3000MHz XMP @ 2 channels, so the compiler difference was ~230H/s vs ~239H/s. It's 3-4%, that's all. But the diff 129H/s with cpuminer-opt vs 239H/s xmrig - that's where real diff lives. Same result with cpuminer-multi and other forks, while xmrstack gives 225H/s with provided stock binaries (I hate the config way there huh).

Will start with component upgrades, thanks for pointing that out.

felixbrucker

hero member

Activity: 700

Merit: 500

Quote from: nizzuu on October 23, 2017, 04:43:03 PM

and around 40-45% on i5/i7 from same gen (Kaby Lake)

that is impressive, thanks for sharing

joblo

legendary

Activity: 1470

Merit: 1114

Quote from: nizzuu on October 23, 2017, 04:43:03 PM

Quote from: joblo on October 21, 2017, 03:47:32 PM

I haven't tried it yet so I don't know how performance compares.

I can tell. With software AES, xmrig gives no performance improvement as compared to cpuminer-opt. With hardware AES, it gives ~15-20% boost on Kaby Lake pentiums, and around 40-45% on i5/i7 from same gen (Kaby Lake), as compared to cpuminer-opt. Thread (-t) number is the same.

Partly it's because of msvc compiler (your binaries are build with gcc), around 3-4% are there. I've personally compiled xmrig with three compilers - latest gcc, msvc 2017 and Intel compiler (trial parallel studio xe 2017 update 4) with different options. Intel and gcc give equal performance while msvc is slightly faster. So, for Windows and cryptonote part of cpuminer-opt it's probably no reason to purchase (or try) Intel compiler.

Wanted to compile cpuminer-opt using these three and test for performance (got different CPU's - G4560/G4600/i5-4570/G1840/G620/i3-4000M/E5-2609v3/E5-2630v4) , but the project structure is a bit complex for me to rework it for direct VS2017 usage. Will give it a try later if joblo won't release instruction (or rework the project)

The Windows binaries I provide are definitely slower than compiled on Linux. Both use gcc 4.8 so I do't think it's the compiler. My guess is
the libraries in my mingw environment are old. I've tried a couple of newer versions of mingw (mostly for SHA) but so far have not been successful
getting cpuminer-opt to compile.

I tried getting MSVC to work early on but gave up. At this point the project files would have to rebuilt from scratch to have any hope of
getting it to work.

nizzuu

full member

Activity: 187

Merit: 100

Cryptocurrency enthusiast

Quote from: joblo on October 21, 2017, 03:47:32 PM

I haven't tried it yet so I don't know how performance compares.

I can tell. With software AES, xmrig gives no performance improvement as compared to cpuminer-opt. With hardware AES, it gives ~15-20% boost on Kaby Lake pentiums, and around 40-45% on i5/i7 from same gen (Kaby Lake), as compared to cpuminer-opt. Thread (-t) number is the same.

Partly it's because of msvc compiler (your binaries are build with gcc), around 3-4% are there. I've personally compiled xmrig with three compilers - latest gcc, msvc 2017 and Intel compiler (trial parallel studio xe 2017 update 4) with different options. Intel and gcc give equal performance while msvc is slightly faster. So, for Windows and cryptonote part of cpuminer-opt it's probably no reason to purchase (or try) Intel compiler.

Wanted to compile cpuminer-opt using these three and test for performance (got different CPU's - G4560/G4600/i5-4570/G1840/G620/i3-4000M/E5-2609v3/E5-2630v4) , but the project structure is a bit complex for me to rework it for direct VS2017 usage. Will give it a try later if joblo won't release instruction (or rework the project)

joblo

legendary

Activity: 1470

Merit: 1114

Quote from: felixbrucker on October 21, 2017, 02:14:52 PM

Maybe i missed it but i just stumbled upon this https://github.com/xmrig/xmrig

as it's based off cpuminer-multi, maybe there are some optimizations which can be used in cpuminer-opt as well, if there are any

It uses double buffering to hash two nonces with one thread. This isn't parallel processing because the double buffering is
done serially. This could theoretically reduce CPU stalls while waiting for data from memory.

I haven't tried it yet so I don't know how performance compares.

felixbrucker

hero member

Activity: 700

Merit: 500

Maybe i missed it but i just stumbled upon this https://github.com/xmrig/xmrig

as it's based off cpuminer-multi, maybe there are some optimizations which can be used in cpuminer-opt as well, if there are any

ecth

newbie

Activity: 2

Merit: 0

Nice, finally managed running it on my machine. Works best with Ubuntu on Windows. Compiled v3.7.0 for my Sandy Bridge-E 3930K (-march=corei7-avx).
On all cores (6 + HT = ~12), with some OC to 160W makes about 3.7-3.8 GHz on all cores/threads. Every thread has about 3.2-3.6 H/s. All together are at 40-42 H/s

