[ANN] [XMG] MAGI | CPU mining | mPoW | mPoS | [MagiPay]

[sleepdog]

Some quick mining benchmarks:

On a machinewith 2 x Xeon E5620, 16 threads. Same .bat for each.

Testz compiled v1.1 : 80 Kh/s
DeMarcus v0921 generic : 89 Kh/s
Spexx : 42 Hh/s

Something weird is happening with Spexx's version. If I run another instance simultaneously it picks up to a combined total of ~70 Kh/s.

[sleepdog]

Maybe Spexx's doesn't handle multi processors well?

On a single X5570 system:

Testz compiled v1.1 : 45 Kh/s
DeMarcus v0921 generic : crashes (needs a CPU with AES support?)
Spexx : 50 Kh/s

EDIT : confirmed on a few other servers that DeMarcus's crashes if there's no AES-NI capable CPU present and Spexx's only utilises half the available hashing power in a dual CPU machine.

[MarcusDe]

Maybe Spexx's doesn't handle multi processors well?

On a single X5570 system:

Testz compiled v1.1 : 45 Kh/s
DeMarcus v0921 generic : crashes (needs a CPU with AES support?)
Spexx : 50 Kh/s

EDIT : confirmed on a few other servers that DeMarcus's crashes if there's no AES-NI capable CPU present and Spexx's only utilises half the available hashing power in a dual CPU machine.

Miners you mention are older versions, I sorted all now ;-)

Please check you CPU architecture - if you run higher version than supported by your CPU, your CPU doesn't support instruction set,  app crashes. It's normal :-)

core2 (should run on all "new" CPUs) -> westmere (runs great on my Xeons) -> sandybridge -> haswell

I can make ivybridge, but it won't be much faster, imo.

[MarcusDe]

Some quick mining benchmarks:

On a machinewith 2 x Xeon E5620, 16 threads. Same .bat for each.

Testz compiled v1.1 : 80 Kh/s
DeMarcus v0921 generic : 89 Kh/s
Spexx : 42 Hh/s

Something weird is happening with Spexx's version. If I run another instance simultaneously it picks up to a combined total of ~70 Kh/s.

Take westmere version on 2 x Xeon E5620. I have same machine and it runs without problems there.

http://graymines.net/miners/magi/

[Spexx]

Some quick mining benchmarks:

On a machinewith 2 x Xeon E5620, 16 threads. Same .bat for each.

Testz compiled v1.1 : 80 Kh/s
DeMarcus v0921 generic : 89 Kh/s
Spexx : 42 Hh/s

Something weird is happening with Spexx's version. If I run another instance simultaneously it picks up to a combined total of ~70 Kh/s.

The weird behavior you noticed was discovered in an earlier version compiled by somebody else. The problem arises when using --threads greater than 1 and you can see what is going on by using the Windows Task Manager (and Process Explorer if you are a proper geek ). Try starting 16 instances of minerd with --threads 1 and you should see an immediate significant improvement. I found that a small further improvement was obtained by forcing each instance to use a predetermined set of three cores, balancing the load. To do this, create a batch file with startup commands thus:-

start /low /min /affinity 0x7 minerd etc
start /low /min /affinity 0xb minerd etc
start /low /min /affinity 0xd minerd etc
start /low /min /affinity 0xe minerd etc
start /low /min /affinity 0x70 minerd etc
start /low /min /affinity 0xb0 minerd etc
start /low /min /affinity 0xd0 minerd etc
start /low /min /affinity 0xe0 minerd etc
start /low /min /affinity 0x700 minerd etc
start /low /min /affinity 0xb00 minerd etc
start /low /min /affinity 0xd00 minerd etc
start /low /min /affinity 0xe00 minerd etc
start /low /min /affinity 0x7000 minerd etc
start /low /min /affinity 0xb000 minerd etc
start /low /min /affinity 0xd000 minerd etc
start /low /min /affinity 0xe000 minerd etc

I found with further testing that it was possible to wring a tiny bit more performance out by having each process/thread connect to a unique worker profile on the pool(s) and the method above will also allow simultaneous connection to more than one pool. I have a quad-core processor with four minerd processes running thus:-

start /low /min /affinity 0x7 minerd --url stratum+tcp::/xmg.suprnova.cc:7127 --user Spexx.Spexx --pass stratocaster --threads 1 --quiet
start /low /min /affinity 0xb minerd --url stratum+tcp::/xmg.suprnova.cc:7127 --user Spexx.Spexx1 --pass stratocaster --threads 1 --quiet
start /low /min /affinity 0xd minerd --url stratum+tcp://mine2.magi.nonce-pool.com:4090 --user Spexx.Spexx --pass stratocaster --threads 1 --quiet
start /low /min /affinity 0xe minerd --url stratum+tcp://mine2.magi.nonce-pool.com:4090 --user Spexx.Spexx1 --pass stratocaster --threads 1 --quiet


This should help a great deal. I would be very interested to see a revised set of results from your machine when using this balanced startup regime with the various versions of minerd now available.

[Spexx]

I can make ivybridge, but it won't be much faster, imo.

I thought that Ivybridge was a "shrink" of Sandybridge with the same instruction sets on both, so anything compiled for one should work on the other. Correct me if I'm wrong.

[Spexx]

Some quick mining benchmarks:

On a machinewith 2 x Xeon E5620, 16 threads. Same .bat for each.

Testz compiled v1.1 : 80 Kh/s
DeMarcus v0921 generic : 89 Kh/s
Spexx : 42 Hh/s

Something weird is happening with Spexx's version. If I run another instance simultaneously it picks up to a combined total of ~70 Kh/s.

The weird behavior you noticed was discovered in an earlier version compiled by somebody else. The problem arises when using --threads greater than 1 and you can see what is going on by using the Windows Task Manager (and Process Explorer if you are a proper geek ). Try starting 16 instances of minerd with --threads 1 and you should see an immediate significant improvement. I found that a small further improvement was obtained by forcing each instance to use a predetermined set of three cores, balancing the load. To do this, create a batch file with startup commands thus:-

start /low /min /affinity 0x7 minerd etc
start /low /min /affinity 0xb minerd etc
start /low /min /affinity 0xd minerd etc
start /low /min /affinity 0xe minerd etc
start /low /min /affinity 0x70 minerd etc
start /low /min /affinity 0xb0 minerd etc
start /low /min /affinity 0xd0 minerd etc
start /low /min /affinity 0xe0 minerd etc
start /low /min /affinity 0x700 minerd etc
start /low /min /affinity 0xb00 minerd etc
start /low /min /affinity 0xd00 minerd etc
start /low /min /affinity 0xe00 minerd etc
start /low /min /affinity 0x7000 minerd etc
start /low /min /affinity 0xb000 minerd etc
start /low /min /affinity 0xd000 minerd etc
start /low /min /affinity 0xe000 minerd etc

I found with further testing that it was possible to wring a tiny bit more performance out by having each process/thread connect to a unique worker profile on the pool(s) and the method above will also allow simultaneous connection to more than one pool. I have a quad-core processor with four minerd processes running thus:-

start /low /min /affinity 0x7 minerd --url stratum+tcp::/xmg.suprnova.cc:7127 --user Spexx.Spexx --pass stratocaster --threads 1 --quiet
start /low /min /affinity 0xb minerd --url stratum+tcp::/xmg.suprnova.cc:7127 --user Spexx.Spexx1 --pass stratocaster --threads 1 --quiet
start /low /min /affinity 0xd minerd --url stratum+tcp://mine2.magi.nonce-pool.com:4090 --user Spexx.Spexx --pass stratocaster --threads 1 --quiet
start /low /min /affinity 0xe minerd --url stratum+tcp://mine2.magi.nonce-pool.com:4090 --user Spexx.Spexx1 --pass stratocaster --threads 1 --quiet


This should help a great deal. I would be very interested to see a revised set of results from your machine when using this balanced startup regime with the various versions of minerd now available.

