I have a request for some focussed testing of cpuminer.
I would like some confirmation that the cpu capabilities checking is correct with no false positives
or false negatives.
I need someone with older cpus to test cpuminer on an algo with multiple supported kernels
looking for two things:
1. cpuminer selected a slower kernel when it should have chosen a faster one. You would observe
this at startup.
2. cpuminer selected a faster kernel on a cpu that doesn't support it. This will likely cause
cpuminer to crash.
Thanks
Topic: [ANN]: cpuminer-opt v3.8.8.1, open source optimized multi-algo CPU miner - page 194. (Read 444067 times)
January 18, 2016, 11:32:42 AM
January 18, 2016, 11:08:48 AM
Yesterday was pretty much a wasted day. Ispent all day trying to integrate cpuminer-neoscrypt
and now I'm back at square 1.
It has a two dimensional matrix of capabilities: 2 arch, and asm or not. And he coded it such
that it is almost impossible to compile more than one variation at a time. eEverithing is hooked
with #If.
I came to the conclusion this is going to require some significant reimplementation. I've decided
to only focus on the highest performing variation for the first release. I'l work out the rest at a
later date.
I will take this approach for cryptonight and argon2 and any other new algo additions in the future.
I got a little cocky with the ease in which I implemented the x11 and quark improvements.
Lesson learned.
and now I'm back at square 1.
It has a two dimensional matrix of capabilities: 2 arch, and asm or not. And he coded it such
that it is almost impossible to compile more than one variation at a time. eEverithing is hooked
with #If.
I came to the conclusion this is going to require some significant reimplementation. I've decided
to only focus on the highest performing variation for the first release. I'l work out the rest at a
later date.
I will take this approach for cryptonight and argon2 and any other new algo additions in the future.
I got a little cocky with the ease in which I implemented the x11 and quark improvements.
Lesson learned.
January 17, 2016, 09:24:02 PM
Things are progressing well with the integration of the faster neoscrypt kernel. It's almost ready
to compile, just cleaning up the declarations and definitions to avoid collisions.
I noticed that the hash functions are all declared in miner.h. Is that necessary? Isn't the scanhash
function the gate to the algos? And more importantly could this be at the root of my multiple
def problems with GRS on linux and whatever else on windows? All these macros get imported
to the hash function and the declaration in miner.h makes them visible to the world.
Wolf, do you think I'm on the right track?
to compile, just cleaning up the declarations and definitions to avoid collisions.
I noticed that the hash functions are all declared in miner.h. Is that necessary? Isn't the scanhash
function the gate to the algos? And more importantly could this be at the root of my multiple
def problems with GRS on linux and whatever else on windows? All these macros get imported
to the hash function and the declaration in miner.h makes them visible to the world.
Wolf, do you think I'm on the right track?
Ice cold, I believe. I haven't checked in depth, but I'm pretty sure your issue is two seperate implementations of the same shit.
January 18, 2016, 10:20:02 AM
cpuSminer S=speed 
or cpuminer-s
or cpuminer-sThanks for posting your results.
Strange is I get many more shares at the other site vs NiceHash?
Wonder why that is? or I should say they seem to come in at a much quicker pace.
Wondering if I can clone the VM and move it to another rig, should be possible I would think.
If there is to much different in the rigs the kernel may crash of the OS. Might set up another
VM on this other laptop for testing that. But then again the virtual HDD is 50GB and my 64GB
USB drive is full with other saved stuff, I'll try transferring it across the network lan.
Edit: Got to clean up my HDD's on this other laptop first didn't realize they were so full
240GB SSD & 500GB IDE both almost full. lol Between Wallets, miners & games it doesn't take long.
Not counting in the many .iso of OS I have downloaded to try out.
The share rate is determined by the difficulty. Pools adjust the difficulty based on the hash rate
to prevent powerfull miners from flooding the server frequent small shares. With very low hash
rates the difficulty can reach the minimum while still not achieving the target share submission rate.
Edit: check the hashrate at the pool. I've noticed that the infrequent share submission results in the
pool reported hash rate fluctuates from -50% to 200% of cpuminer's reported hash rate, which tends
to average out at 100%. If the pool reported hash rate is consistent below this range then there may
be a problem.
A key thing to remember is that the miner reports the share rate not the share size. So the shares
are less frequent but larger, everything else being equal.
Hope this helps.
January 18, 2016, 10:18:43 AM
we can see that when using AVX... the overheat reduce the speed on recent cpus, just like the nvidia 9xx series
some bios allow to reduce or tune the FIVR protection system
some bios allow to reduce or tune the FIVR protection system
January 18, 2016, 10:01:25 AM
Got an accepted share on x11.
But upping the priority didn't affect the progressive slowdown. Hope it's not a hardware problem...
But upping the priority didn't affect the progressive slowdown. Hope it's not a hardware problem...
January 18, 2016, 03:36:55 AM
cpuSminer S=speed or cpuminer-s
or cpuminer-sThanks for posting your results.
Strange is I get many more shares at the other site vs NiceHash?
Wonder why that is? or I should say they seem to come in at a much quicker pace.
Wondering if I can clone the VM and move it to another rig, should be possible I would think.
If there is to much different in the rigs the kernel may crash of the OS. Might set up another
VM on this other laptop for testing that. But then again the virtual HDD is 50GB and my 64GB
USB drive is full with other saved stuff, I'll try transferring it across the network lan.
Edit: Got to clean up my HDD's on this other laptop first didn't realize they were so full
240GB SSD & 500GB IDE both almost full. lol Between Wallets, miners & games it doesn't take long.
Not counting in the many .iso of OS I have downloaded to try out.
January 18, 2016, 03:08:39 AM
cpuSminer S=speed or cpuminer-s
or cpuminer-sThanks for posting your results.
January 18, 2016, 01:28:58 AM
cpuminer-opt also
I like it. I don't have a big ego so I don't need my name in it.I'll look into cpuminer-opt being taken, I was going to go with it..
How about Fast CPU MIner as a brand and keep the cpuminer command name?
It's hard to call it fast compared with GPUs but it's the fastest CPU miner available
(that is a challenge to anyone with anything better) and getting faster.
JMINER---
Or something simple. --scryptr
yup - im all for simple to ...
jbm ... i like it
...#crysx
It's taken, I also like the idea of paying homage to it's original name.
or cpuminer-s
January 17, 2016, 10:08:35 PM
THE FROZEN CPU MINER--
Sorry, I just watched Ted 2 on DVD... and remember an old SNL skit. JBM is perhaps too simple... --scryptr
Phew, I was thinking of another movie with a frozen theme intended for a young audience.
I googled cpuminer-opt and the only thing it found was this discussion. So it's all set.
Thanks to Tanguy for suggesting it.
January 17, 2016, 10:01:45 PM
Things are progressing well with the integration of the faster neoscrypt kernel. It's almost ready
to compile, just cleaning up the declarations and definitions to avoid collisions.
I noticed that the hash functions are all declared in miner.h. Is that necessary? Isn't the scanhash
function the gate to the algos? And more importantly could this be at the root of my multiple
def problems with GRS on linux and whatever else on windows? All these macros get imported
to the hash function and the declaration in miner.h makes them visible to the world.
Wolf, do you think I'm on the right track?
to compile, just cleaning up the declarations and definitions to avoid collisions.
I noticed that the hash functions are all declared in miner.h. Is that necessary? Isn't the scanhash
function the gate to the algos? And more importantly could this be at the root of my multiple
def problems with GRS on linux and whatever else on windows? All these macros get imported
to the hash function and the declaration in miner.h makes them visible to the world.
Wolf, do you think I'm on the right track?
Ice cold, I believe. I haven't checked in depth, but I'm pretty sure your issue is two seperate implementations of the same shit.
There in fact two seperate implementations of the same shit, all the spf stuff. I may have selected one version in one algo
and another version in another. Hmmm, something else to follow up on. I planned on cleaning that up any way, I think
the versions are identical. I didn't want to rush that as there may be a better version.
I still think there is merit to my theory when I think about it more. The macros are all run by the hash function, and being
macros have global scope in the file. Declaring the hash functions miner.h exports it to the world where the linker finds
it and the 5 copies of each macro.
On the other hand it doesn't happen with the other macro based algos so what do I know?
January 17, 2016, 09:00:11 PM
cpuminer-opt also
I like it. I don't have a big ego so I don't need my name in it.I'll look into cpuminer-opt being taken, I was going to go with it..
How about Fast CPU MIner as a brand and keep the cpuminer command name?
It's hard to call it fast compared with GPUs but it's the fastest CPU miner available
(that is a challenge to anyone with anything better) and getting faster.
JMINER---
Or something simple. --scryptr
yup - im all for simple to ...
jbm ... i like it
...#crysx
It's taken, I also like the idea of paying homage to it's original name.
THE FROZEN CPU MINER--
Sorry, I just watched Ted 2 on DVD... and remember an old SNL skit. JBM is perhaps too simple... --scryptr
January 17, 2016, 08:49:11 PM
cpuminer-opt also
I like it. I don't have a big ego so I don't need my name in it.I'll look into cpuminer-opt being taken, I was going to go with it..
How about Fast CPU MIner as a brand and keep the cpuminer command name?
It's hard to call it fast compared with GPUs but it's the fastest CPU miner available
(that is a challenge to anyone with anything better) and getting faster.
JMINER---
Or something simple. --scryptr
yup - im all for simple to ...
jbm ... i like it
...#crysx
It's taken, I also like the idea of paying homage to it's original name.
January 17, 2016, 08:11:58 PM
cpuminer-opt also
I like it. I don't have a big ego so I don't need my name in it.I'll look into cpuminer-opt being taken, I was going to go with it..
How about Fast CPU MIner as a brand and keep the cpuminer command name?
It's hard to call it fast compared with GPUs but it's the fastest CPU miner available
(that is a challenge to anyone with anything better) and getting faster.
JMINER---
Or something simple. --scryptr
yup - im all for simple to ...
jbm ... i like it
...#crysx
January 17, 2016, 08:10:54 PM
cpuminer-opt also
I like it. I don't have a big ego so I don't need my name in it.I'll look into cpuminer-opt being taken, I was going to go with it..
How about Fast CPU MIner as a brand and keep the cpuminer command name?
It's hard to call it fast compared with GPUs but it's the fastest CPU miner available
(that is a challenge to anyone with anything better) and getting faster.
JMINER---
Or something simple. --scryptr
January 17, 2016, 06:51:08 PM
cpuminer-opt also
I like it. I don't have a big ego so I don't need my name in it.I'll look into cpuminer-opt being taken, I was going to go with it..
How about Fast CPU MIner as a brand and keep the cpuminer command name?
It's hard to call it fast compared with GPUs but it's the fastest CPU miner available
(that is a challenge to anyone with anything better) and getting faster.
January 17, 2016, 05:06:36 PM
Things are progressing well with the integration of the faster neoscrypt kernel. It's almost ready
to compile, just cleaning up the declarations and definitions to avoid collisions.
I noticed that the hash functions are all declared in miner.h. Is that necessary? Isn't the scanhash
function the gate to the algos? And more importantly could this be at the root of my multiple
def problems with GRS on linux and whatever else on windows? All these macros get imported
to the hash function and the declaration in miner.h makes them visible to the world.
Wolf, do you think I'm on the right track?
to compile, just cleaning up the declarations and definitions to avoid collisions.
I noticed that the hash functions are all declared in miner.h. Is that necessary? Isn't the scanhash
function the gate to the algos? And more importantly could this be at the root of my multiple
def problems with GRS on linux and whatever else on windows? All these macros get imported
to the hash function and the declaration in miner.h makes them visible to the world.
Wolf, do you think I'm on the right track?
January 17, 2016, 05:01:10 PM
Fx-8320
Aes-mi enabled
Starts at about 500 Kh/s then drops to about 420.
Never submits any share :-/
Aes-mi enabled
Starts at about 500 Kh/s then drops to about 420.
Never submits any share :-/
What algo? Some take a while for the first share. I tested all supported configurations before
release and it submitted shares for all of them.
The drop in hash rate is a bit concerning though. No other major cpu activity?
Sorry, X11. Drop may be cause of other activity (will fix with priority) or some sort of throttling.
Will test more tomorrow.
No problem and thanks for testing.
I've taken a quick look at the faster neoscrypt cpu kernel (the term doesn't really apply to cpus but
whatever) and it seems to have a remarkable resemblence to the ccminer kernel (proper usage of
the word). It may be worth taking a look, it might give some ideas for improving ccminer.
January 17, 2016, 04:48:40 PM
Fx-8320
Aes-mi enabled
Starts at about 500 Kh/s then drops to about 420.
Never submits any share :-/
Aes-mi enabled
Starts at about 500 Kh/s then drops to about 420.
Never submits any share :-/
What algo? Some take a while for the first share. I tested all supported configurations before
release and it submitted shares for all of them.
The drop in hash rate is a bit concerning though. No other major cpu activity?
Sorry, X11. Drop may be cause of other activity (will fix with priority) or some sort of throttling.
Will test more tomorrow.
January 17, 2016, 04:19:14 PM
Fx-8320
Aes-mi enabled
Starts at about 500 Kh/s then drops to about 420.
Never submits any share :-/
Aes-mi enabled
Starts at about 500 Kh/s then drops to about 420.
Never submits any share :-/
What algo? Some take a while for the first share. I tested all supported configurations before
release and it submitted shares for all of them.
The drop in hash rate is a bit concerning though. No other major cpu activity?
[2016-01-17 15:16:07] Starting Stratum on stratum+tcp://x11.usa.nicehash.com:3336
[2016-01-17 15:16:07] 4 miner threads started, using 'x11' algorithm.
[2016-01-17 15:16:07] Stratum authentication failed
[2016-01-17 15:16:07] ...retry after 10 seconds
[2016-01-17 15:16:17] Stratum authentication failed
[2016-01-17 15:16:17] ...retry after 10 seconds
[2016-01-17 15:16:28] Stratum authentication failed
Edit: got it going something was wrong with my bitcoin address.
Jump to: