[XPM] [ANN] Primecoin High Performance | HP14 released!

[gigawatt]

Here's something I came up with while benchmarking some parameter sets.

Before you begin, be sure to delete your debug.log file so you don't have any previous info in it.
Run primecoin with your desired settings and have it run for a few hours.  (Anything less than 2hrs and I can't guarantee accuracy)
In the directory with your debug.log file, run

Code:

grep primemeter debug.log | awk '{s1+=$4;s2+=$6;s3+=$8} END {printf"%10s: %10d\n%10s: %10d\n%10s: %10d\n","Prime/h",s1/NR,"Test/h",s2/NR,"5-Chain/h",s3/NR}'

You'll need to 1) be on a *nix system or 2) have cygwin installed.
This averages all the stats produced in debug.log.  Since the stats are of 60 second intervals, no one data point is relevant, but a few hundred of them are.

Example output:

Code:

   Prime/h:    8138437
    Test/h:   72084650
 5-Chain/h:        330

when you say "run in the directory of your debug.lof file, do you mean from use run>cmd>blahblahdirectoryofdebugYOURCODE>enter?

Do I delete the debug.log file and just add a new.txt and rename it? after each test can I just rename the debug.log file to "test1.txt" and then create a new one? Ill be happy to post all results. Sorry im n00bish

Before you start that, you're going to need to install Cygwin because Windows doesn't have grep or awk.  You'd launch a Cygwin terminal, cd to your primecoin AppData directory, then run the command.  If you want to empty your debug.log file, you can use echo "" > debug.log.

[1695512790]

1695512790

[1695512790]

1695512790

[zax983]

Today I will try:

sievesize=200000
genproclimit=-1
roundsievepercentage=100
sievepercentage=1


00:29:08

getmininginfo


00:29:08

{
"blocks" : 86080,
"chainspermin" : 0,
"currentblocksize" : 1000,
"currentblocktx" : 0,
"difficulty" : 9.28916806,
"errors" : "",
"generate" : true,
"genproclimit" : -1,
"primespersec" : 2686,
"pooledtx" : 0,
"sievepercentage" : 1,
"sievesize" : 200000,
"testnet" : false
}


00:31:23

getmininginfo


00:31:23

{
"blocks" : 86081,
"chainspermin" : 0,
"currentblocksize" : 1000,
"currentblocktx" : 0,
"difficulty" : 9.28909522,
"errors" : "",
"generate" : true,
"genproclimit" : -1,
"primespersec" : 2749,
"pooledtx" : 0,
"sievepercentage" : 1,
"sievesize" : 200000,
"testnet" : false
}


00:32:10

getmininginfo


00:32:10

{
"blocks" : 86081,
"chainspermin" : 0,
"currentblocksize" : 1000,
"currentblocktx" : 0,
"difficulty" : 9.28909522,
"errors" : "",
"generate" : true,
"genproclimit" : -1,
"primespersec" : 2180,
"pooledtx" : 0,
"sievepercentage" : 1,
"sievesize" : 200000,
"testnet" : false
}

Always with any settings so big up and down, why?!

[bitrich]

Here's something I came up with while benchmarking some parameter sets.

Before you begin, be sure to delete your debug.log file so you don't have any previous info in it.
Run primecoin with your desired settings and have it run for a few hours.  (Anything less than 2hrs and I can't guarantee accuracy)
In the directory with your debug.log file, run

Code:

grep primemeter debug.log | awk '{s1+=$4;s2+=$6;s3+=$8} END {printf"%10s: %10d\n%10s: %10d\n%10s: %10d\n","Prime/h",s1/NR,"Test/h",s2/NR,"5-Chain/h",s3/NR}'

You'll need to 1) be on a *nix system or 2) have cygwin installed.
This averages all the stats produced in debug.log.  Since the stats are of 60 second intervals, no one data point is relevant, but a few hundred of them are.

Example output:

Code:

   Prime/h:    8138437
    Test/h:   72084650
 5-Chain/h:        330

when you say "run in the directory of your debug.lof file, do you mean from use run>cmd>blahblahdirectoryofdebugYOURCODE>enter?

Do I delete the debug.log file and just add a new.txt and rename it? after each test can I just rename the debug.log file to "test1.txt" and then create a new one? Ill be happy to post all results. Sorry im n00bish

Before you start that, you're going to need to install Cygwin because Windows doesn't have grep or awk.  You'd launch a Cygwin terminal, cd to your primecoin AppData directory, then run the command.  If you want to empty your debug.log file, you can use echo "" > debug.log.

ok ill see what i can do, ill post results periodically. Im running an i5 2.3ghz 2410M with windows 7

[bitrich]

Here's something I came up with while benchmarking some parameter sets.