On Windows I had problems running more than 3 threads. After upgrading to 1709 (Windows chose me. Update was shown instantly after release, lol Cheesy

) Windows Defender killed the process and deleted the precompiled binary. On Ubuntu (from the Windows App Store) it runs perfectly smooth

joblo

legendary

Activity: 1470

Merit: 1114

Quote from: ybelevi on October 20, 2017, 03:41:18 PM

hello
i've problem to install on debian.please help me

/usr/bin/ld: warning: libssl.so.1.0.2, needed by /usr/lib/gcc/x86_64-linux-gnu/6/../../../x86_64-linux-gnu/libcurl.so, may conflict with libssl.so.1.1
/usr/bin/ld: warning: libcrypto.so.1.0.2, needed by /usr/lib/gcc/x86_64-linux-gnu/6/../../../x86_64-linux-gnu/libcurl.so, may conflict with libcrypto.so.1.1
make[2]: Leaving directory '/root/cpuminer-opt'
make[1]: Leaving directory '/root/cpuminer-opt'

This looks like only a warning, if the executable was built give it a try.

I presume you installed a newer version of openssl-dev for SHA support. It was initially added to 1.1 but appears to have been
ported back to some 1.0 versions.

If 1.1 doesn't work try with the latest 1.0.2. It has been reported to have SHA on Ubuntu 16.04 so it should also work on Debian.
You can confirm SHA support a number of ways. If you're CPU is a Ryzen the default compile arch should include SHA,
or you can add -msha to CFLAGS. You can confirm when the miner is started that SHA is listed in the SW features.

ybelevi

member

Activity: 132

Merit: 10

hello
i've problem to install on debian.please help me

/usr/bin/ld: warning: libssl.so.1.0.2, needed by /usr/lib/gcc/x86_64-linux-gnu/6/../../../x86_64-linux-gnu/libcurl.so, may conflict with libssl.so.1.1
/usr/bin/ld: warning: libcrypto.so.1.0.2, needed by /usr/lib/gcc/x86_64-linux-gnu/6/../../../x86_64-linux-gnu/libcurl.so, may conflict with libcrypto.so.1.1
make[2]: Leaving directory '/root/cpuminer-opt'
make[1]: Leaving directory '/root/cpuminer-opt'

joblo

legendary

Activity: 1470

Merit: 1114

Quote from: 4ward on October 18, 2017, 04:06:43 AM

I noticed that the blake2b exists in the sources, but disabled in algo-gate-api
any reason for that?

I never got it to work. I should not have mentioned it in the change log.

ZenFr

legendary

Activity: 1260

Merit: 1046

Happy o see the new cpuminer-opt version 3.7.0.
I will test it as soon as possible.
Thank you dev, good job :-)!

4ward

member

Activity: 473

Merit: 18

Quote from: joblo on October 17, 2017, 11:14:07 AM

cpuminer-opt-3.7.0

Fixed x14 misalignment bug.
Fixed decred stake version bug.
Getwork fixes for algos that use big endian data encoding: m7m, zr5, neoscrypt,
blake2b, decred.

git: https://github.com/JayDDee/cpuminer-opt

tarball: https://drive.google.com/file/d/0B0lVSGQYLJIZb3FncjV4R0pMaU0/view?usp=sharing

Windows binaries: https://drive.google.com/file/d/0B0lVSGQYLJIZMC1KWFF0V3RQYU0/view?usp=sharing

I noticed that the blake2b exists in the sources, but disabled in algo-gate-api
any reason for that?

joblo

legendary

Activity: 1470

Merit: 1114

cpuminer-opt-3.7.0

Fixed x14 misalignment bug.
Fixed decred stake version bug.
Getwork fixes for algos that use big endian data encoding: m7m, zr5, neoscrypt, decred.

git: https://github.com/JayDDee/cpuminer-opt

tarball: https://drive.google.com/file/d/0B0lVSGQYLJIZb3FncjV4R0pMaU0/view?usp=sharing

Windows binaries: https://drive.google.com/file/d/0B0lVSGQYLJIZMC1KWFF0V3RQYU0/view?usp=sharing

nizzuu

full member

Activity: 187

Merit: 100

Cryptocurrency enthusiast

Famous encoder team is working on AVX-512 support for now. Seems this one could be added to some algos (e.g. lyra2) with ease, what a pity I have no target CPU, but what i see in the code, is to add it in memset and register optimisation in the gate. GCC usage is here (AVX512F).

joblo

legendary

Activity: 1470

Merit: 1114

Quote from: malafaya on October 16, 2017, 01:36:54 PM

Decred mining yields all rejects:

I tried old v3.4.12 with same result. Possibly a change in algo? I recall it used to work on zpool.

Same result obtained with a more recent CPU:

Looks like there was a change in the algo, looking into it.

Topic: [ANN]: cpuminer-opt v3.8.8.1, open source optimized multi-algo CPU miner - page 60. (Read 444040 times)