Topic: [ANN]: cpuminer-opt v3.8.8.1, open source optimized multi-algo CPU miner - page 179. (Read 444067 times)
April 04, 2016, 08:55:07 PM
I have actually always wanted to ask this question. It is somewhat off-topic but to a degree still related. Why is it that Bitcoin mining is now difficult to proceed to with a regular desktop computer? Most around this forum have made a point every time to mention that desktop computers cannot support Bitcoin mining software.
April 03, 2016, 05:08:22 AM
cpuminer-opt v3.1.9 is ready for download
https://drive.google.com/file/d/0B0lVSGQYLJIZYVZqcmZ6TzBPZHM/view?usp=sharing
New in v3.1.9
- decred added (blake256r14)
- myr-gr optimized for AES_NI, marginal increase in hashrate
- groestl, keccak & blakecoin fixed.
- x11gost is added as an alias for sib
https://drive.google.com/file/d/0B0lVSGQYLJIZYVZqcmZ6TzBPZHM/view?usp=sharing
New in v3.1.9
- decred added (blake256r14)
- myr-gr optimized for AES_NI, marginal increase in hashrate
- groestl, keccak & blakecoin fixed.
- x11gost is added as an alias for sib
Thanks!
It seems that ZR5 on v3.1.9 AES-NI compile is a little faster than v3.1.7 (about 5 khash)
But decred is just SSE2 on both Intel and AMD - almost same speed per thread on Intel i7 and Phenom II x4 940?
Strange thing is that both of them don't succeed to submit a share for about 5 minutes. (Maybe the network diff is too high now).
April 02, 2016, 07:13:26 PM
cpuminer-opt v3.1.9 is ready for download
https://drive.google.com/file/d/0B0lVSGQYLJIZYVZqcmZ6TzBPZHM/view?usp=sharing
New in v3.1.9
- decred added (blake256r14)
- myr-gr optimized for AES_NI, marginal increase in hashrate
- groestl, keccak & blakecoin fixed.
- x11gost is added as an alias for sib
Other recently added or improved algos
- x14 added with AES_NI and SSE2 optimisations
- blake added
- blake2s added
- vanilla (blake256r8vnl) added
- sib (x11gost) algo optimised:
+116% AES_NI + SSE2
+63% SSE2
- nist5 algo SSE optimised
+111% AES_NI + SSE2
+68% SSE2
- c11 algo +57% optimized for AES_NI & SSE2
- zr5 algo optimized for AES_NI and SSE2
+98% AES_NI + SSE2
+45% SSE2
https://drive.google.com/file/d/0B0lVSGQYLJIZYVZqcmZ6TzBPZHM/view?usp=sharing
New in v3.1.9
- decred added (blake256r14)
- myr-gr optimized for AES_NI, marginal increase in hashrate
- groestl, keccak & blakecoin fixed.
- x11gost is added as an alias for sib
Other recently added or improved algos
- x14 added with AES_NI and SSE2 optimisations
- blake added
- blake2s added
- vanilla (blake256r8vnl) added
- sib (x11gost) algo optimised:
+116% AES_NI + SSE2
+63% SSE2
- nist5 algo SSE optimised
+111% AES_NI + SSE2
+68% SSE2
- c11 algo +57% optimized for AES_NI & SSE2
- zr5 algo optimized for AES_NI and SSE2
+98% AES_NI + SSE2
+45% SSE2
March 29, 2016, 06:12:35 PM
Nice work, nice bump, went from 890kh/s on this one rig to 940kh/s over 3.1.5 I was using on it.
Intel Duel CPU 21 cores used on EVGA SR2 MB.
Intel Duel CPU 21 cores used on EVGA SR2 MB.
Which algo?
W7 64bit w/ VB Linux
Not sure how that happened, I didn't touch x11. There is no change for me.
Maybe you had fewer processes running.
My AMD rig is seeing a small loss 2291 v3.07 down to 2243 v3.1.8
48 cores full Linux 64bit also x11
If this is a NO_AES_NI compile then, yes there were improvements to many algos.
March 29, 2016, 05:53:49 PM
Nice work, nice bump, went from 890kh/s on this one rig to 940kh/s over 3.1.5 I was using on it.
Intel Duel CPU 21 cores used on EVGA SR2 MB.
Intel Duel CPU 21 cores used on EVGA SR2 MB.
Which algo?
W7 64bit w/ VB Linux
Not sure how that happened, I didn't touch x11. There is no change for me.
Maybe you had fewer processes running.
My AMD rig is seeing a small loss 2291 v3.07 down to 2243 v3.1.8
48 cores full Linux 64bit also x11
March 29, 2016, 05:50:56 PM
Nice work, nice bump, went from 890kh/s on this one rig to 940kh/s over 3.1.5 I was using on it.
Intel Duel CPU 21 cores used on EVGA SR2 MB.
Intel Duel CPU 21 cores used on EVGA SR2 MB.
Which algo?
W7 64bit w/ VB Linux
Not sure how that happened, I didn't touch x11. There is no change for me.
Maybe you had fewer processes running.
March 29, 2016, 05:45:28 PM
Nice work, nice bump, went from 890kh/s on this one rig to 940kh/s over 3.1.5 I was using on it.
Intel Duel CPU 21 cores used on EVGA SR2 MB.
Intel Duel CPU 21 cores used on EVGA SR2 MB.
Which algo?
W7 64bit w/ VB Linux
March 29, 2016, 05:44:46 PM
Nice work, nice bump, went from 890kh/s on this one rig to 940kh/s over 3.1.5 I was using on it.
Intel Duel CPU 21 cores used on EVGA SR2 MB.
Intel Duel CPU 21 cores used on EVGA SR2 MB.
Which algo?
March 29, 2016, 05:43:19 PM
Nice work, nice bump, went from 890kh/s on this one rig to 940kh/s over 3.1.5 I was using on it.
Intel Duel CPU 21 cores used on EVGA SR2 MB 3.333Ghz speed.
Intel Duel CPU 21 cores used on EVGA SR2 MB 3.333Ghz speed.
March 29, 2016, 10:25:57 AM
Thanks!
I used the following:
and
But it seems that still got an error:
With
I got a successful compile on AMD and it mines at 370 khash, which is about 10 khash slower than my scmorse build.
and so far the best is:
which gets 373-375 khash on that Phenom II.
P.S. Stall searching my Ubuntu image for cross-compile.
If I don't find it I will use MSYS and Win 7 x64...
I used the following:
Code:
./configure CFLAGS="-DNO_AES_NI -O3 -march=native" --with-curl --with-crypto
Code:
./configure CFLAGS="-DNO_AES_NI -O3 -march=amdfam10" --with-curl --with-crypto
But it seems that still got an error:
Code:
algo/echo/aes_ni/hash.c:387:4: note: in expansion of macro ‘TRANSFORM’
TRANSFORM(_state[i][j], _k_opt, t1, t2);
^
make[2]: *** [algo/echo/aes_ni/cpuminer-hash.o] Error 1
make[2]: Leaving directory `/home/***/cpuminer-opt-3.1.7'
make[1]: *** [all-recursive] Error 1
make[1]: Leaving directory `/home/***/cpuminer-opt-3.1.7'
make: *** [all] Error 2
With
Code:
./configure CFLAGS="-DNO_AES_NI -O3 -march=core2" --with-curl --with-crypto
and so far the best is:
Code:
./configure CFLAGS="-DNO_AES_NI -O3 -march=k8-sse3" --with-curl --with-crypto
P.S. Stall searching my Ubuntu image for cross-compile.
If I don't find it I will use MSYS and Win 7 x64...
The SSE2 compile should be on par with scmorse.
Perhaps I should change the recommendation to always use -march=core2 for all non-aesni cpus since it works
on AMD as well.
March 29, 2016, 07:03:13 AM
Thanks!
I used the following:
and
But it seems that still got an error:
With
I got a successful compile on AMD and it mines at 370 khash, which is about 10 khash slower than my scmorse build.
and so far the best is:
which gets 373-375 khash on that Phenom II.
P.S. Stall searching my Ubuntu image for cross-compile.
If I don't find it I will use MSYS and Win 7 x64...
I used the following:
Code:
./configure CFLAGS="-DNO_AES_NI -O3 -march=native" --with-curl --with-crypto
Code:
./configure CFLAGS="-DNO_AES_NI -O3 -march=amdfam10" --with-curl --with-crypto
But it seems that still got an error:
Code:
algo/echo/aes_ni/hash.c:387:4: note: in expansion of macro ‘TRANSFORM’
TRANSFORM(_state[i][j], _k_opt, t1, t2);
^
make[2]: *** [algo/echo/aes_ni/cpuminer-hash.o] Error 1
make[2]: Leaving directory `/home/***/cpuminer-opt-3.1.7'
make[1]: *** [all-recursive] Error 1
make[1]: Leaving directory `/home/***/cpuminer-opt-3.1.7'
make: *** [all] Error 2
With
Code:
./configure CFLAGS="-DNO_AES_NI -O3 -march=core2" --with-curl --with-crypto
and so far the best is:
Code:
./configure CFLAGS="-DNO_AES_NI -O3 -march=k8-sse3" --with-curl --with-crypto
P.S. Stall searching my Ubuntu image for cross-compile.
