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

[Rameron]

Hello.

There are some problems with mining with cryptonight algorithm. In 3.3.6 version of miner on many CPU for mining was using AES and AVX instructions if available. But now, even when I launched cpuminer-aes-avx for Ivy/Sandy Bridge or cpuminer-aes-avx2 for Haswell and above, their are always using AES and SSE2 instructions. On i7 3770, on i7 4770, even on Xeon E5-1650 v3. There wasn't any cases when for mining was using AVX instruction in 3.6.0 version.

Is it normal? Or combination of AES and SSE2 instruction are more efficient then AES and AVX?

[onedeveloper]

Dear Dev,

I tried to use cpuminer to mine xmr with my computer.

It shows:

[2017-03-23 13:23:08] Starting Stratum on stratum+tcp://xmr.crypto-pool.fr:3333
[2017-03-23 13:23:08] 24 miner threads started, using 'cryptonight' algorithm.
[2017-03-23 13:23:09] API bind to port 4048 failed - trying again in 20sec
[2017-03-23 13:23:11] Stratum difficulty set to 18000
[2017-03-23 13:23:15] CPU #6: 66 H, 18.39 H/s


Is it ok for API failed?

In my experience, is no problem if you don't want or need to control the miner remotely through the provided API interface. In your case, it seems you are mining with another instance of this same miner, or the original version, and you need to specify a new port for this instance. Use:

Code:

--api-bind <other_port_number>

in your argument list and the problem will be solved.

[joblo]

Hello.

There are some problems with mining with cryptonight algorithm. In 3.3.6 version of miner on many CPU for mining was using AES and AVX instructions if available. But now, even when I launched cpuminer-aes-avx for Ivy/Sandy Bridge or cpuminer-aes-avx2 for Haswell and above, their are always using AES and SSE2 instructions. On i7 3770, on i7 4770, even on Xeon E5-1650 v3. There wasn't any cases when for mining was using AVX instruction in 3.6.0 version.

Is it normal? Or combination of AES and SSE2 instruction are more efficient then AES and AVX?

The algo features are hard coded by me as a guide to users. I changed the definition for cryptonight because there is no AVX
or AVX2 code used by cryptonight. A Haswell won't perform significantly faster that a Sandybridge mining cryptonight but is
faster on algos that do have AVX2 code.

[joblo]

Dear Dev,

I tried to use cpuminer to mine xmr with my computer.

It shows:

[2017-03-23 13:23:08] Starting Stratum on stratum+tcp://xmr.crypto-pool.fr:3333
[2017-03-23 13:23:08] 24 miner threads started, using 'cryptonight' algorithm.
[2017-03-23 13:23:09] API bind to port 4048 failed - trying again in 20sec
[2017-03-23 13:23:11] Stratum difficulty set to 18000
[2017-03-23 13:23:15] CPU #6: 66 H, 18.39 H/s


Is it ok for API failed?

In my experience, is no problem if you don't want or need to control the miner remotely through the provided API interface. In your case, it seems you are mining with another instance of this same miner, or the original version, and you need to specify a new port for this instance. Use:

Code:

--api-bind <other_port_number>

in your argument list and the problem will be solved.

Correct.

[Rameron]

A Haswell won't perform significantly faster that a Sandybridge mining cryptonight but is
faster on algos that do have AVX2 code.

So, is there no difference between Haswell and Sandybridge microarches for cryptonight algo and no sense to use different versions of miner for it (but only in case, when cpu not support aes)?

[joblo]

So, is there no difference between Haswell and Sandybridge microarches for cryptonight algo and no sense to use different versions of miner for it (but only in case, when cpu not support aes)?

That's not what I said.

[Rameron]

That's not what I said.

I know, but what am I wrong about?

[joblo]

That's not what I said.

I know, but what am I wrong about?

You're trolling. My answer was very clear. Your follow up question was just plain stupid.

[Ursul0]

That's not what I said.

I know, but what am I wrong about?

You're trolling. My answer was very clear. Your follow up question was just plain stupid.

he wasn't being stupid. probably just trying to do something other than to run one of the five executables in accordance with the platform they're built for, which may seem in case of Cryptonone as a good idea due to certain lack of clarity...

[joblo]

That's not what I said.

I know, but what am I wrong about?

You're trolling. My answer was very clear. Your follow up question was just plain stupid.

he wasn't being stupid. probably just trying to do something other than to run one of the five executables in accordance with the platform they're built for, which may seem in case of Cryptonone as a good idea due to certain lack of clarity...

Lack of clarity, that's funny. What's not clear about running the exe that matches the CPU's arch?
And it is stupid to suggest it makes no sense to do so. Too many people don't think before asking
questions.

