Topic: Ufasoft Miner - Windows/Linux, x86/x64, SSE2/OpenCL, Open Source - page 51. (Read 631037 times)

— and it has successfully submitted 2 POWs, so the calculations are working ok.

Nice work! 

Cheers,

Ok.

I've gotten some reports that the performance on certain AMD cpus isn't as great as it could be.

If anyone has an AMD which shows this, and knows how to profile, I'd appreciate it. I have this gut feel it's in the core code, so unless I learn what the architecture differences are that cause the problem, I may not have a good fix.

Thanks for the hint.  I grabbed 1.1.0 from maverick, it installed on lucid without any problems.

That helped.  It's running at ~2450 khash/sec on a Core2Duo, as opposed to ~1050 khash/sec with the cryptopp algorithm.  Sweet! 

Cheers,

Check your yasm version:

yasm --version

You probably have 0.8.0. You need 1.1.0. I had to compile it myself for some versions of Debian. YASM can be found here:

http://www.tortall.net/projects/yasm/

That one gives me an error in line 106:


I'm looking forward to try this minerd as soon as I can build it.

Cheers,

So I would like to run the miner on my work computer, but port 8332 is blocked.  I'm able to get around this with distributed.net (cracking RC5-72 keys and OGR 27 nodes) by setting it to port 80.  Would this work with the ufasoft miner?

Thanks,
curator

EDIT: Also if this doesn't work out...could I solo mine with this version?  N00b I know.

I have this odd problem of bitcoin-miner.exe working for a few seconds and then going idle for a few seconds as it works. I'm running A Phenom II X6 on Windows 7 64-bit. I'm using these switches: "bitcoin-miner.exe -t 5 -a 10 -o http://deepbit.net:8332 -u [email protected] -p x" I'm wondering if I'm doing something wrong maybe? Has anyone else had this problem? Any feedback would be greatly appreciated.

-edit: This also happens on my secondary computer running three threads on a quad-core. Is there some specific configuration that is required for 64-bit? That is what the two computers have in common so that's the only thing I can think of, other than some connectivity problem with deepbit. I will try running as just a solo-miner to test.

-edit 2: Running with an address of http://127.0.0.1:8332 fixes the problem. I think this is purely an issue of latency, I have satellite internet and the roundtrip time for anything is ridiculous. For whatever reason, poclbm works okay. I guess I'll have to stick with gpu-only mining until I get better internet. Anyway, as far as performance, I get about 15 MHash/s using 5 threads on my AMD processor, so I don't think this is poorly optimized for AMD or anything. Anyone having performance issues using AMD might be some other issue.

Please take a look at https://github.com/chromicant/cpuminer/tree/sse2

This is only for x86_64 machines running Linux!

I have successfully grafted in most of the SSE2 code from the Windows version into jgarzik's CPU miner. It's still a tad slow (I had to remove some loops to make the porting from the Win64 ABI to SysV's ABI for x86_64), but it's 2x faster than the 4way code in my repository (haven't benchmarked it against jgarzik's latest version).

To use this, you need to checkout the sse2 branch of the github project above. You will also need the most recent version of yasm (1.1.0 as of this writing). I haven't figured out how to add .asm/yasm files to auto*...so you'll have to do some manual steps.

Step 1: cd x86_64, run the build.sh script.
Step 2: autogen.sh / configure / make as you'd normally do for cpuminer
Step 3: ./minerd --url http:blah --userpass your:info --algo sse2_64
Step 4: Enjoy.

Watch this branch, since I have some more speedups that I think will work once I clean up some of the internals. This code was pushing 3200 khash/sec per core on my Core i5 760 @ 2.80GHz. So your mileage may vary. No warranty implied. Etc., etc., etc.

Next after cleaning up the code some is to get a proper merge going. Also, this code should be a good basis for my ARM NEON work...so if you want to burn the batteries up in your iPad2...

Please, if you like my work, donate at the address in my sig!

There appears to be a small memory leak somewhere in this program. I haven't looked at the code yet to try and find it.

It looks like it is leaking on the order of about ~25MB per day.

Specs:
- Windows XP laptop
- Intel Core2 Duo 2.2Ghz
- ~4.7MHash/sec
- Slush's pool

Edit: I've got my request time set to the default of every 15 seconds. And ~25MB/day is ~4KB per 15sec. So, it looks like the leak is probably a handle-leak every time a request is made.

my 2 cores got from 3.5MH/s to 9.8 MH/s, is there plan to further improve the program?

hi  
i cant get this to work on solo mining...dont know why.
putting -o http://127.0.0.1:8332/ results in 0 hash i guess the url is incorrect although i dont get error connecting.
can some please help on this?
i can mine through the pools  .

well got from 5mhash to 14Mhash.
4 cores.

I'll give 50 BTC to the person that implements this code into jgarzik's miner. First 25 BTC when working code is released, the other 25 BTC when it's pushed into jgarzik's cpuminer git repository.

This is WinInet HTTP implementation. It sends "Authorization" field after failed "without authorization" request.

The Core2 Duo laptop that I put this on went from ~1.8MH/s (~900KH/s per thread) to ~4.7MH/s (total)

Not that I can see in the fine manual. The code uses some of MASM's IF defines to help make the code portable, which makes things a tad more annoying. As for speed, on my Core i5 760, the speedup was not insignificant...maybe 2xish.

Ah... I did not, actually.  So I guess that makes it only a 2x speedup (the reported numbers went from 2.1 on just about any miner running two threads, to 8.0 reported by ufasoft's miner also running two threads).  Still... A 2x speedup!  Not complaining.   

8 mhash/sec per thread???

The comparison seems off.  You know that cpuminer only prints out per-thread khash/sec not total, right?

Wow... Your miner took my Core2 machine from a mere 2.1MH/s to over 8!

Even weirder, it performs slightly better than the pure-CPU miners on my Athlon64X2 - Which completely sucks at the standard 4-way SSE version.

I don't have many bitcoins yet (finally passed a whole BTC just this past week), but I've sent a dime your way.  Thank you!

Standard GNU assembler should accept MASM format, with a command line switch or two...

I'm currently poking it. I'm assuming it's in MASM format, and trying to port it to YASM now. If I get something that makes sense, I'll put it in my github repo.