If I don't find it I will use MSYS and Win 7 x64...
March 29, 2016, 06:15:22 AM
Damn!
Just when I get my hopes up for AMD cpus...
It wont compile on AMD Phenom II X4 940.
This is the error that I get:
Can you do something about that?
In the meantime I will find my cross-compile image so I can try to compile AES-NI and SSE2 win64 builds.
Just when I get my hopes up for AMD cpus...
It wont compile on AMD Phenom II X4 940.
This is the error that I get:
Code:
In file included from algo/cryptonight/cryptonight-aesni.c:2:0:
algo/cryptonight/cryptonight.h:65:6: note: expected ‘uint64_t *’ but argument is of type ‘union hash_state *’
void keccakf(uint64_t st[25], int rounds);
^
make[2]: *** [algo/cryptonight/cpuminer-cryptonight-aesni.o] Error 1
make[2]: Leaving directory `/home/***/cpuminer-opt-3.1.7'
make[1]: *** [all-recursive] Error 1
make[1]: Leaving directory `/home/***/cpuminer-opt-3.1.7'
make: *** [all] Error 2
Can you do something about that?
In the meantime I will find my cross-compile image so I can try to compile AES-NI and SSE2 win64 builds.
Add "-DNO_AES_NI" to the CFLAGS.
I haven't gotten Windows to compile since I forked from TPruvot, the project file hasn't been maintained.
Good luck.
March 29, 2016, 03:52:39 AM
Damn!
Just when I get my hopes up for AMD cpus...
It wont compile on AMD Phenom II X4 940.
This is the error that I get:
Can you do something about that?
In the meantime I will find my cross-compile image so I can try to compile AES-NI and SSE2 win64 builds.
Just when I get my hopes up for AMD cpus...
It wont compile on AMD Phenom II X4 940.
This is the error that I get:
Code:
In file included from algo/cryptonight/cryptonight-aesni.c:2:0:
algo/cryptonight/cryptonight.h:65:6: note: expected ‘uint64_t *’ but argument is of type ‘union hash_state *’
void keccakf(uint64_t st[25], int rounds);
^
make[2]: *** [algo/cryptonight/cpuminer-cryptonight-aesni.o] Error 1
make[2]: Leaving directory `/home/***/cpuminer-opt-3.1.7'
make[1]: *** [all-recursive] Error 1
make[1]: Leaving directory `/home/***/cpuminer-opt-3.1.7'
make: *** [all] Error 2
Can you do something about that?
In the meantime I will find my cross-compile image so I can try to compile AES-NI and SSE2 win64 builds.
March 29, 2016, 03:40:54 AM
Very nice bump in hashrate for ZR5! 
v3.1.7 - goes to 830-840 khash
v3.1.8 - goes to 825-830 khash
Read something about SSE2 optimisations on some algos - does that means that I can use your miner in AMD cpus that have SSE2 (since they lack AES-NI support)?
P.S. You are very welcome for the scmorse's tip.
Right now I'm making your miner default for my i7 cpu.
v3.1.7 - goes to 830-840 khash
v3.1.8 - goes to 825-830 khash
Read something about SSE2 optimisations on some algos - does that means that I can use your miner in AMD cpus that have SSE2 (since they lack AES-NI support)?
P.S. You are very welcome for the scmorse's tip.
Right now I'm making your miner default for my i7 cpu.