Oh hold on a minute. Your processor has four cores, so it would be best to use just 4 threads i.e. just the first four command lines of the startup batch file above. The remaining command lines will not work for you - they attempt to use 12 further cores which you don't have.

[MarcusDe]

I can make ivybridge, but it won't be much faster, imo.

I thought that Ivybridge was a "shrink" of Sandybridge with the same instruction sets on both, so anything compiled for one should work on the other. Correct me if I'm wrong.

‘sandybridge’
MMX, SSE, SSE2, SSE3, SSSE3, SSE4.1, SSE4.2, POPCNT, AVX, AES and PCLMUL

‘ivybridge’
MMX, SSE, SSE2, SSE3, SSSE3, SSE4.1, SSE4.2, POPCNT, AVX, AES, PCLMUL, FSGSBASE, RDRND and F16C


Anyway, lets see if miner will use any of those 3 new instructions, omw to make ivybridge version ;-)

[cakir]

I can make ivybridge, but it won't be much faster, imo.

I thought that Ivybridge was a "shrink" of Sandybridge with the same instruction sets on both, so anything compiled for one should work on the other. Correct me if I'm wrong.

‘sandybridge’
MMX, SSE, SSE2, SSE3, SSSE3, SSE4.1, SSE4.2, POPCNT, AVX, AES and PCLMUL

‘ivybridge’
MMX, SSE, SSE2, SSE3, SSSE3, SSE4.1, SSE4.2, POPCNT, AVX, AES, PCLMUL, FSGSBASE, RDRND and F16C


Anyway, lets see if miner will use any of those 3 new instructions, omw to make ivybridge version ;-)

@MarcusDe You're awesome!

I'm getting ~24 khs now!
Processor: i5 2410m and miner version: sandybridge 0922

[bbr]

Newest WIN64 miners for all new intel architectures:

http://graymines.net/miners/magi/

This haswell release is getting me 74kh/s but your previous release is getting me 81kh/s

[24core]

Maybe Spexx's doesn't handle multi processors well?

On a single X5570 system:

Testz compiled v1.1 : 45 Kh/s
DeMarcus v0921 generic : crashes (needs a CPU with AES support?)
Spexx : 50 Kh/s

EDIT : confirmed on a few other servers that DeMarcus's crashes if there's no AES-NI capable CPU present and Spexx's only utilises half the available hashing power in a dual CPU machine.

I have a dual Xeon ES server 2.3Ghz 32 core and I get 153 Kh/s

Where is the link to the Spexx version so I can test ?

[Spexx]

I can make ivybridge, but it won't be much faster, imo.

I thought that Ivybridge was a "shrink" of Sandybridge with the same instruction sets on both, so anything compiled for one should work on the other. Correct me if I'm wrong.

‘sandybridge’
MMX, SSE, SSE2, SSE3, SSSE3, SSE4.1, SSE4.2, POPCNT, AVX, AES and PCLMUL

‘ivybridge’
MMX, SSE, SSE2, SSE3, SSSE3, SSE4.1, SSE4.2, POPCNT, AVX, AES, PCLMUL, FSGSBASE, RDRND and F16C


Anyway, lets see if miner will use any of those 3 new instructions, omw to make ivybridge version ;-)

Thanks for that and I stand corrected The documentation provided with the source code does mention using the AVX instruction set if available. Your build for Haswell crashes on an AMD FX4100 (as expected) so the compiler must be using some of the advanced features available under Haswell. All your other versions are stable on the FX4100 and on an Intel i7 4510u but run at slightly different speeds. I am very interested in which compiler you have been using. For my 64 bit version I used Cygwin with CFLAGS="-O3 -march=generic -mtune=generic" CXXFLAGS="-O3 -march=generic -mtune=generic".

Edit: Oops - CFLAGS="-O3 mtune=generic" CXXFLAGS="-O3 -mtune=generic" and the ./configure command line included --build=x86_64-pc-cygwin

[kondiomir]

Newest WIN64 miners for all new intel architectures:

http://graymines.net/miners/magi/

Nice

AMD FX ?

[Spexx]

Maybe Spexx's doesn't handle multi processors well?

On a single X5570 system:

Testz compiled v1.1 : 45 Kh/s
DeMarcus v0921 generic : crashes (needs a CPU with AES support?)
Spexx : 50 Kh/s

EDIT : confirmed on a few other servers that DeMarcus's crashes if there's no AES-NI capable CPU present and Spexx's only utilises half the available hashing power in a dual CPU machine.

I have a dual Xeon ES server 2.3Ghz 32 core and I get 153 Kh/s

Where is the link to the Spexx version so I can test ?

https://www.dropbox.com/s/hiadf8o5u7bsit7/XMGpoolminerWin64.rar?dl=0

[MarcusDe]

Newest WIN64 miners for all new intel architectures:

http://graymines.net/miners/magi/

Nice! Ivybridge?

Here you go!

http://graymines.net/miners/magi/cpuminer_win64_magi_byMarcusDe-ivybridge-0922.zip

[MarcusDe]

Newest WIN64 miners for all new intel architectures:

http://graymines.net/miners/magi/

Nice

AMD FX ?

Can u paste cpu-z image or just give me detailed info what model?

[24core]

Maybe Spexx's doesn't handle multi processors well?

On a single X5570 system:

Testz compiled v1.1 : 45 Kh/s
DeMarcus v0921 generic : crashes (needs a CPU with AES support?)
Spexx : 50 Kh/s

EDIT : confirmed on a few other servers that DeMarcus's crashes if there's no AES-NI capable CPU present and Spexx's only utilises half the available hashing power in a dual CPU machine.

I have a dual Xeon ES server 2.3Ghz 32 core and I get 153 Kh/s

Where is the link to the Spexx version so I can test ?

https://www.dropbox.com/s/hiadf8o5u7bsit7/XMGpoolminerWin64.rar?dl=0

only got 41kh/s, with my other miner I get 153 kh/s!!! - lol

[Lightsplasher]

@joelao95
Exactly those miners that should stay are slowly stop mining.
Only 2 guys with over 10H/s are more then half of rate that you are targeting and they seems are going to stay.

Also, why pools rewarding blockfinders?

edit: Nice. Nethashrate under 40 and those both guys stays and well rewarded by pool.
However idk how nethashrate can be under 40MH/s when both pools have around 40.

Hey maybe a nice public service announcement will work:  The target hash rate is 50 MH/sec and anyone taking a significant percentage of this is harming the community.  Let all play fair; hugs and kisses, unicorns, rainbows, spread the love people.  - Glenda the good.

Or perhaps:  By edict of the high command any party caught taking more than their fair share of the communal hash rate will have their ip banned their equipment hung drawn and quartered and the useless hulk displayed on the tower of London bridge for the rest of the community to behold.  - King Henry the Ninth.

Considering the market cap of this coin – just too funny.    But at least we are burning less power as a community now. 

[24core]

Newest WIN64 miners for all new intel architectures:

http://graymines.net/miners/magi/

Cheers !!!

I am now getting 180 kh/s !!!

[Spexx]

Maybe Spexx's doesn't handle multi processors well?

On a single X5570 system:

Testz compiled v1.1 : 45 Kh/s
DeMarcus v0921 generic : crashes (needs a CPU with AES support?)
Spexx : 50 Kh/s

EDIT : confirmed on a few other servers that DeMarcus's crashes if there's no AES-NI capable CPU present and Spexx's only utilises half the available hashing power in a dual CPU machine.

I have a dual Xeon ES server 2.3Ghz 32 core and I get 153 Kh/s

Where is the link to the Spexx version so I can test ?

https://www.dropbox.com/s/hiadf8o5u7bsit7/XMGpoolminerWin64.rar?dl=0

only got 41kh/s, with my other miner I get 153 kh/s!!! - lol

Pmsl. Now stop laughing and follow the instructions for starting multiple instances of minerd with one thread per instance, as mentioned here:-

https://bitcointalk.org/index.php?topic=735170.msg8926975#msg8926975

I bet you one XMG that you get more than 153 Kh/s