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

[joblo]

What with this error:

Code:

[2016-09-29 00:57:08] Scrypt buffer allocation failed
[2016-09-29 00:57:08] Scrypt buffer allocation failed
[2016-09-29 00:57:08] Scrypt buffer allocation failed
[2016-09-29 00:57:08] Scrypt buffer allocation failed
[2016-09-29 00:57:08] FAIL: thread 46987800 failed to initialize

Windows 7 x64 with 8GB RAM.

My script:

Code:

cpuminer-btver1 -a scrypt:1048576 --benchmark

That's a memory allocation error, how much do you have free?

Edit: I don't know much about scrypt but that nfactor doesn't look right. The default is 6.

Edit: there is a bug in the error message, the thread id was missing.

[joblo]

@scryptr. Please also post the -msse2 compile that failed. This isn't a simple matter of telling you
the correct way, we need to figure out the correct way first.

@scryptr
The correct commandline for building for your CPU Sempron 145 and Athlon X II is:

Code:

./autogen.sh && CFLAGS="-O3 -march=btver1" CXXFLAGS="$CFLAGS -std=gnu++11" ./configure --with-curl && make && strip cpuminer

Try that and you will have your binary.
Personally I use that on my Phenom's and Sempron 145.

This commandline works flawlessly even on older AMD Athlon 64 X2 6000+ and 5000+

A SUCCESSFUL BUILD RESULTED!!! --

The following screen captures are easier to post than a pages-long compile log:

https://postimg.org/image/olsjajmyj/
Initial launch of CPUMiner-OPT 3.4.8-dev.

https://postimg.org/image/3lwnwwy77/
First accepts for CPUMiner-OPT 3.4.8-dev while mining CryptoNight at NiceHash.com.

Thank you very much for the compile scrypt.  I would appreciate a little explanation of the flags used.  After spending  several hours studying GCC flags for various CPU versions, and attempting to compile with different combinations, your help is a gift.

I am also trying to compile NiceHash's version of tsiv's CCminer for CryptoNight.  Perhaps their git repository is a bit skewed, tsiv's compiles with just "autogen.sh", "configure", and "make".

Thanks again!       --scryptr

Thanks for posting the results. I am also curious about why certain flags didn't work, and especially why something that compiled
crashed with an illegal instruction, and which instruction it crashed on.

The compile flags for cpuminer-opt are more complicated because it supports so many algos written by different people to different
standards.

AMD CPUs, well that's another story. Thanks to th3.r00t for making some sense out of it.

[Nik4691]

That's a memory allocation error, how much do you have free?

Edit: there is a bug in the error message, the thread id was missing.

About 4.5GB. Restarted windows too.

Code:

CPU: AMD FX(tm)-8320 Eight-Core Processor
CPU features: SSE2 AES AVX
SW built on Sep 22 2016 with GCC 4.8.3
SW features: SSE2
Algo features: SSE2
Start mining with SSE2

[2016-09-29 02:26:16] Starting Stratum on stratum+tcp://poolinat0r.com:4444
[2016-09-29 02:26:16] Scrypt buffer allocation failed
[2016-09-29 02:26:16] Scrypt buffer allocation failed
[2016-09-29 02:26:16] 8 miner threads started, using 'scrypt' algorithm.
[2016-09-29 02:26:16] Scrypt buffer allocation failed
[2016-09-29 02:26:16] Scrypt buffer allocation failed
[2016-09-29 02:26:16] Scrypt buffer allocation failed
[2016-09-29 02:26:16] FAIL: thread 52623896 failed to initialize
[2016-09-29 02:26:16]

[joblo]

That's a memory allocation error, how much do you have free?

Edit: there is a bug in the error message, the thread id was missing.

About 4.5GB. Restarted windows too.

Code:

CPU: AMD FX(tm)-8320 Eight-Core Processor
CPU features: SSE2 AES AVX
SW built on Sep 22 2016 with GCC 4.8.3
SW features: SSE2
Algo features: SSE2
Start mining with SSE2

[2016-09-29 02:26:16] Starting Stratum on stratum+tcp://poolinat0r.com:4444
[2016-09-29 02:26:16] Scrypt buffer allocation failed
[2016-09-29 02:26:16] Scrypt buffer allocation failed
[2016-09-29 02:26:16] 8 miner threads started, using 'scrypt' algorithm.
[2016-09-29 02:26:16] Scrypt buffer allocation failed
[2016-09-29 02:26:16] Scrypt buffer allocation failed
[2016-09-29 02:26:16] Scrypt buffer allocation failed
[2016-09-29 02:26:16] FAIL: thread 52623896 failed to initialize
[2016-09-29 02:26:16]

The n factor doesn't look right, the default is 6 for scrypt.

[Nik4691]

That's a memory allocation error, how much do you have free?

Edit: there is a bug in the error message, the thread id was missing.

About 4.5GB. Restarted windows too.

Code:

CPU: AMD FX(tm)-8320 Eight-Core Processor
CPU features: SSE2 AES AVX
SW built on Sep 22 2016 with GCC 4.8.3
SW features: SSE2
Algo features: SSE2
Start mining with SSE2

[2016-09-29 02:26:16] Starting Stratum on stratum+tcp://poolinat0r.com:4444
[2016-09-29 02:26:16] Scrypt buffer allocation failed
[2016-09-29 02:26:16] Scrypt buffer allocation failed
[2016-09-29 02:26:16] 8 miner threads started, using 'scrypt' algorithm.
[2016-09-29 02:26:16] Scrypt buffer allocation failed
[2016-09-29 02:26:16] Scrypt buffer allocation failed
[2016-09-29 02:26:16] Scrypt buffer allocation failed
[2016-09-29 02:26:16] FAIL: thread 52623896 failed to initialize
[2016-09-29 02:26:16]

The n factor doesn't look right, the default is 6 for scrypt.

It is this coin: https://bitcointalk.org/index.php?topic=1540023.0
It says

Code:

Algo: scrypt² (N of 1024² or 128MB per thread)

So, isn't it 1024X1024=1048576?

[joblo]

It is this coin: https://bitcointalk.org/index.php?topic=1540023.0
It says

Code:

Algo: scrypt² (N of 1024² or 128MB per thread)

So, isn't it 1024X1024=1048576?

It says it's a new algo, it won't work with old scrypt.

[vaccin]

Thanks, managed to compile it on ubuntu 16.4.
Two different machines, working great on one, not so on the other:

Pleased with it on my xeon E5-1620 v2
https://i.gyazo.com/e51bae8d0dfa86c3e971d288b2c81524.png

Not so much on my other E3-1231 v3. Here I'll stick with https://github.com/perl5577/cpuminer-multi

[felicita]

i got this error when i try to compile your miner 



someone knows this ??

minerd to compile was no porblem but it has no crptonigth algorithm ...



kind regards

[joblo]

i got this error when i try to compile your miner 



someone knows this ??

minerd to compile was no porblem but it has no crptonigth algorithm ...



kind regards

Missing dependencies, consult README.md and ensure all packages are installed.

[Nik4691]

It is this coin: https://bitcointalk.org/index.php?topic=1540023.0
It says

Code:

Algo: scrypt² (N of 1024² or 128MB per thread)

So, isn't it 1024X1024=1048576?

It says it's a new algo, it won't work with old scrypt.

It's scrypt-N with N factor=1048576.
As a guy suggested in the other thread witch i mention, i set virtual memory at 32GB and now it works.
No way to use such amount though... I still have lots of free memory.

[joblo]

It is this coin: https://bitcointalk.org/index.php?topic=1540023.0
It says

Code:

Algo: scrypt² (N of 1024² or 128MB per thread)

So, isn't it 1024X1024=1048576?

It says it's a new algo, it won't work with old scrypt.

It's scrypt-N with N factor=1048576.
As a guy suggested in the other thread witch i mention, i set virtual memory at 32GB and now it works.
No way to use such amount though... I still have lots of free memory.

Did it actually work for him? What miner did he use?

ScryptN is not supported in cpuminer-opt. The code is present but untested. If they have a working miner I suggest you try it.
I was able to run a benchmark with your command line on Linux with 16 GB mem, however, there is no guarantee it is hashing
correctly. Your memory allocation issue may not be related to the miner.

[Nik4691]

It is this coin: https://bitcointalk.org/index.php?topic=1540023.0
It says

Code:

Algo: scrypt² (N of 1024² or 128MB per thread)

So, isn't it 1024X1024=1048576?

It says it's a new algo, it won't work with old scrypt.

It's scrypt-N with N factor=1048576.
As a guy suggested in the other thread witch i mention, i set virtual memory at 32GB and now it works.
No way to use such amount though... I still have lots of free memory.

Did it actually work for him? What miner did he use?

ScryptN is not supported in cpuminer-opt. The code is present but untested. If they have a working miner I suggest you try it.
I was able to run a benchmark with your command line on Linux with 16 GB mem, however, there is no guarantee it is hashing
correctly. Your memory allocation issue may not be related to the miner.

Yes it works with virtual memory at 32GB. I run the benchmark, and then mine some in their pool. Shares submitted.
They suggest your miner only.
There is a miner in the wallet too, but gives me half the hashes of cpuminer-opt.

[felicita]

i got this error when i try to compile your miner 



someone knows this ??

minerd to compile was no porblem but it has no crptonigth algorithm ...



kind regards

Missing dependencies, consult README.md and ensure all packages are installed.

ok i think i missed some of these
Dependencies

build-essential  (for Ubuntu, Development Tools package group on Fedora)
automake
libjansson-dev
libgmp-dev
libcurl4-openssl-dev
libssl-dev
pthreads
zlib


can you gude me how to instal all them .....
iam really new in linux.....-.-


kind regards

[vaccin]

i got this error when i try to compile your miner 

http://www0.xup.in/tn/2016_09/11791853.png

someone knows this ??

minerd to compile was no porblem but it has no crptonigth algorithm ...



kind regards

Missing dependencies, consult README.md and ensure all packages are installed.

ok i think i missed some of these
Dependencies

build-essential  (for Ubuntu, Development Tools package group on Fedora)
automake
libjansson-dev
libgmp-dev
libcurl4-openssl-dev
libssl-dev
pthreads
zlib


can you gude me how to instal all them .....
iam really new in linux.....-.-


kind regards

Code:

sudo apt-get install build-essential libssl-dev libcurl4-openssl-dev libjansson-dev libgmp-dev automake

[NiceHashSupport]

joblo: we have detected that cryptonight would run faster in most cases when not using all available threads. This has most likely to do with CPU cache; the bigger it is, the more threads it can run fast. Are you aware of this?

Speed increase can be around 10-15% when using less than all available threads and that is not something to simply ignore.

[felixbrucker]

joblo: we have detected that cryptonight would run faster in most cases when not using all available threads. This has most likely to do with CPU cache; the bigger it is, the more threads it can run fast. Are you aware of this?

Speed increase can be around 10-15% when using less than all available threads and that is not something to simply ignore.

i have observed the following:

e3-1265lv2: half threads gives best result
i5-3330, fx-8320e, a10-6800k: max threads gives best results

this is indeed in regards to the cache (noted somewhere in this forum, maybe even this thread, not sure), though some cpus still need the default max threads to return best results even though they have plenty cache

[bit815792215]

2mb per thread

[K1lo]

joblo: we have detected that cryptonight would run faster in most cases when not using all available threads. This has most likely to do with CPU cache; the bigger it is, the more threads it can run fast. Are you aware of this?

Speed increase can be around 10-15% when using less than all available threads and that is not something to simply ignore.

Really? I did some benchmarking on my i7-3770 (8 MB L3 cache) in 4,6,7 and 8 thread usage (4C HT).

My results were as follows.

4 threads average : 123.21 standard deviation 1.36
6 threads average : 139.18 standard deviation 13.57
7 threads average : 144.74 standard deviation 9.21
8 threads average : 157.94 standard deviation 21.55



This is data collected from three different runs over several minutes, approx 50 data points per number of threads.

As you can see the hashing speed for cryptonight does increase when more threads are used, however the consistency of the data lowers. When I was running 4 threads almost all the results were consistent (e.g. 122.3, 124, 122.5, 123.6, 124 .. etc), however as the number of threads increased to cover hyperthreaded cores that value started to fluctuated wildly. For example, for 8 threads I saw values ranging from a min of 125.0 to a max of 208.4! However, when averaged out you can still see that more threads = faster hashing rate. I didn't try more threads than the number of physical and logical cores on my machine.

[joblo]

joblo: we have detected that cryptonight would run faster in most cases when not using all available threads. This has most likely to do with CPU cache; the bigger it is, the more threads it can run fast. Are you aware of this?

Speed increase can be around 10-15% when using less than all available threads and that is not something to simply ignore.

Really? I did some benchmarking on my i7-3770 (8 MB L3 cache) in 4,6,7 and 8 thread usage (4C HT).

My results were as follows.

4 threads average : 123.21 standard deviation 1.36
6 threads average : 139.18 standard deviation 13.57
7 threads average : 144.74 standard deviation 9.21
8 threads average : 157.94 standard deviation 21.55



This is data collected from three different runs over several minutes, approx 50 data points per number of threads.

As you can see the hashing speed for cryptonight does increase when more threads are used, however the consistency of the data lowers. When I was running 4 threads almost all the results were consistent (e.g. 122.3, 124, 122.5, 123.6, 124 .. etc), however as the number of threads increased to cover hyperthreaded cores that value started to fluctuated wildly. For example, for 8 threads I saw values ranging from a min of 125.0 to a max of 208.4! However, when averaged out you can still see that more threads = faster hashing rate. I didn't try more threads than the number of physical and logical cores on my machine.

My results are consistent with Nicehash. I get best performance with 4 threads with  i7 and 3 threads with i5 for cryptonight.
Your performance seems low for an i7 no matter the number of threads you used. My i5-2400 is faster than your i7-3770.

[K1lo]

My results are consistent with Nicehash. I get best performance with 4 threads with  i7 and 3 threads with i5 for cryptonight.
Your performance seems low for an i7 no matter the number of threads you used. My i5-2400 is faster than your i7-3770.

huh really? This PC is running a 480 which is mining also, but looking at the stats it doesn't seem to be using the CPU at all. Additionally I've checked and the CPU is not thermally throttling or anything like that. 

I'm not doing anything particulary strange with my cpuminer config. I'm running the most recent 3.4.7 build using cpuminer-core-avx-i.exe with the command:

cpuminer-core-avx-i.exe -a cryptonight -o stratum+tcp://xmr-eu.dwarfpool.com:8005 -u [wallet address] -q

Any ideas why I might be getting such a low hash rate on this machine? On another machine with an i7-4770 I get around 200 H/s using the cpuminer-core-avx2 binary.