March 28, 2016, 03:53:09 PM
Hot on the heels of cpuminer-opt v3.1.7 comes v3.1.8 with more optimisations.
https://drive.google.com/file/d/0B0lVSGQYLJIZVllmU01wSGpvRjA/view?usp=sharing
New in v3.1.8
- sib algo optimised:
+116% AES_NI & SSE2
+63% SSE2
- nist5 algo SSE optimised, in addition to previous AES_NI optimizations
+111% AES_NI & SSE2
+68% SSE2
- c11 algo +57% optimized for AES_NI & SSE2
https://drive.google.com/file/d/0B0lVSGQYLJIZVllmU01wSGpvRjA/view?usp=sharing
New in v3.1.8
- sib algo optimised:
+116% AES_NI & SSE2
+63% SSE2
- nist5 algo SSE optimised, in addition to previous AES_NI optimizations
+111% AES_NI & SSE2
+68% SSE2
- c11 algo +57% optimized for AES_NI & SSE2
March 28, 2016, 10:32:12 AM
The new version 3.1.6 returns only invalid (low-diff) shares with zr5 algo on ziftrpool.io
AKAIK v3.1.5 worked just fine with that pool.
There is some more room for improvement, especially on zr5.
With scmorse/ziftr-cpu which is ig0tik3d/ziftr-cpu with stratum,
I get about 640 khash on Core i7-4790K CPU @ 4.40GHz and with your miner I got as close to 600 khash...
Just my 2 cents.
AKAIK v3.1.5 worked just fine with that pool.
There is some more room for improvement, especially on zr5.
With scmorse/ziftr-cpu which is ig0tik3d/ziftr-cpu with stratum,
I get about 640 khash on Core i7-4790K CPU @ 4.40GHz and with your miner I got as close to 600 khash...
Just my 2 cents.
Thanks for the feedback, I'll look into it.
Edit: No rejects for me with default options at yiimp. I also reviewed the code and found no changes that would affect zr5.
Can you confirm that 3.1.5 & 3.1.6 behave differently? Could you also try yiimp?
When I mine the diff is reported as .013. You could try setting the diff manually. On yaamp clones it is in the password
string ( -p d=n ), but your pool may be different.
Edit2: I took a look at the scmorse implementation and i might be able to do better. His implemetation does not use the
aes-ni optimized groestl. If I can integrate that it should hash even faster.
Merging my optimizations with scmorse gave a 33% increase to zr5 but I'll wait to release it until your issue is understood.
Well after recompile of v3.1.6 on my Ubuntu box there is no low-diff rejects on ziftrpool.io
v3.1.5 is also good, but little bit (about 5 khash) slower.
Guess it was something wrong with the previous compile...
(I think I used ./build.sh script)33% more harshrate on ZR5 sounds just AWESOME!
I expect that my Core i7-4790K will go upto ~800 khash...
Just to compare - an older GPU, AMD HD6870 makes 860-900 khash average with wolf's GPU miner @ 845 MHz...
I haven't looked at build.sh since the fork, I should clean it up. Thanks for the tip about scmorse.
March 28, 2016, 10:28:15 AM
cpuminer-opt v3.1.7 is available for download:
https://drive.google.com/file/d/0B0lVSGQYLJIZdVZlczRsVlFGUWs/view?usp=sharing
New in v3.1.7
For users:
- zr5 algo +33% AES_NI and SSE2 optimised, +98% since v3.1.4
For developpers:
- better handling of algo aliases
- better handling of hashrate display
Recently added algos:
- x14 with AES_NI optimisations
- blake
- blake2s
- vanilla (blake256r14vnl).
- x17, blakecoin & fresh algos, benchmark tested only.
Recently improved algos:
- nist5 +54% AES_NI optimised.
- c11 +27% AES_NI optimised.
- SSE2 hashrate increases in many algos.
See first post and for more details including normalisation factors.
https://drive.google.com/file/d/0B0lVSGQYLJIZdVZlczRsVlFGUWs/view?usp=sharing
New in v3.1.7
For users:
- zr5 algo +33% AES_NI and SSE2 optimised, +98% since v3.1.4
For developpers:
- better handling of algo aliases
- better handling of hashrate display
Recently added algos:
- x14 with AES_NI optimisations
- blake
- blake2s
- vanilla (blake256r14vnl).
- x17, blakecoin & fresh algos, benchmark tested only.
Recently improved algos:
- nist5 +54% AES_NI optimised.
- c11 +27% AES_NI optimised.
- SSE2 hashrate increases in many algos.
See first post and for more details including normalisation factors.
March 28, 2016, 06:13:25 AM
The new version 3.1.6 returns only invalid (low-diff) shares with zr5 algo on ziftrpool.io
AKAIK v3.1.5 worked just fine with that pool.
There is some more room for improvement, especially on zr5.
With scmorse/ziftr-cpu which is ig0tik3d/ziftr-cpu with stratum,
I get about 640 khash on Core i7-4790K CPU @ 4.40GHz and with your miner I got as close to 600 khash...
