[ANN][MTC] High Performance CPU Miner for Metiscoin (Linux + Windows)

[gigawatt]

I've compiled my optimized code for each of the following architectures: core2, corei7, corei7-avx, core-avx-i, core-avx2, k8, k8-sse3, barcelona, bdver1, and bdver2
I've also compiled it via Visual Studio 2010 with performance guided optimizations for AMD, Intel, and blend processors (with and without AVX).


The end result is a 10 - 25% speed increase over the stock jh00 miner from yPool!  (Even more if you're on Linux)

Donations appreciated!  MTC: MEu8jBkkVvTLwvpiPjWC9YntyDH2u5KwVy


Which version is for me?

For Linux users: compile the code from github

Code:

git clone https://github.com/llamasoft/xptminer-linux.git
cd xptminer-linux
CFLAGS="-flto -funroll-loops" make

Alternately, check out my All-In-One EZ Setup Script.

For Windows users:
For starters, you cannot go wrong with any of the "favor" versions.
They are compiled in a generic format that should work on all processor architectures.

If you have an Intel processor, use "favorIntel".
If you have an AMD processor, try "favorBlend" then "favorAMD".


Architecture Specific Builds: (oldest to newest)
Intel

• core2
• corei7
• corei7-avx - a.k.a. "sandybridge"
• core-avx-i - a.k.a. "ivybridge"
• core-avx2 - a.k.a. "haswell / 4th generation i7"

   
AMD

• k8 - a.k.a. "opteron / athlon64"
• k8-sse3
• barcelona - a.k.a. "k10"
• bdver1 - a.k.a. "bulldozer"
• bdver2 - a.k.a. "piledriver"

Start with the oldest architecture of your processor manufacturer.
If the program works, write down the hash rate after 5-10 minutes* and try the next version.
If the program crashes, stop and use the previous version with the highest hash rate.
*: Seriously.  At least 5 minutes.  Before then and the hash rate is very inaccurate.

Compare the results to the corresponding "favor" version.
Keep whichever is best!



Where do I get the binaries?

The Windows 64-bit binaries are available for download from two locations:
https://mega.co.nz/#F!YsZSHYKA!IC8LK_MBGwqC-gWOpO7zoQ
https://www.dropbox.com/sh/wtxvxvkirxax2vj/9P_Rxb9V1y (dropbox mirror)

There is no 32-bit version of the optimized code for Windows.
Part of the optimizations force the hash algorithms to use 64-bit mode.  For more info, see this post.



Benchmarks

User	CPU	Miner Version	kHash/s
============	========================	============	============
gigawatt	AMD Phenom II x4 965 @ Stock	favorBlend	637
gigawatt	Intel i5-3210M @ Stock	favorIntel	292
gigawatt	Amazon AWS - c1.xlarge	github	690

Want your benchmark added?  Just post it in the thread and I'll update the list.

[1714131911]

1714131911

[1714131911]

1714131911

[1714131911]

1714131911

[1714131911]

1714131911

[1714131911]

1714131911

[hero18688]

Can you post a avx2 version for 4th generation i7？Thanks!

[gigawatt]

Can you post a avx2 version for 4th generation i7？Thanks!

I've added support for Intel processors with AVX2 support.
I've also added builds for Bulldozer and Piledriver versions of AMD processors.

Enjoy!

[Lowkek]

Thanks for this, downloading now

[vortexfoto]

huh strange, best hashrate i can get from FX-6300 is 650

[gigawatt]

huh strange, best hashrate i can get from FX-6300 is 650

Try the favorAMD and favorBlend builds.  On newer AMD models the favorBlend version should be faster.  For some reason favorAMD means "favor older AMD models" like Opteron and Athlon.

[dragoon1808]

How can I change/remove the 2.5% dev-fee that shows up in ypool?  The -d option doesn't work like in jh00's/clintars version

[hero18688]

Can you post a avx2 version for 4th generation i7？Thanks!

I've added support for Intel processors with AVX2 support.
I've also added builds for Bulldozer and Piledriver versions of AMD processors.

Enjoy!

Thanks.But I found a bit performance lost on avx2 version,with xptMiner_vs2010_favorBLEND my i7-4770 got 955k/s,but with avx2,it drops to 888k/s.I've used this for 1 hour.will watch it later then report.

[Victoo]

Thanks for this! Good work.

[vegascoins]

gigawatt ... have you seen this?

https://github.com/jh000/xptMiner/commit/c3a111715dfe161590575d4073c4576982a33a50

[gigawatt]

I've made some memory and cache optimizations.  I'm now getting an extra 25 kh/s on my Phenom II x4 965.
Was averaging 610 kh/s, now I'm at 635 kh/s.

I've also added Visual Studio AVX builds for your folks with AVX capable processors!  Feel free to give them a try.


The changes have been pushed to github and the downloads updated.

[Wipeout2097]

huh strange, best hashrate i can get from FX-6300 is 650

Using xpt-bdver1 I'm getting 740 Kh/s out of this same CPU, but at 4Ghz. My rig is fucked CPU mining wise though, memory in single channel  

[gigawatt]

huh strange, best hashrate i can get from FX-6300 is 650

Using xpt-bdver1 I'm getting 740 Kh/s out of this same CPU, but at 4Ghz. My rig is fucked CPU mining wise though, memory in single channel  

Try favorBlend and AVX_favorBlend.  You might be able to get a few more khash out of it.

[Wipeout2097]

huh strange, best hashrate i can get from FX-6300 is 650

Using xpt-bdver1 I'm getting 740 Kh/s out of this same CPU, but at 4Ghz. My rig is fucked CPU mining wise though, memory in single channel  

Try favorBlend and AVX_favorBlend.  You might be able to get a few more khash out of it.

I did. xpt-bdver1 gives the highest rate. I don't get it...  

740 Kh/s is good though, considering single-channel ram and how it compares with the i7-4770 above

[crendore]

Awesome but seems slower than the jh00 miner.  925 Kh/s on my i7 4770k, and i was getting 1065 with jh00.
Tested both your avx2 and i7 avx binaries.