Before you begin, be sure to delete your debug.log file so you don't have any previous info in it.
Run primecoin with your desired settings and have it run for a few hours.  (Anything less than 2hrs and I can't guarantee accuracy)
In the directory with your debug.log file, run

Code:

grep primemeter debug.log | awk '{s1+=$4;s2+=$6;s3+=$8} END {printf"%10s: %10d\n%10s: %10d\n%10s: %10d\n","Prime/h",s1/NR,"Test/h",s2/NR,"5-Chain/h",s3/NR}'

You'll need to 1) be on a *nix system or 2) have cygwin installed.
This averages all the stats produced in debug.log.  Since the stats are of 60 second intervals, no one data point is relevant, but a few hundred of them are.

Example output:

Code:

   Prime/h:    8138437
    Test/h:   72084650
 5-Chain/h:        330

when you say "run in the directory of your debug.lof file, do you mean from use run>cmd>blahblahdirectoryofdebugYOURCODE>enter?

Do I delete the debug.log file and just add a new.txt and rename it? after each test can I just rename the debug.log file to "test1.txt" and then create a new one? Ill be happy to post all results. Sorry im n00bish

Before you start that, you're going to need to install Cygwin because Windows doesn't have grep or awk.  You'd launch a Cygwin terminal, cd to your primecoin AppData directory, then run the command.  If you want to empty your debug.log file, you can use echo "" > debug.log.

ok ill see what i can do, ill post results periodically. Im running an i5 2.3ghz 2410M with windows 7

ok i got cygwin installed, launched a cygwin terminal. how do I "cd to my primecoin directory"? I dont know what that means. Sorry

[gigawatt]

ok i got cygwin installed, launched a cygwin terminal. how do I "cd to my primecoin directory"? I dont know what that means. Sorry

cd is the command for "change directory".
Cygwin uses *nix style directory paths, so your primecoin directory should be something like:
cd "/cygdrive/c/Users/your_username/AppData/Roaming/Primecoin"

[bitrich]

I got it figured out. Thanks will post some results by tomorrow

[bitrich]

Ok started with these settings and ran for 2 hours

roundsievepercentage=10
sievepercentage=1
sievesize=100000
pps=around 2800
cpm=1-2

 Prime/h:    9799228
    Test/h:  126765132
 5-Chain/h:         94

Yes I know these are low settings but figured I would start from there and work my way up.

[Tommo_Aus]

Hi guys,

I've made the switch to the HP8 client and noted a large increase in p/s but I can't find an answer to a few basics of Primecoin.

To judge performace am I only interested in p/sec or does chainspermin matter as well?

What is sievesize? chainspermin and p/s change if I vary the sievesize but what does this actually mean? I've seen some wanting to go for the largest size possible such as 4 000 000 but my p/s goes down a lot if I try this. On my i7 3820 if I use the lowest sievesize, 100 000, I get the best results.

2 000 000:

    "blocks" : 83331,
    "chainspermin" : 8,
    "currentblocksize" : 1820,
    "currentblocktx" : 1,
    "difficulty" : 9.26782423,
    "errors" : "",
    "generate" : true,
    "genproclimit" : 7,
    "roundsievepercentage" : 30,
    "primespersec" : 3866,
    "pooledtx" : 1,
    "sievepercentage" : 10,
    "sievesize" : 2000000,
    "testnet" : false

100 000:

    "blocks" : 83318,
    "chainspermin" : 11,
    "currentblocksize" : 1000,
    "currentblocktx" : 0,
    "difficulty" : 9.26746053,
    "errors" : "",
    "generate" : true,
    "genproclimit" : 7,
    "roundsievepercentage" : 30,
    "primespersec" : 7495,
    "pooledtx" : 0,
    "sievepercentage" : 10,
    "sievesize" : 100000,
    "testnet" : false

For those interested there are my results on different hardware:

2x Xeon E5520 (4 cores/8 threads each, running 14 threads):

{
    "blocks" : 86410,
    "chainspermin" : 10,
    "currentblocksize" : 1000,
    "currentblocktx" : 0,
    "difficulty" : 9.28923029,
    "errors" : "",
    "generate" : true,
    "genproclimit" : 14,
    "roundsievepercentage" : 20,
    "primespersec" : 8445,
    "pooledtx" : 0,
    "sievepercentage" : 8,
    "sievesize" : 100000,
    "testnet" : false
}

2x Opteron 6128 HE (8 cores each, running 14 threads):