Just my 2 cents.
AKAIK v3.1.5 worked just fine with that pool.
There is some more room for improvement, especially on zr5.
With scmorse/ziftr-cpu which is ig0tik3d/ziftr-cpu with stratum,
I get about 640 khash on Core i7-4790K CPU @ 4.40GHz and with your miner I got as close to 600 khash...
Just my 2 cents.
Thanks for the feedback, I'll look into it.
Edit: No rejects for me with default options at yiimp. I also reviewed the code and found no changes that would affect zr5.
Can you confirm that 3.1.5 & 3.1.6 behave differently? Could you also try yiimp?
When I mine the diff is reported as .013. You could try setting the diff manually. On yaamp clones it is in the password
string ( -p d=n ), but your pool may be different.
Edit2: I took a look at the scmorse implementation and i might be able to do better. His implemetation does not use the
aes-ni optimized groestl. If I can integrate that it should hash even faster.
Merging my optimizations with scmorse gave a 33% increase to zr5 but I'll wait to release it until your issue is understood.
Well after recompile of v3.1.6 on my Ubuntu box there is no low-diff rejects on ziftrpool.io
v3.1.5 is also good, but little bit (about 5 khash) slower.
Guess it was something wrong with the previous compile...
(I think I used ./build.sh script)33% more harshrate on ZR5 sounds just AWESOME!
I expect that my Core i7-4790K will go upto ~800 khash...
Just to compare - an older GPU, AMD HD6870 makes 860-900 khash average with wolf's GPU miner @ 845 MHz...
March 27, 2016, 12:35:34 PM
The new version 3.1.6 returns only invalid (low-diff) shares with zr5 algo on ziftrpool.io
AKAIK v3.1.5 worked just fine with that pool.
There is some more room for improvement, especially on zr5.
With scmorse/ziftr-cpu which is ig0tik3d/ziftr-cpu with stratum,
I get about 640 khash on Core i7-4790K CPU @ 4.40GHz and with your miner I got as close to 600 khash...
Just my 2 cents.
AKAIK v3.1.5 worked just fine with that pool.
There is some more room for improvement, especially on zr5.
With scmorse/ziftr-cpu which is ig0tik3d/ziftr-cpu with stratum,
I get about 640 khash on Core i7-4790K CPU @ 4.40GHz and with your miner I got as close to 600 khash...
Just my 2 cents.
Thanks for the feedback, I'll look into it.
Edit: No rejects for me with default options at yiimp. I also reviewed the code and found no changes that would affect zr5.
Can you confirm that 3.1.5 & 3.1.6 behave differently? Could you also try yiimp?
When I mine the diff is reported as .013. You could try setting the diff manually. On yaamp clones it is in the password
string ( -p d=n ), but your pool may be different.
Edit2: I took a look at the scmorse implementation and i might be able to do better. His implemetation does not use the
aes-ni optimized groestl. If I can integrate that it should hash even faster.
Merging my optimizations with scmorse gave a 33% increase to zr5 but I'll wait to release it until your issue is understood.
March 27, 2016, 08:36:52 AM
The new version 3.1.6 returns only invalid (low-diff) shares with zr5 algo on ziftrpool.io
AKAIK v3.1.5 worked just fine with that pool.
There is some more room for improvement, especially on zr5.
With scmorse/ziftr-cpu which is ig0tik3d/ziftr-cpu with stratum,
I get about 640 khash on Core i7-4790K CPU @ 4.40GHz and with your miner I got as close to 600 khash...
Just my 2 cents.
AKAIK v3.1.5 worked just fine with that pool.
There is some more room for improvement, especially on zr5.
With scmorse/ziftr-cpu which is ig0tik3d/ziftr-cpu with stratum,
I get about 640 khash on Core i7-4790K CPU @ 4.40GHz and with your miner I got as close to 600 khash...
Just my 2 cents.
Thanks for the feedback, I'll look into it.
Edit: No rejects for me with default options at yiimp. I also reviewed the code and found no changes that would affect zr5.
Can you confirm that 3.1.5 & 3.1.6 behave differently? Could you also try yiimp?
When I mine the diff is reported as .013. You could try setting the diff manually. On yaamp clones it is in the password
string ( -p d=n ), but your pool may be different.
Edit2: I took a look at the scmorse implementation and i might be able to do better. His implemetation does not use the
aes-ni optimized groestl. If I can integrate that it should hash even faster.
Jump to: