[ANN]: cpuminer-opt v3.8.8.1, open source optimized multi-algo CPU miner - page 136.

th3.r00t

sr. member

Activity: 312

Merit: 250

Quote from: Hurtman on September 14, 2016, 11:25:14 AM

doesnt work with AMD Phenom II Cry

I have no problem compiling for AMD Phenom II under Ubuntu and Windows
I have 2 Phenom's and they're compiling fine with -march=btver1, 'cause -march=native don't work on AMD's

Under Ubuntu I use this:

Code:

./autogen.sh && ./configure CFLAGS="-O3 -march=btver1" --with-curl --with-crypto && make && strip cpuminer

Under Windows I use my own mingw64 build environment with these:

Code:

./autogen.sh
CFLAGS="-O3 -march=btver1" CXXFLAGS="$CFLAGS -std=gnu++11" ./configure --with-curl --build=x86_64-w64-mingw32
make
strip -p --strip-debug --strip-unneeded cpuminer.exe

Hurtman

full member

Activity: 180

Merit: 100

doesnt work with AMD Phenom II Cry

Voljanin

newbie

Activity: 28

Merit: 0

Code:

It's probably because the brand string is not implemented on that CPU. There is a test to confirm it is supported but not
implemented correctly. There isn't much to be gained by testing for support, if the brand string isn't available the only
thing that can be done is to tell the users it's unavailable.

cpu_detect seems to have worked around the lack of brand string support, but it's not worth the effort for cpuminer.

As far as AMD compiling goes, I'm thinking it might be easier to just use the Intel architecture compile that matches
the feature set on AMD, unless someone can show a proper AMD compile performs better.

To my regret I could not make it work gcc 6.2 works only gcc 4.8.5 Huh

joblo

legendary

Activity: 1470

Merit: 1114

Quote from: felixbrucker on September 12, 2016, 09:46:14 AM

hi joblo

it seems x11evo benchmark (not sure about actual mining) does take 5 minutes to report hashrate, just fyi, algo does not seem to be profitable for cpu

this is with 3.4.5 where benchmarks work in general

this is the output after 5 minutes:

That's weird. I see the save behaviour. The hash count of the first hash report is much higher than subsequent reports indicating a
longer scan time. However, it is only from one thread, what were the other 3 doing for 5 minutes? Also the high first hashcount
only represents about 1:45 of the 5 minutes at the reported hashrate.

I see no reason for one algo to behave like that, there are no benchmark hooks in x11evo code, it runs the same code as real
mining which works normally. yet lack of hash reports indicate that it spent all that time in x11evo code.

I'll take a look at the code but I'm not opptimistic I'll find anything.

felixbrucker

hero member

Activity: 700

Merit: 500

hi joblo

it seems x11evo benchmark (not sure about actual mining) does take 5 minutes to report hashrate, just fyi, algo does not seem to be profitable for cpu

this is with 3.4.5 where benchmarks work in general

this is the output after 5 minutes:

joblo

legendary

Activity: 1470

Merit: 1114

Quote from: Hurtman on September 11, 2016, 02:09:21 AM

Quote from: joblo on September 10, 2016, 08:52:14 PM

Quote from: Hurtman on September 10, 2016, 08:12:21 PM

where it is possible to download the previous "cpuminer-opt" ?
AMD FX 9590 7 threads [XMR]
ver 3.3.8 = 330 h/s
ver 3.4.6 = 120 h/s ((

If you're using the Windows binaries try the optimized Intel builds. There is no change in cryptonight
between these versions of cpuminer-opt.

Thank you!
avx-i and core-avx2 normal

At startup cpuminer lists the optimization features available on your CPU, the SW build and the algo
you wish to mine. If your CPU does not have AVX2 don't use the AVX2 build. Although it works with cryptonight
it's only because there is no AVX2 code in that algo. Trying to mine an AVX2 optimized algo using the AVX2
optimized build will crash if your CPU does not support AVX2.

torsh

newbie

Activity: 5

Merit: 0

Quote from: joblo on September 10, 2016, 05:35:16 PM

Quote from: torsh on September 10, 2016, 04:12:24 PM

how to do so that would start at cpuminer gets loaded conf.json

-c, --config=FILE load a JSON-format configuration file

automatically without installing -c, --config = FILE how cpuminer-multi

Hurtman

full member

Activity: 180

Merit: 100

Quote from: joblo on September 10, 2016, 08:52:14 PM

Quote from: Hurtman on September 10, 2016, 08:12:21 PM

where it is possible to download the previous "cpuminer-opt" ?
AMD FX 9590 7 threads [XMR]
ver 3.3.8 = 330 h/s
ver 3.4.6 = 120 h/s ((

If you're using the Windows binaries try the optimized Intel builds. There is no change in cryptonight
between these versions of cpuminer-opt.

Thank you!
avx-i and core-avx2 normal

joblo

legendary

Activity: 1470

Merit: 1114

Quote from: Hurtman on September 10, 2016, 08:12:21 PM

where it is possible to download the previous "cpuminer-opt" ?
AMD FX 9590 7 threads [XMR]
ver 3.3.8 = 330 h/s
ver 3.4.6 = 120 h/s ((

If you're using the Windows binaries try the optimized Intel builds. There is no change in cryptonight
between these versions of cpuminer-opt.

Hurtman

full member

Activity: 180

Merit: 100

where it is possible to download the previous "cpuminer-opt" ?
AMD FX 9590 7 threads [XMR]
ver 3.3.8 = 330 h/s
ver 3.4.6 = 120 h/s ((

joblo

legendary

Activity: 1470

Merit: 1114

Quote from: torsh on September 10, 2016, 04:12:24 PM

how to do so that would start at cpuminer gets loaded conf.json

-c, --config=FILE load a JSON-format configuration file

torsh

newbie

Activity: 5

Merit: 0

how to do so that would start at cpuminer gets loaded conf.json

joblo

legendary

Activity: 1470

Merit: 1114

Quote from: th3.r00t on September 10, 2016, 12:22:42 PM

Quote from: joblo on September 10, 2016, 11:39:32 AM

As far as AMD compiling goes, I'm thinking it might be easier to just use the Intel architecture compile that matches
the feature set on AMD, unless someone can show a proper AMD compile performs better.

I am pretty sure that there are performance loss when compiling with Intel's feature set for AMD.
I read recently (don't remember where) that even that the SSE and SSE2 implementations of both major CPU brands are (almoast exactly) the same
the Intel-Intel compile is faster on many ocasions than Intel-AMD.

When I compile with mingw64 and Intel's architecture similar to my Phenom i get a slower binary than AMD's btver1.

The important part is whether Intel compile/AMD CPU is slower than AMD compileAMD CPU. If you have any more data I'd
like to take a look at it to try to figure out why.

Some of the key algos for measuring performance of various features:

AVX2: lyra2v2, hodl
AVX: lyra2v2, hodl
AES: cryptonight, hodl, quark

th3.r00t

sr. member

Activity: 312

Merit: 250

Quote from: joblo on September 10, 2016, 11:39:32 AM

As far as AMD compiling goes, I'm thinking it might be easier to just use the Intel architecture compile that matches
the feature set on AMD, unless someone can show a proper AMD compile performs better.

I am pretty sure that there are performance loss when compiling with Intel's feature set for AMD.
I read recently (don't remember where) that even that the SSE and SSE2 implementations of both major CPU brands are (almoast exactly) the same
the Intel-Intel compile is faster on many ocasions than Intel-AMD.

When I compile with mingw64 and Intel's architecture similar to my Phenom i get a slower binary than AMD's btver1.

joblo

legendary

Activity: 1470

Merit: 1114

Quote from: th3.r00t on September 10, 2016, 10:51:08 AM

I think I've found a somekind a bug in CPU detection.

I've found some old PC with AMD Athlon 64 X2 6000+
After some fails to compile with -march=native, -march=k8-sse3 and -march=nocona I finally had a successful compile with -march=core2.

Starting the binary reports the cpu like this:

Code:

         **********  cpuminer-opt 3.4.6  ***********
     A CPU miner with multi algo support and optimized for CPUs
     with AES_NI and AVX extensions.
     BTC donation address: 12tdvfF7KmAsihBXQXynT6E6th2c2pByTT
     Forked from TPruvot's cpuminer-multi with credits
     to Lucas Jones, elmad, palmd, djm34, pooler, ig0tik3d,
     Wolf0, Jeff Garzik and Optiminer.

CPU: c@
CPU features: SSE2
SW built on Sep 10 2016 with GCC 4.8.4
SW features: SSE2
Algo features: SSE2 AES AVX AVX2
Start mining with SSE2

cpu-detect reports correct CPU string:

Code:

urban@mystique:~$ ~/cpu-detect/cpu-detect
Vendor: AuthenticAMD
Brand: AMD Athlon(tm) 64 X2 Dual Core Processor 6000+
CPU features:
 MMX SSE SSE2 SSE3

It's probably because the brand string is not implemented on that CPU. There is a test to confirm it is supported but not
implemented correctly. There isn't much to be gained by testing for support, if the brand string isn't available the only
thing that can be done is to tell the users it's unavailable.

cpu_detect seems to have worked around the lack of brand string support, but it's not worth the effort for cpuminer.

As far as AMD compiling goes, I'm thinking it might be easier to just use the Intel architecture compile that matches
the feature set on AMD, unless someone can show a proper AMD compile performs better.

th3.r00t

sr. member

Activity: 312

Merit: 250

I think I've found a somekind a bug in CPU detection.

I've found some old PC with AMD Athlon 64 X2 6000+
After some fails to compile with -march=native, -march=k8-sse3 and -march=nocona I finally had a successful compile with -march=core2.

Starting the binary reports the cpu like this:

Code:

         **********  cpuminer-opt 3.4.6  ***********
     A CPU miner with multi algo support and optimized for CPUs
     with AES_NI and AVX extensions.
     BTC donation address: 12tdvfF7KmAsihBXQXynT6E6th2c2pByTT
     Forked from TPruvot's cpuminer-multi with credits
     to Lucas Jones, elmad, palmd, djm34, pooler, ig0tik3d,
     Wolf0, Jeff Garzik and Optiminer.

CPU: c@
CPU features: SSE2
SW built on Sep 10 2016 with GCC 4.8.4
SW features: SSE2
Algo features: SSE2 AES AVX AVX2
Start mining with SSE2

cpu-detect reports correct CPU string:

Code:

urban@mystique:~$ ~/cpu-detect/cpu-detect
Vendor: AuthenticAMD
Brand: AMD Athlon(tm) 64 X2 Dual Core Processor 6000+
CPU features:
 MMX SSE SSE2 SSE3

joblo

legendary

Activity: 1470

Merit: 1114

Quote from: jkminkov on September 10, 2016, 06:58:23 AM

myr-gr algo is all rejects on zpool for windows build, only sse2 works, i7 cpu.

It looks like the AES version of myr-gr is broken and has been for a while, probably since I implemented it.
I recall being disappointed in the performance gain, of lack of, with AES since it forms a major
part of myr-gr. The only thing I can think of is that I mistakenly ran the SSE code when I thought I was testing
the AES code. That would explain both the performance disappointment and the failure to detect that AES
was broken.

Next release will disable AES for myr-gr. In the meantime use the SSE2 Windows binary or compile specifically
for SSE2 with "-march=core2".

jkminkov

hero member

Activity: 698

Merit: 500

myr-gr algo is all rejects on zpool for windows build, only sse2 works, i7 cpu.

joblo

legendary

Activity: 1470

Merit: 1114

Quote from: ljglug on September 08, 2016, 06:34:53 AM

Is there a version of VRM that can be mined?

Not that I am aware of.

ljglug

full member

Activity: 168

Merit: 100

Is there a version of VRM that can be mined?

Topic: [ANN]: cpuminer-opt v3.8.8.1, open source optimized multi-algo CPU miner - page 136. (Read 444067 times)