I presume we gona see a lot more development in future, looks like there is always something new to add and optimize even just for CPU
Topic: [XPM] [ANN] Primecoin High Performance | HP14 released! - page 34. (Read 397645 times)
gr8 work as usual !
I presume we gona see a lot more development in future, looks like there is always something new to add and optimize even just for CPU
I presume we gona see a lot more development in future, looks like there is always something new to add and optimize even just for CPU
-hp10 released! Downloads are on the first page as usual.
Changes in -hp10:
* Major rewrite of the sieve to support extending the sieve (originally implemented by jh000).
* Added a new parameter 'sieveextensions' which controls how many times the sieve is extended.
* Removed the problematic and unnecessary fast-division test.
* Changed the value of nL1CacheElements to 224000 which is slightly faster (thanks to nushor on IRC).
* The 'roundsievepercentage' parameter is gone. The round primorial is now automatically adjusted.
* New RPC command 'listtopprimes' by Sunny.
Changes in -hp10:
* Major rewrite of the sieve to support extending the sieve (originally implemented by jh000).
* Added a new parameter 'sieveextensions' which controls how many times the sieve is extended.
* Removed the problematic and unnecessary fast-division test.
* Changed the value of nL1CacheElements to 224000 which is slightly faster (thanks to nushor on IRC).
* The 'roundsievepercentage' parameter is gone. The round primorial is now automatically adjusted.
* New RPC command 'listtopprimes' by Sunny.
Thanks for the update!
-hp10 released! Downloads are on the first page as usual.
Changes in -hp10:
* Major rewrite of the sieve to support extending the sieve (originally implemented by jh000).
* Added a new parameter 'sieveextensions' which controls how many times the sieve is extended.
* Removed the problematic and unnecessary fast-division test.
* Changed the value of nL1CacheElements to 224000 which is slightly faster (thanks to nushor on IRC).
* The 'roundsievepercentage' parameter is gone. The round primorial is now automatically adjusted.
* New RPC command 'listtopprimes' by Sunny.
Changes in -hp10:
* Major rewrite of the sieve to support extending the sieve (originally implemented by jh000).
* Added a new parameter 'sieveextensions' which controls how many times the sieve is extended.
* Removed the problematic and unnecessary fast-division test.
* Changed the value of nL1CacheElements to 224000 which is slightly faster (thanks to nushor on IRC).
* The 'roundsievepercentage' parameter is gone. The round primorial is now automatically adjusted.
* New RPC command 'listtopprimes' by Sunny.
blockchain download is VERY slow right now, my guess is that a lot of people just compiled and we are all downloading at the same time.
I'm sure there is a way to keep the already downloaded blockchain with hp9, but I wasn't sure so started fresh after compiling.
edit :
Here is what I get with HP10 on one server :
CPD went from 1.31 to 2.75
PPS went from 2k to 2.7k
Next is tweaking sieveextensions...
I'm sure there is a way to keep the already downloaded blockchain with hp9, but I wasn't sure so started fresh after compiling.
edit :
Here is what I get with HP10 on one server :
CPD went from 1.31 to 2.75
PPS went from 2k to 2.7k
Next is tweaking sieveextensions...
Running in vmware
Getting
"chainspermin" : 10
"chainsperday" : 1.14342394
"primespersec" : 1157
With
"genproclimit" : 4
How do these numbers look to you?
Getting
"chainspermin" : 10
"chainsperday" : 1.14342394
"primespersec" : 1157
With
"genproclimit" : 4
How do these numbers look to you?
Yeah, I'm giving the man the time.. More RAM because I will still do VM linux inside Windows. Any preferences for the good VM?
It will come, give the man some time.
Why more ram ?
Why more ram ?
I second that request for the compiled Windows binary... Or I'll have buy more RAM and go linux!
What, more RAM for Linux??
I pushed some big optimizations for -hp10 to github so that more people can test them.
Any chance of getting a compiled Windows binary? I'd like to test on a 32 core Opteron rig and a 3930k.
I second that request for the compiled Windows binary... Or I'll have buy more RAM and go linux!
ok I got the sources, thx !
I'm compiling it the same way that you showed on your Linux Compilation Guide. Is this the right way to do it ?
I'm compiling it the same way that you showed on your Linux Compilation Guide. Is this the right way to do it ?
It should still work just fine. You just need to change in which directory you compile.
I deleted everything on one server just to be sure and it seems it worked fine.
Need to download the whole blockchain again but that is a minor pain.
getmininginfo shows the new "sieveextensions" : 6, can't wait to see the change
Will just transfering the new primecoind to the vps that can't compile work ?
Does anyone have optimized compilation flags for the AMD bulldozer architecture? This processor is severely under-performing, worse than CPUs years older!
PS: I can confirm hp10 is faster
PS: I can confirm hp10 is faster
ok I got the sources, thx !
I'm compiling it the same way that you showed on your Linux Compilation Guide. Is this the right way to do it ?
I'm compiling it the same way that you showed on your Linux Compilation Guide. Is this the right way to do it ?
It should still work just fine. You just need to change in which directory you compile.
ok I got the sources, thx !
I'm compiling it the same way that you showed on your Linux Compilation Guide. Is this the right way to do it ?
I'm compiling it the same way that you showed on your Linux Compilation Guide. Is this the right way to do it ?
holy crap hp10 is 2x as fast as hp9
wow, are the sources available ? I'm not used to git :/
You can download a ZIP from github if you want.
Do you compile with make -f makefile.unix USE_UPNP=- or is there some optimization flags to set? A bunch of other miners and posts mention msse2 -O3 and some others I think.
Compiling the client with -O3, -msse2 or -march=native is not going to make a big difference. The sieve currently doesn't benefit from SSE2 or any other special instruction set. The default -O2 optimization level is more than enough.
GMP does benefit from compiler flags. The "configure" script that comes with it tries to detect the best settings if you compile it yourself.
holy crap hp10 is 2x as fast as hp9
wow, are the sources available ? I'm not used to git :/
Do you compile with make -f makefile.unix USE_UPNP=- or is there some optimization flags to set? A bunch of other miners and posts mention msse2 -O3 and some others I think.
holy crap hp10 is 2x as fast as hp9
"Roundsievepercentage" doesn't show on the mininginfo anymore.
It's replaced by "sieveextensions" (default value : 6 )
Any info on that?
It's replaced by "sieveextensions" (default value : 6 )
Any info on that?
"Roundsievepercentage" is gone. The round primorial is now adjusted automatically.
"Sieveextensions" is a new parameter which controls the new big optimization. jh00 discovered that the sieve can be "extended" efficiently to numbers that are two times bigger. That parameter determines how many times the sieve is extended (each further extension will produce numbers two times bigger than the previous extension). Bigger numbers will have a lower probability of hitting primes and prime chains but each extension is pretty cheap. Six extensions seemed like the optimal value for mainnet.
"Roundsievepercentage" doesn't show on the mininginfo anymore.
It's replaced by "sieveextensions" (default value : 6 )
Any info on that?
It's replaced by "sieveextensions" (default value : 6 )
Any info on that?
Jump to: