[XPM] [ANN] Primecoin High Performance | HP14 released! - page 83.

Tamis

sr. member

Activity: 476

Merit: 250

Quote from: palmface on July 21, 2013, 02:19:33 PM

Quote from: Tamis on July 21, 2013, 02:15:54 PM

Great work mikaelh !

I'll see if this fixes my strange crash with sievesize higher than 3.6M.

Sorry Tamis, still getting that same crash with the new update. I get the same crash with hp4, hp5 and hp6. All are the winx64 versions and it seemed to coincide with difficulty hitting 9.

GL

I was wondering if this was related to difficulty hitting 9+ !
pps is down too so I'm guessing this is also related to the difficulty ?!

Thanks palmface

mikaelh can you give us some info about the sievepercentage parameter ?

palmface

newbie

Activity: 32

Merit: 0

Quote from: Tamis on July 21, 2013, 02:15:54 PM

Great work mikaelh !

I'll see if this fixes my strange crash with sievesize higher than 3.6M.

Sorry Tamis, still getting that same crash with the new update. I get the same crash with hp4, hp5 and hp6. All are the winx64 versions and it seemed to coincide with difficulty hitting 9.

GL

Tamis

sr. member

Activity: 476

Merit: 250

Great work mikaelh !

I'll see if this fixes my strange crash with sievesize higher than 3.6M.

n4ru

sr. member

Activity: 350

Merit: 250

Quote from: mikaelh on July 21, 2013, 01:53:23 PM

-hp6 released!

Lots of changes in this release. Here's a list of notable changes:
* Added fast divisibility tests before doing the expensive Fermat's test
* Lots of other small optimizations
* Introduced a new experimental tuning parameter "sievepercentage". Default value is 10, minimum is 1, maximum is 100.
* Added new RPC command "getchainspermin"
* Added the following information to "getmininginfo": chainspermin, difficulty, sievepercentage, and sievesize
* Attempt to fix the random crash while mining

My PPS on mainnet went up 23% with this release with difficulty >= 9 on mainnet. This release attempts to fix the random crash while mining. I'm not sure if it's actually fixed because I have heard both positive and negative reports. The crash is more likely with high number of cores. 16 and 32 core machines are suffering the most.

The "sievepercentage" parameter is there for people to experiment with. It limits how many base primes are used to filter out candidate multipliers in the sieve. 8% seems optimal for testnet but I have no idea about mainnet.

Thanks for the update, will be testing it on my farm.

mikaelh

sr. member

Activity: 301

Merit: 250

-hp6 released!

Lots of changes in this release. Here's a list of notable changes:
* Added fast divisibility tests before doing the expensive Fermat's test
* Lots of other small optimizations
* Introduced a new experimental tuning parameter "sievepercentage". Default value is 10, minimum is 1, maximum is 100.
* Added new RPC command "getchainspermin"
* Added the following information to "getmininginfo": chainspermin, difficulty, sievepercentage, and sievesize
* Attempt to fix the random crash while mining

My PPS on mainnet went up 23% with this release with difficulty >= 9 on mainnet. This release attempts to fix the random crash while mining. I'm not sure if it's actually fixed because I have heard both positive and negative reports. The crash is more likely with high number of cores. 16 and 32 core machines are suffering the most.

The "sievepercentage" parameter is there for people to experiment with. It limits how many base primes are used to filter out candidate multipliers in the sieve. 8% seems optimal for testnet but I have no idea about mainnet.

notlist3d

legendary

Activity: 1456

Merit: 1000

Thank you guys!

ivanlabrie

hero member

Activity: 812

Merit: 1000

"Maybe im nuts... but can u tell me where to find primecoin.conf. Sorry again for taking up space in forum."

Hit win key, type %appdata%, hit enter.
Look for primecoin folder. Create it if it's not there already.
New text document, rename it to primecoin.conf (you need to see file extensions to do so).

Tamis

sr. member

Activity: 476

Merit: 250

Quote from: notlist3d on July 21, 2013, 01:07:01 PM

Maybe im nuts... but can u tell me where to find primecoin.conf. Sorry again for taking up space in forum.

It's in Appdata/Roaming/Primecoin

Tamis

sr. member

Activity: 476

Merit: 250

something weird is happening with hp5, I set sievesize to 4.8M started mining and went for a nap.
Two hours later I find out the the client crashed. I try to restart it and it crashes again.
I rebooted the computer, same. I cannot put any sievesize higher than 3.6M or the client crashes as soon as it starts mining.
Was running fine a few hours ago at 4.6M and even found a block with this setting.

notlist3d

legendary

Activity: 1456

Merit: 1000

Quote from: Tamis on July 21, 2013, 12:57:28 PM

Quote from: notlist3d on July 21, 2013, 12:29:18 PM

Quote from: Tuck Fheman on July 21, 2013, 09:17:40 AM

Quote from: mikaelh on July 21, 2013, 08:28:16 AM

Use multiples of 200k for sieve size with HP5. One of the internal loops goes over 200k elements. If there's a small bug there somewhere, it might crash when sieve size is not a multiple of 200k.

Thanks for that info!

Sorry for such a newb question. What is command to change sieve size? I didn't see config file for it.

in primecoin.conf add "sievesize=...."

Maybe im nuts... but can u tell me where to find primecoin.conf. Sorry again for taking up space in forum.

Tamis

sr. member

Activity: 476

Merit: 250

Quote from: notlist3d on July 21, 2013, 12:29:18 PM

Quote from: Tuck Fheman on July 21, 2013, 09:17:40 AM

Quote from: mikaelh on July 21, 2013, 08:28:16 AM

Use multiples of 200k for sieve size with HP5. One of the internal loops goes over 200k elements. If there's a small bug there somewhere, it might crash when sieve size is not a multiple of 200k.

Thanks for that info!

Sorry for such a newb question. What is command to change sieve size? I didn't see config file for it.

in primecoin.conf add "sievesize=...."

notlist3d

legendary

Activity: 1456

Merit: 1000

Quote from: Tuck Fheman on July 21, 2013, 09:17:40 AM

Quote from: mikaelh on July 21, 2013, 08:28:16 AM

Use multiples of 200k for sieve size with HP5. One of the internal loops goes over 200k elements. If there's a small bug there somewhere, it might crash when sieve size is not a multiple of 200k.

Thanks for that info!

Sorry for such a newb question. What is command to change sieve size? I didn't see config file for it.

notlist3d

legendary

Activity: 1456

Merit: 1000

Quote from: krikunalta on July 21, 2013, 11:45:28 AM

hp5 client crashes on win 7 64bit random

Same here i went back to 4 after a few crashes.

krikunalta

newbie

Activity: 16

Merit: 0

hp5 client crashes on win 7 64bit random

nmersulypnem

full member

Activity: 238

Merit: 100

Quote from: Tuck Fheman on July 21, 2013, 09:17:40 AM

Quote from: mikaelh on July 21, 2013, 08:28:16 AM

Use multiples of 200k for sieve size with HP5. One of the internal loops goes over 200k elements. If there's a small bug there somewhere, it might crash when sieve size is not a multiple of 200k.

Thanks for that info!

I think their might be some race conditions that don't exist on testnet. Does anyone have timing numbers on how long the sieve takes to wieve and how long we have, on average, to find a block before someone else solves it?

Tuck Fheman

sr. member

Activity: 363

Merit: 250

Quote from: mikaelh on July 21, 2013, 08:28:16 AM

Use multiples of 200k for sieve size with HP5. One of the internal loops goes over 200k elements. If there's a small bug there somewhere, it might crash when sieve size is not a multiple of 200k.

Thanks for that info!

mikaelh

sr. member

Activity: 301

Merit: 250

Quote from: Tuck Fheman on July 21, 2013, 07:46:40 AM

HP5 - 6144000 - 7876 - 29 (4 at a time usually + 6 in 1 minute) - same result using "-1" or "6" threads.

Use multiples of 200k for sieve size with HP5. One of the internal loops goes over 200k elements. If there's a small bug there somewhere, it might crash when sieve size is not a multiple of 200k.

Tamis

sr. member

Activity: 476

Merit: 250

Thanks for the details Tuck Fheman !

Here are my recent tests : https://bitcointalksearch.org/topic/m.2772839

It's intersting to note that we do not have the same results with the same sievesize on testnet, that's could be cpu dependent or like you say testnet is not reliable at all.

Tuck Fheman

sr. member

Activity: 363

Merit: 250

Quote from: Tamis on July 21, 2013, 04:06:46 AM

Quote from: Tuck Fheman on July 20, 2013, 09:06:51 PM

On testnet, I found that a sievesize=6144000 outperformed every other sievesize I tried (incl. 6000000 & 6246400) by 30% using an i7.

However, when i try to use this sievesize mining live, h5 crashes as soon as mining starts (every time). It's unfortunate because it showed such promise on testnet.

Blocks were coming in 4-5 chunks at a time steadily, even hit 6 blocks in 1 minute (no other setting hit above 4 blocks/min).

same here, 6144000 crashed h5 as soon as it attemps to mine on a i7 3770k with 16gb of ram.

Did you get better results on testnet with 4M than 1M ?

How do you activate testnet ? I can't check it on the debug window.

The best sievesize's I found were ...

Build - Sievesize - PPS - Blocks in 10 minutes
HP3 - 1000000 - 7387 - 14
HP5 - 2000000 - 7092 - 14
HP5 - 4000000 - 7911 - 15 (3 at a time usually)
HP4 - 6000000 - 7819 - 15
HP4 - 6144000 - 7430 - 20
HP3 - 6553600 - 5042 - 20 (steady 2 per minute)
HP5 - 5000000 - 7815 - 22
HP5 - 6000000 - 6919 - 23
HP5 - 6144000 - 7876 - 29 (4 at a time usually + 6 in 1 minute) - same result using "-1" or "6" threads.

Nothing above 4608000 worked for me mining live w/HP5 and using higher settings live w/HP3 worked (without crashing) but was yielding very low PPS (in the hundreds) live.

I get a higher PPS @ 1M or 2M live on HP5.

I have no idea if the testnet actually "test" anything to be honest. Sad

None of the data I collected translates over into real world mining ... especially since the client shuts down and no blocks are being found for days using the same settings that worked so well on the testnet.

So I'm trying the highest testnet settings that yielded the best results that will still work live and seeing how it goes. So far (6 hours or so, no blocks).

To answer your question ... add "-testnet" to the command line.

mikaelh

sr. member

Activity: 301

Merit: 250

Quote from: maco on July 21, 2013, 05:18:50 AM

/primecoin-hp/src# primecoind stop
error: incorrect rpcuser or rpcpassword (authorization failed)

Your primecoin daemon was started with a different password in the config file. Just kill it off by typing:

Code:

killall primecoind

Topic: [XPM] [ANN] Primecoin High Performance | HP14 released! - page 83. (Read 397616 times)