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

[AngryDwarf]

Superb work. Some algo's on my ageing CPU are almost half an ageing GPU.

Don't know whether it's possible to report the actual CPU frequency rather than the default CPU frequency in the stats. Some people are probably wondering how I get so much hash out of my CPU! (It's OC'd to 4.6 GHz).

You really should look at getting this onto github. It's quite simple to use. I'd suggest the following:
Create a github account and install git (sudo apt-get install git).
Fork the original source code using the web front end
Clone the fork into a new directory. (git clone <repo>)
Copy your code over the cloned fork code.
Submit the changes. (git commit -n "comment", I think)
Push the changes to github. (git push with a couple of parameters)

It's also easy to track changes made to the original and integrate them after that.

Expect a star from me!

[1714042397]

1714042397

[1714042397]

1714042397

[1714042397]

1714042397

[1714042397]

1714042397

[1714042397]

1714042397

[joblo]

Superb work. Some algo's on my ageing CPU are almost half an ageing GPU.

Don't know whether it's possible to report the actual CPU frequency rather than the default CPU frequency in the stats. Some people are probably wondering how I get so much hash out of my CPU! (It's OC'd to 4.6 GHz).

You really should look at getting this onto github. It's quite simple to use. I'd suggest the following:
Create a github account and install git (sudo apt-get install git).
Fork the original source code using the web front end
Clone the fork into a new directory. (git clone <repo>)
Copy your code over the cloned fork code.
Submit the changes. (git commit -n "comment", I think)
Push the changes to github. (git push with a couple of parameters)

It's also easy to track changes made to the original and integrate them after that.

Expect a star from me!

Thanks for the tips.

I agree with CPU freq. Simply displaying the default freq is of no use, it just happens to be part of the brand string.
I'm sure the actual freq can be found, the code may already exist.

About github, I presume your suggestion to fork the original cpuminer-multi is to maintain its history, this is a good thing.
I wanted to wait until the development slowed down so I wouldn't be trying to  learn github while trying to fix bugs in cpuminer.

Now may be a good time to dig into it. I need to play in a sandbox for a while before I comit to it.

[AngryDwarf]

Superb work. Some algo's on my ageing CPU are almost half an ageing GPU.

Don't know whether it's possible to report the actual CPU frequency rather than the default CPU frequency in the stats. Some people are probably wondering how I get so much hash out of my CPU! (It's OC'd to 4.6 GHz).

You really should look at getting this onto github. It's quite simple to use. I'd suggest the following:
Create a github account and install git (sudo apt-get install git).
Fork the original source code using the web front end
Clone the fork into a new directory. (git clone <repo>)
Copy your code over the cloned fork code.
Submit the changes. (git commit -n "comment", I think)
Push the changes to github. (git push with a couple of parameters)

It's also easy to track changes made to the original and integrate them after that.

Expect a star from me!

Thanks for the tips.

I agree with CPU freq. Simply displaying the default freq is of no use, it just happens to be part of the brand string.
I'm sure the actual freq can be found, the code may already exist.

About github, I presume your suggestion to fork the original cpuminer-multi is to maintain its history, this is a good thing.
I wanted to wait until the development slowed down so I wouldn't be trying to  learn github while trying to fix bugs in cpuminer.

Now may be a good time to dig into it. I need to play in a sandbox for a while before I comit to it.

github not to hard to learn, at its basics. Original developer might be terse and not open for discussion. Which is not good for a serial forker.

[joblo]

Superb work. Some algo's on my ageing CPU are almost half an ageing GPU.

Don't know whether it's possible to report the actual CPU frequency rather than the default CPU frequency in the stats. Some people are probably wondering how I get so much hash out of my CPU! (It's OC'd to 4.6 GHz).

Thanks for the tips.

I agree with CPU freq. Simply displaying the default freq is of no use, it just happens to be part of the brand string.
I'm sure the actual freq can be found, the code may already exist.

CPU freq is not so simple. It can be read diectly from the /sys/ or /proc/ file systems on Linux but I don't know of an API
for Windows.

Using /sys/devices/system/cpu/cpu0/cpufreq/cpuinfo_cur_freq would be preferred because it returns a simple numeric string.
But it is root access only.

/proc/cpuinfo is accessible and also includes the CPU freq but it needs to be filtered from two pages of output.

This will require more thought and more work.

[Epsylon3]

yep the only reliable way found was dmidecode | grep MHz

found no way on windows to get the freq and the temperature :/ it seems to require some low level drivers to access some protected asm

[joblo]

cpuminer-opt v3.3.7 released.

https://drive.google.com/file/d/0B0lVSGQYLJIZc0x3V2JPUXhPLWs/view?usp=sharing

- lyra2 AES optimized +7%
- small increase, ~1%, in most chained algos (x11, quark, lyra2v2, etc)
- lyra2v2 now reported as AES optimized
- added a decimal place to the accepted rate percentage if less than 100%

Windows binaries:

http://cryptomining-blog.com/wp-content/download/cpuminer-opt-3-3-7-windows.zip

[felixbrucker]

hi joblo,

it seems decred (and maybe other algos?) generate 100% rejected on nicehash with the following error:
"reject reason: Invalid extranonce2 size."

is this a bug?

br

[Epsylon3]

nicehash is using its own protocol. Even if it looks correct, i think its not better and might create problems on future blockchain votes.

My implementation was accepted/reviewed by the decred team, they have seen the vote problem (the last bit of the vote field should never be changed by the miner, and blocks will not be accepted by "proper" pools)
First i thought it was always 1 but it can be 0 too. its a small thing but could be important later.

Nicehash made a different one without contacting decred team

Else for multi gpus, my initial protocol seems better/easier to handle

[joblo]

hi joblo,

it seems decred (and maybe other algos?) generate 100% rejected on nicehash with the following error:
"reject reason: Invalid extranonce2 size."

is this a bug?

br

Thanks for reporting this. It appears to be specific to decred and nicehash. I will look into this and do a round of testing
on nicehash to confirm only decred is broken. I'm not very motivated to implement a workaround if it only applies to one
algo in one pool.

I should also point out that decred, and other blake algos, are the worst performing algos for CPU mining.

In summary this is a very low priority issue.

[joblo]

I can confirm that mining decred at nicehash is broken yet it works at zpool. I have also found some of the
other blake based algos are broken at nicehash.

I will have to do some more investigation.

[Epsylon3]

i heard skein2 too

[joblo]

i heard skein2 too

Yup, I broke that in 3.3.7. I knew that one of my optimizations would break it but forgot to follow up.
I had removed the automatic ctx init in the close function of all sph algos because it was unnecessary
in most cases. But it is needed in algos like skein2 that run the same algo multiple times. I simply forgot
to add an explicit init before the second skein round.

[felixbrucker]

thanks for looking into it, yes its a low prio issue, maybe its fixed in some future release

[felixbrucker]

i stumbled upon this as i was running some benchmark with nicehash for my cpu autoswitching miner control software (which uses cpuminer-opt)

[felixbrucker]

is there a way to statically build/link the miner (linux) ?

i wasn't able to get it running

[joblo]

is there a way to statically build/link the miner (linux) ?

i wasn't able to get it running

A native build is always prefereable to ensure you're taking advantage of all the optimizations supported
by your CPU.

Describe the problem you're having. Does it fail to compile, or fail to run after compiling? Post any error
messages.

[Godson_Mansa]

hey joblo, i run a windows 32 bit machine, considering the sw is for 64 bit i can't use it, is there a possible workaround ?
thanks

[yubsep]

Good.. I'm running CPU-Miner at i7 6700K 8M on windows 7 64x and didn't have any problem...

[joblo]

hey joblo, i run a windows 32 bit machine, considering the sw is for 64 bit i can't use it, is there a possible workaround ?
thanks

No and it's not worth it if there was.

[Fuzzbawls]

hey joblo, i run a windows 32 bit machine, considering the sw is for 64 bit i can't use it, is there a possible workaround ?
thanks

probably time to upgrade the computer/OS or replace it entirely. If the CPU is indeed 32-bit, then mining with it will be academic at best. Latest Intel 32-bit CPUs didn't handle thermal design very well, and overheating was quite common in the high clockspeed pentium 4's. Latest AMD 32-bit CPUs lack support for SSE2, which this miner also uses quite heavily.

If you have a 64-bit CPU, but are using a 32-bit OS...upgrade your OS.