[Ursul0]

lol. well.. I can see clearly that you're optimizing algorithms in asm nowdays:) no ambiguity there.
I guess it's something like that: If I start EXE on cryptonote for AVX2 on a CPU that supports it I still see that it's supposedly runs with SSE2 optimization, as it's a 'feature' of the algo implementation. So the question remains: should the best results be achieved with the miner that is "optimized for SSE2" even though it runs on AVX2 capable platform?

[felixbrucker]

lol. well.. I can see clearly that you're optimizing algorithms in asm nowdays:) no ambiguity there.
I guess it's something like that: If I start EXE on cryptonote for AVX2 on a CPU that supports it I still see that it's supposedly runs with SSE2 optimization, as it's a 'feature' of the algo implementation. So the question remains: should the best results be achieved with the miner that is "optimized for SSE2" even though it runs on AVX2 capable platform?

afaik the cpuminer bins distributed are compiled for this arch (features), so using a AVX2 bin with a AVX2 cpu on a SSE2 only algo yields the same results as using a SSE2 only bin as the miner uses the same code paths (but the AVX2 bin can use the AVX2 paths as well, but for this algo there arent any, so its the same)

[Ursul0]

lol. well.. I can see clearly that you're optimizing algorithms in asm nowdays:) no ambiguity there.
I guess it's something like that: If I start EXE on cryptonote for AVX2 on a CPU that supports it I still see that it's supposedly runs with SSE2 optimization, as it's a 'feature' of the algo implementation. So the question remains: should the best results be achieved with the miner that is "optimized for SSE2" even though it runs on AVX2 capable platform?

afaik the cpuminer bins distributed are compiled for this arch (features), so using a AVX2 bin with a AVX2 cpu on a SSE2 only algo yields the same results as using a SSE2 only bin as the miner uses the same code paths (but the AVX2 bin can use the AVX2 paths as well, but for this algo there arent any, so its the same)

Thx:) I'd normally guess that this should be the case, however I have no idea how optimizations are implemented/executables built.

...app can "use appropriate code path" only if its there and for that you need to know that there's 0 avx instructions written by the coder or generated otherwise, during the AVX exe build for this particular algo. 
EDIT: otherwise using AVX exe on SSE2 only capable CPU is a no go, even in cases where the algo itself has no AVX optimizations... in any case it all boils to the straight forward: use EXE build for your platform and hope for the best results

[joblo]

lol. well.. I can see clearly that you're optimizing algorithms in asm nowdays:) no ambiguity there.
I guess it's something like that: If I start EXE on cryptonote for AVX2 on a CPU that supports it I still see that it's supposedly runs with SSE2 optimization, as it's a 'feature' of the algo implementation. So the question remains: should the best results be achieved with the miner that is "optimized for SSE2" even though it runs on AVX2 capable platform?

afaik the cpuminer bins distributed are compiled for this arch (features), so using a AVX2 bin with a AVX2 cpu on a SSE2 only algo yields the same results as using a SSE2 only bin as the miner uses the same code paths (but the AVX2 bin can use the AVX2 paths as well, but for this algo there arent any, so its the same)

Thx:) I'd normally guess that this should be the case, however I have no idea how optimizations are implemented/executables built.

...app can "use appropriate code path" only if its there and for that you need to know that there's 0 avx instructions written by the coder or generated otherwise, during the AVX exe build for this particular algo. 
EDIT: otherwise using AVX exe on SSE2 only capable CPU is a no go, even in cases where the algo itself has no AVX optimizations... in any case it all boils to the straight forward: use EXE build for your platform and hope for the best results

Please stop this speculartion. If you are really concerned about these details do your own experiments. Don't invent issues
that don't exist, it only confuses others.

[giagge]

Someone who has ryzen 1700, can kindly tell me the best parameters for overclocking? I'd like to stay within the 65 watts.

[joblo]

Looking for Ryzen owners to test SHA acceleration.

Ryzen CPUs have HW accelerated SHA, something Intel's mainstream CPUs won't have until Cannonlake.

Requirements:
  Ryzen CPU
  Linux
  Compiler version with support for sha
  Openssl development package with support for sha
  Cpuminer-opt source code with support for sha

I can provide the last one. GCC is supposed to have support for -msha starting in v 5, and Openssl added support in 2014
so both should be available in Ubuntu 16.04.

The following algos use sha256 and should benefit from HW acceleration: sha256t, lbry, skein, myr-groestl, m7m.

[LukkasG]

Wave, person,

Would anyone have a tutorial on how to use download, compile and use this miner in UBuntu?

[nizzuu]

Seems that lyra2rev2 is broken in 3.6.0 avx2 build. When I launch the miner, cli doesn't show cpu threads at all, then miner quits after ~15 seconds. It's OK on 3.5.12 avx2 with same settings:

3.6.0:
cpuminer-aes-avx2 -a lyra2rev2 -t 4 -o stratum+tcp://hub.miningpoolhub.com:20507 -u qwerty -p qwerty

CPU: Intel(R) Core(TM) i5-4570 CPU @ 3.20GHz
CPU features: SSE2 AES AVX AVX2
SW built on Mar 15 2017 with GCC 4.8.3
SW features: SSE2 AES AVX AVX2
Algo features: SSE2 AES AVX AVX2
Start mining with SSE2 AES AVX AVX2

[2017-03-30 12:23:07] Starting Stratum on stratum+tcp://hub.miningpoolhub.com:20
507
[2017-03-30 12:23:07] 4 miner threads started, using 'lyra2rev2' algorithm.
[2017-03-30 12:23:07] Stratum difficulty set to 4

Then, miner quits after a short time. No crashes, just quit.

3.5.12:
cpuminer-aes-avx2 -a lyra2rev2 -t 4 -o stratum+tcp://hub.miningpoolhub.com:20507 -u qwerty -p qwerty

[2017-03-30 12:25:49] Starting Stratum on stratum+tcp://hub.miningpoolhub.com:20
507
[2017-03-30 12:25:49] 4 miner threads started, using 'lyra2rev2' algorithm.
[2017-03-30 12:25:50] Stratum difficulty set to 4
[2017-03-30 12:25:53] Stratum difficulty set to 1.79259
[2017-03-30 12:26:03] CPU #3: 2097.15 kH, 221.35 kH/s
[2017-03-30 12:26:03] CPU #0: 2097.15 kH, 219.95 kH/s
[2017-03-30 12:26:03] CPU #2: 2097.15 kH, 218.08 kH/s
[2017-03-30 12:26:04] CPU #1: 2097.15 kH, 197.22 kH/s
[2017-03-30 12:26:09] Accepted 1/1 (100%), 7509.29 kH, 851.53 kH/s

It's OK with 3.6.0 avx and other versions thought (slower, of course). Hope that info helps.

[joblo]

Seems that lyra2rev2 is broken in 3.6.0 avx2 build. When I launch the miner, cli doesn't show cpu threads at all, then miner quits after ~15 seconds. It's OK on 3.5.12 avx2 with same settings:

3.6.0:
cpuminer-aes-avx2 -a lyra2rev2 -t 4 -o stratum+tcp://hub.miningpoolhub.com:20507 -u qwerty -p qwerty

CPU: Intel(R) Core(TM) i5-4570 CPU @ 3.20GHz
CPU features: SSE2 AES AVX AVX2
SW built on Mar 15 2017 with GCC 4.8.3
SW features: SSE2 AES AVX AVX2
Algo features: SSE2 AES AVX AVX2
Start mining with SSE2 AES AVX AVX2

[2017-03-30 12:23:07] Starting Stratum on stratum+tcp://hub.miningpoolhub.com:20
507
[2017-03-30 12:23:07] 4 miner threads started, using 'lyra2rev2' algorithm.
[2017-03-30 12:23:07] Stratum difficulty set to 4

Then, miner quits after a short time. No crashes, just quit.

3.5.12:
cpuminer-aes-avx2 -a lyra2rev2 -t 4 -o stratum+tcp://hub.miningpoolhub.com:20507 -u qwerty -p qwerty

[2017-03-30 12:25:49] Starting Stratum on stratum+tcp://hub.miningpoolhub.com:20
507
[2017-03-30 12:25:49] 4 miner threads started, using 'lyra2rev2' algorithm.
[2017-03-30 12:25:50] Stratum difficulty set to 4
[2017-03-30 12:25:53] Stratum difficulty set to 1.79259
[2017-03-30 12:26:03] CPU #3: 2097.15 kH, 221.35 kH/s
[2017-03-30 12:26:03] CPU #0: 2097.15 kH, 219.95 kH/s
[2017-03-30 12:26:03] CPU #2: 2097.15 kH, 218.08 kH/s
[2017-03-30 12:26:04] CPU #1: 2097.15 kH, 197.22 kH/s
[2017-03-30 12:26:09] Accepted 1/1 (100%), 7509.29 kH, 851.53 kH/s

It's OK with 3.6.0 avx and other versions thought (slower, of course). Hope that info helps.

You're right, crashes on Windows, but works on Linux. I need to investigate.

Edit:

Lyra2REv2 AVX2 works on Windows with 3.5.13 and it might be a little faster than 3.5.12.

There are no changes to the Lyra2 code between 3.5.13 and 3.6.0 so I don't have any clues.
It's going to take a while to figure this out.

[integrale]

Wave, person,

Would anyone have a tutorial on how to use download, compile and use this miner in UBuntu?

hi LukkasG
i could try to help you out, but need some info about your OS and hardware  pm me if you need help

cheers Integrale