Topic: [XPM] [ANN] Primecoin High Performance | HP14 released! - page 115. (Read 397647 times)
Optimal sievesize might be 1000000, some people discussed on this topic.
New version is out. It mainly adds a tuning parameter called -sievesize. Use it like this: primecoind -sievesize=2000000
@mikaelh to make sure ... the change of sievesize is not what Sunny was talking about in https://bitcointalksearch.org/topic/m.2716964 , right?
I have reduced sievesize from 2.5M to 50000 in several steps and the PSS indeed rises (not linearly). But I hope it's not just an illusion.
Sieve size isn't directly what Sunny was talking about, but it is related. If you lower the amount of work you perform or the shorten the time you spend working on it before giving up, then you will throw away a lot of potential solutions unnecessarily.
Lowering sieve size to 50,000 will raise PPS but it is an illusion.
New version is out. It mainly adds a tuning parameter called -sievesize. Use it like this: primecoind -sievesize=2000000
@mikaelh to make sure ... the change of sievesize is not what Sunny was talking about in https://bitcointalksearch.org/topic/m.2716964 , right?
I have reduced sievesize from 2.5M to 50000 in several steps and the PSS indeed rises (not linearly). But I hope it's not just an illusion.
update
hp3 falls over alot
hp3 falls over alot
What do you mean falls over?
I'm having the same problem on my two computers; It throws random errors and quits every few hours. I Have to manually start the program again and start mining.
update
hp3 falls over alot
hp3 falls over alot
What do you mean falls over?
update
hp3 falls over alot
hp3 falls over alot
I heard about the a "hp3" build? Is this in the git repo for mikaelh2?
yes: https://github.com/mikaelh2/primecoin
thx one other dumb Q. Do I have to do anything to enable hp3 or is it built by default?
I got an exception running it with maybe 12 threads on 4 core i7 2660. Sorry, removed the error message and restarted the primecoin-qt. Will report details if it happens again.
I heard about the a "hp3" build? Is this in the git repo for mikaelh2?
yes: https://github.com/mikaelh2/primecoin
I'm getting about the same rate but my CPU is way down like less than 20% most of the time. Anyway to allow it to use full CPU. Win 8 by the way.
In the console of the primecoin-qt you can enter
Code:
setgenerate true -1
That allows all cores to be used. Do you have a "primecoin.conf" file on that machine?
I heard about the a "hp3" build? Is this in the git repo for mikaelh2?
I'm getting about the same rate but my CPU is way down like less than 20% most of the time. Anyway to allow it to use full CPU. Win 8 by the way.
does any one have a script to start
primcoind --deamon
if it falls over?
if(check/5 mins if on watch process!)
{primcoind --deamon;}
primcoind --deamon
if it falls over?
if(check/5 mins if on watch process!)
{primcoind --deamon;}
I'm getting pretty abysmal performance with this build on osx (using the osx makefile). 1player's build seems to be working better on osx while all other machines are better using mikhael's hp3. Any suggestions / optimizations to the osx makefile would be much appreciated.
I'm tempted to VM an ubuntu build and see how that goes but I'd rather run it natively in osx with this high performance build.
I'm tempted to VM an ubuntu build and see how that goes but I'd rather run it natively in osx with this high performance build.
That's the route I took - VirtualBox/ubuntu - been running 4 of my cores at about 3500pps with Mikaelh2 build. Found 3 blocks but nothing in the last 24 hours
Well, to be honest his build is probably the best, but I don't have the guts to run a binary made by a new guy at bitcointalk.
I'm getting pretty abysmal performance with this build on osx (using the osx makefile). 1player's build seems to be working better on osx while all other machines are better using mikhael's hp3. Any suggestions / optimizations to the osx makefile would be much appreciated.
I'm tempted to VM an ubuntu build and see how that goes but I'd rather run it natively in osx with this high performance build.
I'm tempted to VM an ubuntu build and see how that goes but I'd rather run it natively in osx with this high performance build.
That's the route I took - VirtualBox/ubuntu - been running 4 of my cores at about 3500pps with Mikaelh2 build. Found 3 blocks but nothing in the last 24 hours
Finally managed to build it on Ubuntu 13.04 amd64, it gives me around 4500 pps on my FX 8350 
Congrats!
Next project; improve HP3???
In my case it's weird, if I compile with march=native the pps rate is lower.
Remember, PPS isn't an accurate measurement.
http://www.reddit.com/r/primecoin/comments/1iaak8/psa_primes_per_second_just_dropped_carry_on/
After all, I think that the performance is lower with -march=native, it could be related to my AMD processor fake 8 cores. I could still give a try to march=native to try.
Should the sievesize be a prime number?
Finally managed to build it on Ubuntu 13.04 amd64, it gives me around 4500 pps on my FX 8350
Congrats!
Next project; improve HP3???
In my case it's weird, if I compile with march=native the pps rate is lower.
Remember, PPS isn't an accurate measurement.
http://www.reddit.com/r/primecoin/comments/1iaak8/psa_primes_per_second_just_dropped_carry_on/
I was poking through prime.h and prime.cpp and they're not really all that long. Someone who knows c++, OpenCL and / or CUDA should be able to make the conversion. I would be willing to bet it has already happened and they're raking it in at the moment.
I found that installing libmpfr-dev which it required for Freicoin, also satisfies the gmp requirement.
Is there a difference in speed between the mpfr and gmp libraries?
Is there a difference in speed between the mpfr and gmp libraries?
libmpfr-dev pulls libgmp as a dependency. The actual mpfr does floating point calculations. Comparing mpfr and gmp is like comparing apples and bananas.
Thank you.
I did not know that it used gmp as a dependency.
I see so many libraries and packages that appear to do the same things. Some better at certain applications while worse at others.
I guess the next major step would be to rewrite the mining routine in OpenCL. Though I would not doubt somebody has already did it and is not sharing.
Can't say I would blame them at this point.
Yeah, I was thinking that earlier. It seems a likely explanation to the skyrocketing difficulty (that surprises even Sunny).
I'm curious - how / why were OCL miners even publicly released anyway? What caused the person / group to release it/them and not horde anymore?
Remember that having too much control of a decentralized network by a single person reduces the trust that the public has on the network, that is bad for the guy hoarding the miner. Best is to mine a nice time with the OpenCL miner then when you have your stash just release the miner to the public to make your stash worth more.
Jump to: