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

[felixbrucker]

yup those numbers could add up to the actual hashrate, try with very low core nums like 3,4,5 and post the results, this algo is very memory bound, more cpu power doesnt speed things up but can make things worse actually

also if this system is using multiple cpu you will need to spread the few threads on the other processors to achive maximum hashpower (to use multiple memories), i dont know if this is possible with plain threads (procs work fine) other option would be to start multiple cpuminer for each cpu with low thread count

[mikhan]

@felix
you mean your fork or joblo's one?
btw i got 256 gb of ram on that dual xeon server
will follow your advices later...

[felixbrucker]

@felix
you mean your fork or joblo's one?
btw i got 256 gb of ram on that dual xeon server
will follow your advices later...

i only have a copy of the src in my github acc, no modifications made, im talking about joblos cpuminer-opt

the amount of memory is largely irrelevant, its more the speed from cpu to memory that dictates the hashpower

my xeon e3 has 8 cores, though using more than 4 cores does not result in more hashpower

with 2x 24 threads (2 cpu) i assume the optimum thread count per cpu is about 6-10, maybe a bit more or less depending on the power of each individual core

now on linux placing procs on specific cpus is easy, not sure how to do this on windows.

im unsure how cpuminer-opt handles the placement of threads in a multi cpu environment, if you only specify 12 threads (6 per cpu) it might spread them on cpu0 and cpu1, but it might also spread them only on cpu0 which would result in only half the hashpower

[t2yax]

i've tried zcash mining in windows with mobile ivy bridge cpu.it just stops.i mean windows error.

[joblo]

@felix
you mean your fork or joblo's one?
btw i got 256 gb of ram on that dual xeon server
will follow your advices later...

i only have a copy of the src in my github acc, no modifications made, im talking about joblos cpuminer-opt

the amount of memory is largely irrelevant, its more the speed from cpu to memory that dictates the hashpower

my xeon e3 has 8 cores, though using more than 4 cores does not result in more hashpower

with 2x 24 threads (2 cpu) i assume the optimum thread count per cpu is about 6-10, maybe a bit more or less depending on the power of each individual core

now on linux placing procs on specific cpus is easy, not sure how to do this on windows.

im unsure how cpuminer-opt handles the placement of threads in a multi cpu environment, if you only specify 12 threads (6 per cpu) it might spread them on cpu0 and cpu1, but it might also spread them only on cpu0 which would result in only half the hashpower

Given the 12 thread performance was similar to 48 it looks like it was using both CPUs. With only one CPU I would expect the
performance to drop by half.

I take that back. Adding another CPU doesn't increase memory bandwidth. A dual CPU is overkill, an i7 is overkill. The only way
to improve performance is with LGA2011 which has a 4 channel memory controller.

[joblo]

The branding Intel uses for Celeron CPUs is misleading and is causing confusion with some users.
First the model number don't match the generation, ie some G4xxx models are branded as Skylake
but Skylake Pentiums are the 6xxx series.

But more importantly Celerons are cost reduced and stripped of some advanced technology, including AVX
and AVX2. Therefore the AVX and AVX2 builds will not work on Celerons. The Westmere build is likely the
most featured compile that can be used with Celerons.

I will update the readme file included in the binary package of the next release to clarify.

I would also like to remind users to always read the readme file before reporting problems and to provide a
clear problem description and supporting data. I will ignore any reports without data.

[felixbrucker]

Given the 12 thread performance was similar to 48 it looks like it was using both CPUs. With only one CPU I would expect the
performance to drop by half.

I take that back. Adding another CPU doesn't increase memory bandwidth. A dual CPU is overkill, an i7 is overkill. The only way
to improve performance is with LGA2011 which has a 4 channel memory controller.

afaik each cpu has its own memory, so using multiple cpu actually should multiply the hashpower by the amount of cpus used. if one cpu wants to access the memory of another cpu it would need to do so via qpi. each memory bus per cpu is unique and shouldnt slow down other cpus, only qpi access slows down significantly.

increasing cpu <-> memory speeds should also increase the hashpower

[joblo]

Given the 12 thread performance was similar to 48 it looks like it was using both CPUs. With only one CPU I would expect the
performance to drop by half.

I take that back. Adding another CPU doesn't increase memory bandwidth. A dual CPU is overkill, an i7 is overkill. The only way
to improve performance is with LGA2011 which has a 4 channel memory controller.

afaik each cpu has its own memory, so using multiple cpu actually should multiply the hashpower by the amount of cpus used. if one cpu wants to access the memory of another cpu it would need to do so via qpi. each memory bus per cpu is unique and shouldnt slow down other cpus, only qpi access slows down significantly.

increasing cpu <-> memory speeds should also increase the hashpower

That wasn't the point. A single CPU has access to 4 memory channels instead of 2 doubling the memory bandwidth, The 2nd CPU
is still overkill.

[felixbrucker]

Given the 12 thread performance was similar to 48 it looks like it was using both CPUs. With only one CPU I would expect the
performance to drop by half.

I take that back. Adding another CPU doesn't increase memory bandwidth. A dual CPU is overkill, an i7 is overkill. The only way
to improve performance is with LGA2011 which has a 4 channel memory controller.

afaik each cpu has its own memory, so using multiple cpu actually should multiply the hashpower by the amount of cpus used. if one cpu wants to access the memory of another cpu it would need to do so via qpi. each memory bus per cpu is unique and shouldnt slow down other cpus, only qpi access slows down significantly.

increasing cpu <-> memory speeds should also increase the hashpower

That wasn't the point. A single CPU has access to 4 memory channels instead of 2 doubling the memory bandwidth, The 2nd CPU
is still overkill.

i dont seem to understand your point


taken a system with multiple (say 2) cpu sockets exists and both cpus are present:

- each cpu has dual or quad memory channel (if its dual or quad doesnt matter in this scenario)
- each cpu gives X H/s, like a normal single cpu system would, just two cpu+ram on one mobo
- each cpu uses their own dual/quad memory channel

using one cpu produces X H/s, using both cpu produces 2x X H/s

taken a system with a single cpu socket exists:

- the cpu has dual memory channels

upgrading the dual channel to quad channel through a mobo/cpu upgrade results in (likely) doubled hashrate


im missing the point where a second cpu doesnt speed up the total hashrate of the system if the second cpu has its own memory channels

or do you imply the one cpu should use the other dual channel memory for the other cpu and thus doubling its memory bandwith? cause that wont work afaik

[joblo]

Given the 12 thread performance was similar to 48 it looks like it was using both CPUs. With only one CPU I would expect the
performance to drop by half.

I take that back. Adding another CPU doesn't increase memory bandwidth. A dual CPU is overkill, an i7 is overkill. The only way
to improve performance is with LGA2011 which has a 4 channel memory controller.

afaik each cpu has its own memory, so using multiple cpu actually should multiply the hashpower by the amount of cpus used. if one cpu wants to access the memory of another cpu it would need to do so via qpi. each memory bus per cpu is unique and shouldnt slow down other cpus, only qpi access slows down significantly.

increasing cpu <-> memory speeds should also increase the hashpower

That wasn't the point. A single CPU has access to 4 memory channels instead of 2 doubling the memory bandwidth, The 2nd CPU
is still overkill.

i dont seem to understand your point


taken a system with multiple (say 2) cpu sockets exists and both cpus are present:

- each cpu has dual or quad memory channel (if its dual or quad doesnt matter in this scenario)
- each cpu gives X H/s, like a normal single cpu system would, just two cpu+ram on one mobo
- each cpu uses their own dual/quad memory channel

using one cpu produces X H/s, using both cpu produces 2x X H/s

taken a system with a single cpu socket exists:

- the cpu has dual memory channels

upgrading the dual channel to quad channel through a mobo/cpu upgrade results in (likely) doubled hashrate


im missing the point where a second cpu doesnt speed up the total hashrate of the system if the second cpu has its own memory channels

or do you imply the one cpu should use the other dual channel memory for the other cpu and thus doubling its memory bandwith? cause that wont work afaik

My mistake, I was assuming shared memory.

[felixbrucker]

My mistake, I was assuming shared memory.

ah yes, with shared memory a second cpu would not speed things up

i suppose shared memory only exists in specific systems nowadays?

[My9bot]

hey can you look at zcash algo??
https://github.com/nicehash/nheqminer
thx

[alexcrownjr]

cant build on Centos 7 6x

Code:

In file included from algo/hodl/hodl-gate.h:1:0,
                 from algo/hodl/hodl.cpp:2:
./algo-gate-api.h:126:49: error: using typedef-name ‘json_t’ after ‘struct’
 bool ( *work_decode )            ( const struct json_t*, struct work* );
                                                 ^
In file included from ./miner.h:38:0,
                 from algo/hodl/hodl.cpp:1:
/usr/include/jansson.h:53:3: note: ‘json_t’ has a previous declaration here
 } json_t;
   ^
In file included from algo/hodl/hodl.cpp:1:0:
./miner.h:526:20: warning: ‘algo_names’ defined but not used [-Wunused-variable]
 static const char *algo_names[] = {
                    ^
make[2]: *** [algo/hodl/cpuminer-hodl.o] Error 1
make[2]: Leaving directory `/home/stas/cpuminer-opt'
make[1]: *** [all-recursive] Error 1
make[1]: Leaving directory `/home/stas/cpuminer-opt'
make: *** [all] Error 2

installed
sudo yum groupinstall 'Development Tools'

sudo yum install tmux gmp-devel jansson-devel openssl-devel boost-devel git automake gcc make curl-devel

[joblo]

cant build on Centos 7 6x

Code:

In file included from algo/hodl/hodl-gate.h:1:0,
                 from algo/hodl/hodl.cpp:2:
./algo-gate-api.h:126:49: error: using typedef-name ‘json_t’ after ‘struct’
 bool ( *work_decode )            ( const struct json_t*, struct work* );
                                                 ^
In file included from ./miner.h:38:0,
                 from algo/hodl/hodl.cpp:1:
/usr/include/jansson.h:53:3: note: ‘json_t’ has a previous declaration here
 } json_t;
   ^
In file included from algo/hodl/hodl.cpp:1:0:
./miner.h:526:20: warning: ‘algo_names’ defined but not used [-Wunused-variable]
 static const char *algo_names[] = {
                    ^
make[2]: *** [algo/hodl/cpuminer-hodl.o] Error 1
make[2]: Leaving directory `/home/stas/cpuminer-opt'
make[1]: *** [all-recursive] Error 1
make[1]: Leaving directory `/home/stas/cpuminer-opt'
make: *** [all] Error 2

installed
sudo yum groupinstall 'Development Tools'

sudo yum install tmux gmp-devel jansson-devel openssl-devel boost-devel git automake gcc make curl-devel

I think there were errors before this, I need to see the first ones. Also please provide some of the more mundane
info like the version, where you downloaded from and the build commands you used.

Edit: Try uninstalling jansson-devel. It looks like there is a bug in configure that can't handle when jansson is installed
on the system.

[Marvell1]

ok this makes no sense :
intel G1820 using cpuminer-core2 is getting 569h or 10h/s
the G3250s using cpuminer-core2 are also getting ss high as  500 h or 9.8 h/s
and a brand new skylarke I-7 600k using cpuminer-core-avx2   is getting a max of 580 hs or 9.9 h/s cores hash faster but max out at aroun 90h
a core i3 4170 using cpuminer-core-avx2 gets only 450 h or 8.9 hs/s  - slower hash but each core does 120 h

any ideas as to why the i3 and the skylarke under perform so much on lyra2z

[joblo]

ok this makes no sense :
intel G1820 using cpuminer-core2 is getting 569h or 10h/s
the G3250s using cpuminer-core2 are also getting ss high as  500 h or 9.8 h/s
and a brand new skylarke I-7 600k using cpuminer-core-avx2   is getting a max of 580 hs or 9.9 h/s cores hash faster but max out at aroun 90h
a core i3 4170 using cpuminer-core-avx2 gets only 450 h or 8.9 hs/s  - slower hash but each core does 120 h

any ideas as to why the i3 and the skylarke under perform so much on lyra2z

Yes, I have discussed it extensively. It all about memory bandwidth, not compute power. For kicks try reducing the number of threads
on your 6700K until you see the total hashrate drop. Hint: it's less than 4.

[alexcrownjr]

I think there were errors before this, I need to see the first ones. Also please provide some of the more mundane
info like the version, where you downloaded from and the build commands you used.

Edit: Try uninstalling jansson-devel. It looks like there is a bug in configure that can't handle when jansson is installed
on the system.

without jansson-devel got 1 warning but working, thanks

Code:

centos cpuminer-opt]$ gcc --version
gcc (GCC) 4.8.5 20150623 (Red Hat 4.8.5-4)
Copyright (C) 2015 Free Software Foundation, Inc.
This is free software; see the source for copying conditions.  There is NO
warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.

first warning

Code:

./autogen.sh
aclocal: warning: couldn't open directory 'm4': No such file or directory

[Marvell1]

ok this makes no sense :
intel G1820 using cpuminer-core2 is getting 569h or 10h/s
the G3250s using cpuminer-core2 are also getting ss high as  500 h or 9.8 h/s
and a brand new skylarke I-7 600k using cpuminer-core-avx2   is getting a max of 580 hs or 9.9 h/s cores hash faster but max out at aroun 90h
a core i3 4170 using cpuminer-core-avx2 gets only 450 h or 8.9 hs/s  - slower hash but each core does 120 h

any ideas as to why the i3 and the skylarke under perform so much on lyra2z

Yes, I have discussed it extensively. It all about memory bandwidth, not compute power. For kicks try reducing the number of threads
on your 6700K until you see the total hashrate drop. Hint: it's less than 4.

Wow youre rate basically the same hashrate with 3 cores running, but I notice that shares seem to be submitted slower with less cores

or is that just a display issue ?

basically this CPU is useless for mining this and zcoin which is why I got it what a wast of $299

[Marvell1]

ok this makes no sense :
intel G1820 using cpuminer-core2 is getting 569h or 10h/s
the G3250s using cpuminer-core2 are also getting ss high as  500 h or 9.8 h/s
and a brand new skylarke I-7 600k using cpuminer-core-avx2   is getting a max of 580 hs or 9.9 h/s cores hash faster but max out at aroun 90h
a core i3 4170 using cpuminer-core-avx2 gets only 450 h or 8.9 hs/s  - slower hash but each core does 120 h

any ideas as to why the i3 and the skylarke under perform so much on lyra2z

Yes, I have discussed it extensively. It all about memory bandwidth, not compute power. For kicks try reducing the number of threads
on your 6700K until you see the total hashrate drop. Hint: it's less than 4.

At my local store that had some broad well - E cards 6 core ones for only $100 more I exchanged my 6700 for a 6800k are boy were you right , even with like 50% usage on the cores I'm seeing speeds

of like 10000 h and 30 h/s  I guess these cards have 28 pc-e lanes vs the normal 16 ?  I cant image the higer end ones wit 40 pc-e lanes thoughs could probably do a sold 60 h/s , crazy these are only $100 more than the i7 6700k  I did need to buy are more
expensive mb though.

ok maybe this was a fluke after windows 10 updates run seems to have slowed down a ton trying to figure out whats what

**edit no fluke
 got the same high hash rates agin but for some reason after a while the miner just stops when it hits those max rates of 9k h or more any ideas ?

seems that the pool kicks the rig off due to shares being such high hash rates ? I get stratum timeouts when it gets to the higher rates

[joblo]

ok this makes no sense :
intel G1820 using cpuminer-core2 is getting 569h or 10h/s
the G3250s using cpuminer-core2 are also getting ss high as  500 h or 9.8 h/s
and a brand new skylarke I-7 600k using cpuminer-core-avx2   is getting a max of 580 hs or 9.9 h/s cores hash faster but max out at aroun 90h
a core i3 4170 using cpuminer-core-avx2 gets only 450 h or 8.9 hs/s  - slower hash but each core does 120 h

any ideas as to why the i3 and the skylarke under perform so much on lyra2z

Yes, I have discussed it extensively. It all about memory bandwidth, not compute power. For kicks try reducing the number of threads
on your 6700K until you see the total hashrate drop. Hint: it's less than 4.

At my local store that had some broad well - E cards 6 core ones for only $100 more I exchanged my 6700 for a 6800k are boy were you right , even with like 50% usage on the cores I'm seeing speeds

of like 10000 h and 30 h/s  I guess these cards have 28 pc-e lanes vs the normal 16 ?  I cant image the higer end ones wit 40 pc-e lanes thoughs could probably do a sold 60 h/s , crazy these are only $100 more than the i7 6700k  I did need to buy are more
expensive mb though.

ok maybe this was a fluke after windows 10 updates run seems to have slowed down a ton trying to figure out whats what

**edit no fluke
 got the same high hash rates agin but for some reason after a while the miner just stops when it hits those max rates of 9k h or more any ideas ?

seems that the pool kicks the rig off due to shares being such high hash rates ? I get stratum timeouts when it gets to the higher rates

Stratum problems are a pool issue.