An (even more) optimized version of cpuminer (pooler's cpuminer, CPU-only)

[muqali]

Looking to get started on this with W7 x64 and an i7 3930K, I tried running the minerd.exe and get an json_rpc_call failed message. This is with or without a cfg.json file present in the directory. I ran through the first 6 pages or so of this post and didn't see any full Windows setup. If someone could give me a step by step please I would be happy as a pig in shit.

edit - got the miner setup. Using 8 threads at stock speeds I'm getting about 67KH/s. What does the --s flag do?

[Pontius]

[...] This time I've also built a couple static binaries for Linux (x86 and x86-64), for those people who really don't want to bother compiling the miner themselves.

Static binaries? Uh, nice!
Due to the libcurl deps I'm failing badly to a build static bin myself (on RHEL5). What flags did you set?

[muqali]

[...] This time I've also built a couple static binaries for Linux (x86 and x86-64), for those people who really don't want to bother compiling the miner themselves.

Static binaries? Uh, nice!
Due to the libcurl deps I'm failing badly to a build static bin myself (on RHEL5). What flags did you set?

Don't you need to set LDFLAGS to compile statically on Linux? It's been a while, so I could be mistaken.

[pooler]

• A --benchmark option is now available to benchmark a system without the need of connecting to a work provider.

can you tell more about that option?

There's not much to say, really. The option allows you to do some testing without the need of connecting to a pool. No shares are generated.

Thanks Pooler!
Gained 1 kh/s .

i've lost 0.7 kh/s

Does your processor have AVX?

yes, my Bulldozer x6 an x8 does, haven't tested on x8 yet tho

AMD processors use a different implementation, which (as you can see if you look at the commits to the git repository) has remained unchanged since last version, and actually since version 2.2. For this reason, I'm obliged to respectfully ignore your comment.

Looking to get started on this with W7 x64 and an i7 3930K, I tried running the minerd.exe and get an json_rpc_call failed message. This is with or without a cfg.json file present in the directory. I ran through the first 6 pages or so of this post and didn't see any full Windows setup. If someone could give me a step by step please I would be happy as a pig in shit.

The online help is your friend, just run "minerd --help". You will get a detailed description of all supported options, and see that you have to use "-c" if you want to use a configuration file.

What does the --s flag do?

There is no --s flag. There's a --scantime flag, whose short version is -s. The miner will accept --s because in the Windows version there's just one long option that starts with "s", and getopt is indulgent. Please note that, since this may change in the future, you should not be using shortened long options such as --s when writing scripts or batch files.
That said, you can safely ignore the option when mining at a pool, since it is ineffective when long polling is enabled. If you are solo mining, you can set it to as low as 1 to minimize the likeliness of generating orphaned blocks.

[pooler]

[...] This time I've also built a couple static binaries for Linux (x86 and x86-64), for those people who really don't want to bother compiling the miner themselves.

Static binaries? Uh, nice!
Due to the libcurl deps I'm failing badly to a build static bin myself (on RHEL5). What flags did you set?

Compiling libcurl on Linux should be as simple as running configure & make. I disabled some features such as SSL to keep the size down, but that's not at all necessary.
Anyway, once I have a libcurl.a, I just put its path in the environment variable LIBCURL along with -lrt (and possibly other libraries, depending on what features libcurl.a was built with) and build the miner as usual.

[Pontius]

[...]
Compiling libcurl on Linux should be as simple as running configure & make. I disabled some features such as SSL to keep the size down, but that's not at all necessary.
Anyway, once I have a libcurl.a, I just put its path in the environment variable LIBCURL along with -lrt (and possibly other libraries, depending on what features libcurl.a was built with) and build the miner as usual.

Arrrgh. RHEL 'libcurl' depends on '-lgssapi_krb5', that one I was missing.

[ErebusBat]

I get this error in OS X Mountain Lion:

dyld: Library not loaded: /opt/local/lib/libidn.11.dylib

libidn is installed via fink, so I'm not sure what's up?

This also happens in Lion.

Forcing it to work with brew then causes an error on SSL.  

Is there an official way to make this work?

EDIT: I got pissed and forced it all, but I am pretty sure this isn't the way to do it.  You may need to brew install some packages if they are not already on your system:

Code:

mkdir -p /opt/local/lib
sudo ln /usr/local/Cellar/libidn/1.23/lib/libidn.11.dylib /opt/local/lib/libidn.11.dylib
sudo ln /usr/local/Cellar/openssl/1.0.1a/lib/libssl.1.0.0.dylib /opt/local/lib/libssl.1.0.0.dylib
sudo ln /usr/local/Cellar/openssl/1.0.1a/lib/libcrypto.1.0.0.dylib /opt/local/lib/libcrypto.1.0.0.dylib

[muqali]

Given that AMD now has Piledriver core Ax chips out, will there be FMA support added to the miner? Would FMA even help with the scrypt algorithm?

[pieppiep]

The FMA instruction set is the name of a future extension to the 128 and 256-bit SIMD instructions in the X86 microprocessor instruction set to perform fused multiply–add (FMA) operations.

and

The FMA operation has the form:
d=a+b*c

I know sha doesn't use multiplications and as fas as I know scrypt doesn't either.
So these instructions can't be used.

[muqali]

The FMA instruction set is the name of a future extension to the 128 and 256-bit SIMD instructions in the X86 microprocessor instruction set to perform fused multiply–add (FMA) operations.

and

The FMA operation has the form:
d=a+b*c

I know sha doesn't use multiplications and as fas as I know scrypt doesn't either.
So these instructions can't be used.

Yea, I asked ufasoft if it'd help his miner and he said no. He said scrypt depends a lot on RAM latency. Is it just latency, or does bandwidth count too? ie my quad channel X79 has some benefit? Or should I look more at shaving CAS latency as low as possible? Would either make any real difference?

[pooler]

I know sha doesn't use multiplications and as fas as I know scrypt doesn't either.
So these instructions can't be used.

Yea, I asked ufasoft if it'd help his miner and he said no. He said scrypt depends a lot on RAM latency. Is it just latency, or does bandwidth count too? ie my quad channel X79 has some benefit? Or should I look more at shaving CAS latency as low as possible? Would either make any real difference?

If the processor's caches are large enough (and they usually are), RAM doesn't even come into play. This is the main advantage that CPUs have over GPUs.

[muqali]

I know sha doesn't use multiplications and as fas as I know scrypt doesn't either.
So these instructions can't be used.

Yea, I asked ufasoft if it'd help his miner and he said no. He said scrypt depends a lot on RAM latency. Is it just latency, or does bandwidth count too? ie my quad channel X79 has some benefit? Or should I look more at shaving CAS latency as low as possible? Would either make any real difference?

If the processor's caches are large enough (and they usually are), RAM doesn't even come into play. This is the main advantage that CPUs have over GPUs.

256KB of L2 per core and 12MB of L3, I'd hope that at the least the L3 would be big enough. I don't want to pay a grand for the 3960X

[pieppiep]

The scratchpad used for scrypt is 128.5kB, so a 256kB L2 cache is enough to not have to use the L3 cache. (except maybe when a lot of task switching is done)

For processors without cache or a very small cache you want lower latency more than a higher bandwidth.

[stepkrav]

hello. where can i find a sample configuration file for minerd binary on Linux 32?

[goxed]

The scratchpad used for scrypt is 128.5kB, so a 256kB L2 cache is enough to not have to use the L3 cache. (except maybe when a lot of task switching is done)

For processors without cache or a very small cache you want lower latency more than a higher bandwidth.

It's 256kB but 8-WAY set associative, essentially 32KB per set.

[goxed]

Thanks pooler! Saw improvement right away on a Core i5-3570 K. Thanks

[stepkrav]

i have an Intel i3 2310M with two cores and four threads. Is it normal, to have only 2kilohash/s ?

[pooler]

hello. where can i find a sample configuration file for minerd binary on Linux 32?

A sample configuration file (example-cfg.json) is included in the source tarball. Remember that you have to use the -c option if you want to use a configuration file. See --help output for more info.

i have an Intel i3 2310M with two cores and four threads. Is it normal, to have only 2kilohash/s ?

If that is per-thread and in 32-bit mode, it sounds reasonable.

[muqali]

i have an Intel i3 2310M with two cores and four threads. Is it normal, to have only 2kilohash/s ?

I don't think it is. Sounds very low, even for an i3.

[stepkrav]

hello. where can i find a sample configuration file for minerd binary on Linux 32?

A sample configuration file (example-cfg.json) is included in the source tarball. Remember that you have to use the -c option if you want to use a configuration file. See --help output for more info.

oh, thanks for that i'll try it.

i have an Intel i3 2310M with two cores and four threads. Is it normal, to have only 2kilohash/s ?

If that is per-thread and in 32-bit mode, it sounds reasonable.

it's not per-thread, its a total. I guess i'm doing something wrong. I'll test it again with the .json.