{
    "blocks" : 86414,
    "chainspermin" : 15,
    "currentblocksize" : 1000,
    "currentblocktx" : 0,
    "difficulty" : 9.28927737,
    "errors" : "",
    "generate" : true,
    "genproclimit" : 14,
    "roundsievepercentage" : 20,
    "primespersec" : 13677,
    "pooledtx" : 0,
    "sievepercentage" : 8,
    "sievesize" : 100000,
    "testnet" : false
}

i7 3820 (4 cores/8 threads, running 7 threads):

{
    "blocks" : 86415,
    "chainspermin" : 10,
    "currentblocksize" : 1000,
    "currentblocktx" : 0,
    "difficulty" : 9.28940433,
    "errors" : "",
    "generate" : true,
    "genproclimit" : 7,
    "roundsievepercentage" : 20,
    "primespersec" : 8297,
    "pooledtx" : 0,
    "sievepercentage" : 8,
    "sievesize" : 100000,
    "testnet" : false
}

For some reason it really likes the Opteron's, even though they're lower powered than the Xeon's in normal applications.

On last bit, I've found one block over the last couple of days and its displaying as immature, 2591 confirms. When does it mature so I can send it to another wallet? Is it at a certain number of confirms?

[Dsfyu]

On last bit, I've found one block over the last couple of days and its displaying as immature, 2591 confirms. When does it mature so I can send it to another wallet? Is it at a certain number of confirms?

If you go to transactions and hover your mouse over the dial there it should tell you how many more before it matures. If I remember correctly it's something like 3200 or something in that range.

[Trillium]

Yes a single Opteron dual core @ 2.8 will easily outperform 2x Xeon DP (single core, HT) @ 3.6 GHz. By a factor of 2-3x. AMD CPU do more work per cycle than intel designs. I get 1600+ and 600-700 PPS respectively from these. The memory and memory subsystem is slightly faster on the AMD too, but not really by much. Both are 64 bit.

As to Tommo_Aus, you can't just paste a huge list of 'results' giving 1 result for each hardware setup. If you look at the getmininginfo it is refreshed once per minute. If you give that command again in 1 minutes time the CPM and PPS will be different.

For example CPM might go like this: 4, 4, 4, 5, 8, 1, 4, 4, 8, 6, 6, 5, 5, 4, 7

So if you had used getmining info when it was a low value... or a high value for that matter... you are misleading yourself.

The thing to do is take an average of MANY CPM counts (or PPS counts).

[zax983]

How I can use only 2-3 from my threads, not need to use all 4 or just not need to use all 100% from my processor !

[gigawatt]

How I can use only 2-3 from my threads, not need to use all 4 or just not need to use all 100% from my processor !

In your primecoin.conf file, add "genproclimit=N" where N is how many threads you want.  -1 means "use them all".

[gigawatt]

Yes a single Opteron dual core @ 2.8 will easily outperform 2x Xeon DP (single core, HT) @ 3.6 GHz. By a factor of 2-3x. AMD CPU do more work per cycle than intel designs. I get 1600+ and 600-700 PPS respectively from these. The memory and memory subsystem is slightly faster on the AMD too, but not really by much. Both are 64 bit.

As to Tommo_Aus, you can't just paste a huge list of 'results' giving 1 result for each hardware setup. If you look at the getmininginfo it is refreshed once per minute. If you give that command again in 1 minutes time the CPM and PPS will be different.

For example CPM might go like this: 4, 4, 4, 5, 8, 1, 4, 4, 8, 6, 6, 5, 5, 4, 7

So if you had used getmining info when it was a low value... or a high value for that matter... you are misleading yourself.

The thing to do is take an average of MANY CPM counts (or PPS counts).

You should see my awk command, it does exactly that.
Another thing to note: no setting is "good" or "bad".  Each benchmark should be considered in relation to previous benchmarks taken under the same circumstances.  Comparing it to somebody else's hardware and setup is pointless, only compare it to your own previous tests.

[zax983]

How I can use only 2-3 from my threads, not need to use all 4 or just not need to use all 100% from my processor !

In your primecoin.conf file, add "genproclimit=N" where N is how many threads you want.  -1 means "use them all".

Oh, thanks! Hm, I try with 3 and -3 before any time and still 100%, but now with 2 it's 90% it's ok but strange, coz in pool with 3 I have 80%, with 2 I have 50% utilization and I have no other processes that to make more... But better from 100%, anyway...

Still need to know what other best settings for my poor i3-2125

[maco]

How I can use only 2-3 from my threads, not need to use all 4 or just not need to use all 100% from my processor !

In your primecoin.conf file, add "genproclimit=N" where N is how many threads you want.  -1 means "use them all".

Oh, thanks! Hm, I try with 3 and -3 before any time and still 100%, but now with 2 it's 90% it's ok but strange, coz in pool with 3 I have 80%, with 2 I have 50% utilization and I have no other processes that to make more... But better from 100%, anyway...

Still need to know what other best settings for my poor i3-2125

You can play around with it on testnet mode which will give you a faster or more accurate answer.. it is good you are asking the community though

[gigawatt]

How I can use only 2-3 from my threads, not need to use all 4 or just not need to use all 100% from my processor !

In your primecoin.conf file, add "genproclimit=N" where N is how many threads you want.  -1 means "use them all".

Oh, thanks! Hm, I try with 3 and -3 before any time and still 100%, but now with 2 it's 90% it's ok but strange, coz in pool with 3 I have 80%, with 2 I have 50% utilization and I have no other processes that to make more... But better from 100%, anyway...

If you get desperate enough, you can force the primecoin process to only have access to certain CPU cores.
Just open task manager, right click the primecoin process, then click "Set Affinity".  Select which cores you want it to run on, then hit ok.

[LZ]

HP8 crashed twice on Linux and Windows with the same error.

Code:

Assertion failed!
File: src/checkqueue.h, Line 171
Expression pqueue->nTotal == pqueue->nIdle

[mikaelh]

HP8 crashed twice on Linux and Windows with the same error.

Code:

Assertion failed!
File: scr/checkqueue.h, Line 171
Expression pqueue->nTotal == pqueue->nIdle

Yeah, that's an old bug which is still around unfortunately. It's been there since the first official release of Primecoin. I tried to fix it once already but my attempt failed.

[Dsfyu]

HP8 crashed twice on Linux and Windows with the same error.

Code:

Assertion failed!
File: scr/checkqueue.h, Line 171
Expression pqueue->nTotal == pqueue->nIdle

Yeah, that's an old bug which is still around unfortunately. It's been there since the first official release of Primecoin. I tried to fix it once already but my attempt failed.

It's working fine for me so far. I have six machines currently running hp8 and haven't gotten any crashes yet while in hp5/6 I was getting crashes on my desktop after running the program for about an hour. As far as I can tell the problem has been fixed.


Edit: Scratch that, I just had the error just now. Still, good job fixing it, it happens much less frequently now than it did before.

[masterOfDisaster]

HP8 crashed twice on Linux and Windows with the same error.

Code:

Assertion failed!
File: scr/checkqueue.h, Line 171
Expression pqueue->nTotal == pqueue->nIdle

Yeah, that's an old bug which is still around unfortunately. It's been there since the first official release of Primecoin. I tried to fix it once already but my attempt failed.

Maybe it's not completely fixed, but my experience since version hp7 is that it runs more stable. In fact I can not remember having a crashed program ever since. I assume it has nothing to do with the time the program is running and it's more related to chance. Because I run the program rarely for more than one day. The quite stable difficulty is fine to run some tests with various settings on the mainnet
But from what I can say up to now: the default settings are very well. But I encountered a remarkable increase of the 5-chains rate by reducing the "sievesize" to 500k on a phenom 2 x4 955 (the 500 k output is from the last 24 hours, whereas the 1 M output is from the preceding 24 hours; 100 k, 800 k and 2 M have not been that well...):
<edit>
Another investigation revealed, that the 100 k sievesize beats all other tested sievesizes. It's time to start editing  "sievepercentage" or "roundsievepercentage"
</edit>

Code:

grep primemeter ~/.primecoin/debug.log.1M | awk '{s1+=$4;s2+=$6;s3+=$8} END {printf"%10s: %10d\n%10s: %10d\n%10s: %10d\n","Prime/h",s1/NR,"Tes
t/h",s2/NR,"5-Chain/h",s3/NR}'

   Prime/h:    8922316
    Test/h:   72432708
 5-Chain/h:        428

grep primemeter ~/.primecoin/debug.log.100k | awk '{s1+=$4;s2+=$6;s3+=$8} END {printf"%10s: %10d\n%10s: %10d\n%10s: %10d\n","Prime/h",s1/NR,"Tes
t/h",s2/NR,"5-Chain/h",s3/NR}'

   Prime/h:   20544758
    Test/h:  188487687
 5-Chain/h:        616

grep primemeter ~/.primecoin/debug.log.500k | awk '{s1+=$4;s2+=$6;s3+=$8} END {printf"%10s: %10d\n%10s: %10d\n%10s: %10d\n","Prime/h",s1/NR,"Tes
t/h",s2/NR,"5-Chain/h",s3/NR}'

   Prime/h:   15523602
    Test/h:  132365699
 5-Chain/h:        601

I have not yet started messing around with "sievepercentage" or "roundsievepercentage".