[ANN][GAP] Gapcoin - Prime Gap Search - New Math Algo - CPU / GPU - Zero Premine

[Jamil.Kamil]

I don't want to be any gap between me and GAP...

[1714007730]

1714007730

[1714007730]

1714007730

[1714007730]

1714007730

[pdazzl]

I tried mining some with some shifts on the range around 380 but miners gave me 'illegal instruction core dump' error.

I then noticed that it gives me that error for every shift bigger than 192. So i put my miners running with shift 187-192. Lets see what happens.

It seems to work. Any reason why i get this error? Can my binaries be the problem?

My PPS is halved btw.

Looks like you've found a world record already on a shift 192--length 7546

https://bchain.info/GAP/block/8af4723fc2e9b9f460455ab09d38dcc9649d9a06e2927783a0e89c6efd522075

[j0nn9]

I tried mining some with some shifts on the range around 380 but miners gave me 'illegal instruction core dump' error.

I then noticed that it gives me that error for every shift bigger than 192. So i put my miners running with shift 187-192. Lets see what happens.

It seems to work. Any reason why i get this error? Can my binaries be the problem?

My PPS is halved btw.

Yes that's probably the binary, or some library. With what hardware, system and binary do you mine?

Thanks, will be great to see CRT improvements.  Will using default --sieve-primes and --sieve-size still be what to use on larger shifts with the new miner?

Currently I'm testing with the default values, but they're probably not suitable for any shift value.

[GreekBitcoin]

I tried mining some with some shifts on the range around 380 but miners gave me 'illegal instruction core dump' error.

I then noticed that it gives me that error for every shift bigger than 192. So i put my miners running with shift 187-192. Lets see what happens.

It seems to work. Any reason why i get this error? Can my binaries be the problem?

My PPS is halved btw.

Looks like you've found a world record already on a shift 192--length 7546

https://bchain.info/GAP/block/8af4723fc2e9b9f460455ab09d38dcc9649d9a06e2927783a0e89c6efd522075

Yep i checked and i think its mine Hurray!


j0nn9, pdazzl

I have a few VMs with AMD processors. I think i have plenty of memory and thats not the problem. Running Ubuntu 14.04.  Unfortunately i cant install dependencies so i can only use binaries. Now since i am quite noob on those things, i know just the basics, when i made some binaries from the source  on my personal computer i couldnt run it on my VMs because of missing dependencies. See, i dont really know how to build nice static binaries (damn i am not sure if those are called like this, i am just a physics guy ) So i got permissions to install dependencies on one of my VMs once and somehow i built a binary that managed to work on the other VMs. The binary was like 15% slower but it did the job.

So what i think i need is a guide to show me how to static build gapminer...


Btw, i was thinking about an idea that probably will bring more people aboard. I am not sure if it can be done or maybe if it is bad so i am just writing it here for discussion. Is it possible to automatically know who (if he wants ofc) is mining the world record blocks so that we can list the records like this: Gapcoin[Username]
You know, the idea is that you are mining gaps for the coins, for science and for a position on the 'leaderboards'...

[pdazzl]

Btw, i was thinking about an idea that probably will bring more people aboard. I am not sure if it can be done or maybe if it is bad so i am just writing it here for discussion. Is it possible to automatically know who (if he wants ofc) is mining the world record blocks so that we can list the records like this: Gapcoin[Username]
You know, the idea is that you are mining gaps for the coins, for science and for a position on the 'leaderboards'...

A username would probably have to be embedded in the blockchain somehow.  Either as a custom wallet address with a person's username or another optional field setting on a solved block which I've read can be done with Bitcoin and this coin is a fork of it.  I'm sure j0nn9 knows more.

[pdazzl]

Currently I'm testing with the default values, but they're probably not suitable for any shift value.

Someone did a shift 480 with some ridiculously large adder.  j0nn9 - this isn't your new miner is it?

https://bchain.info/GAP/block/f524a95290029ba9c6de5708392464b3522db02519be850612bac4287644ec8e

[j0nn9]

Currently I'm testing with the default values, but they're probably not suitable for any shift value.

Someone did a shift 480 with some ridiculously large adder.  j0nn9 - this isn't your new miner is it?

https://bchain.info/GAP/block/f524a95290029ba9c6de5708392464b3522db02519be850612bac4287644ec8e

Yes, that was me running a test on mainnet. Unfortunately it wasn't a new record.

Btw, i was thinking about an idea that probably will bring more people aboard. I am not sure if it can be done or maybe if it is bad so i am just writing it here for discussion. Is it possible to automatically know who (if he wants ofc) is mining the world record blocks so that we can list the records like this: Gapcoin[Username]
You know, the idea is that you are mining gaps for the coins, for science and for a position on the 'leaderboards'...

A username would probably have to be embedded in the blockchain somehow.  Either as a custom wallet address with a person's username or another optional field setting on a solved block which I've read can be done with Bitcoin and this coin is a fork of it.  I'm sure j0nn9 knows more.

Theoretically this could be done using OP_RETURN, but it would require serious changes in the miner.
And if you want to bind a username to a specific miner it gets a lot harder. So at least it's not a short time goal.

[pdazzl]

@GreekBitcoin - I'm counting 19 records you've set already, nice!

[j0nn9]

New experimental GapMiner release (CPU only)

Revision 5: capable of using the Chinese Remainder Theorem to speed up mining with large shift.


windows: https://github.com/gapcoin/GapMiner/releases/download/crt-rev5/gapminer-windows-rev5.zip
md5: dbae859b1e8825f4f9b3f693004833c4

linux: https://github.com/gapcoin/GapMiner/releases/download/crt-rev5/gapminer-linux-rev5.zip
md5: 909f7fcb141a23d60b04ba8c0df1f391

source code: https://github.com/gapcoin/GapMiner/archive/crt-rev5.zip
source code PoWCore: https://github.com/gapcoin/Gapcoin-PoWCore/archive/master.zip


The usage should be pretty self explaining from the start-gapminer file.
There are different CRT files containing values optimized for a specific shift.


There are 3 new verbose outputs:

• gaps/s:  The amount of gaps scanned  per second
• gaplist:  The number of sieved gaps ready for scanning
• block:  The percent of the calculation amount of an average block

The primes per second and the block percent values are estimated using the theoretical speed improve from the CRT, so they don't have to be correct.

Mining with the CRT is optimized for solo mining.
You can mine at the pool, but you will probably mine less shares, and the pps and block values are probably incorrect, but your chance of finding a block is almost the same as when mining solo.

Mining with the CRT is splitted in sieve and primality testing using separate threads for each:

Code:

--threads 4 --fermat-threads 1

means: 3 sieve and 1 gap scan thread. So you have to use at least 2 threads.

The scan threads always pick the most promising gap from the gaplist, therefore the gaplist value should always be at least over 100, but a too high gaplist value can slow down mining, (for example over 9000). You can alter --sieve-primes, --threads or --fermat-threads to achieve this.
(--sieve-size is not used for CRT mining)
FYI: The gaplist is implemented as an heap


The is also an logging version, which produces additional output in a gapminer.log file.
If something goes wrong, you can switch versions, and look whats wrong, or at least give feedback.

[j0nn9]

So what i think i need is a guide to show me how to static build gapminer...

https://bitcointalk.org/index.php?topic=149479.0 should be a good starting point,
for linux you have to change some of the config parameters

You have to statically compile all dependent libraries: boost, gmp, mpfr, openssl, curl.
And then link it together.

An other option is to remove the -march=native switch when compiling dynamically. The resultant binary should be system independent.

[pdazzl]

New experimental GapMiner release (CPU only)

Revision 5: capable of using the Chinese Remainder Theorem to speed up mining with large shift.


windows: https://github.com/gapcoin/GapMiner/releases/download/crt-rev5/gapminer-windows-rev5.zip
md5: dbae859b1e8825f4f9b3f693004833c4

linux: https://github.com/gapcoin/GapMiner/releases/download/crt-rev5/gapminer-linux-rev5.zip
md5: 909f7fcb141a23d60b04ba8c0df1f391

source code: https://github.com/gapcoin/GapMiner/archive/crt-rev5.zip


The usage should be pretty self explaining from the start-gapminer file.
There are different CRT files containing values optimized for a specific shift.


There are 3 new verbose outputs:

• gaps/s:  The amount of gaps scanned  per second
• gaplist:  The number of sieved gaps ready for scanning
• block:  The percent of the calculation amount of an average block

The primes per second and the block percent values are estimated using the theoretical speed improve from the CRT, so they don't have to be correct.

Mining with the CRT is optimized for solo mining.
You can mine at the pool, but you will probably mine less shares, and the pps and block values are probably incorrect, but your chance of finding a block is almost the same as when mining solo.

Mining with the CRT is splitted in sieve and primality testing using separate threads for each:

Code:

--threads 4 --fermat-threads 1

means: 3 sieve and 1 gap scan thread. So you have to use at least 2 threads.

The scan threads always pick the most promising gap from the gaplist, therefore the gaplist value should always be at least over 100, but a too high gaplist value can slow down mining, (for example over 9000). You can alter --sieve-primes, --threads or --fermat-threads to achieve this.
(--sieve-size is not used for CRT mining)
FYI: The gaplist is implemented as an heap


The is also an logging version, which produces additional output in a gapminer.log file.
If something goes wrong, you can switch versions, and look whats wrong, or at least give feedback.

Thanks j0nn9, trying some large shifts.  Seems like almost the same performance on really enormous shifts vs fairly large ones, pretty neat.

[pdazzl]

Got my first big CRT block, a record too.  Shift 512, length 13306.  On these really big shifts, a much higher overall Fermat thread ratio appears to yield better results.

https://bchain.info/GAP/block/cca2fe349c7c75098a0fba31139260e6b64f055a6e34e8b564f9070b9d0aec1c

@j0nn9 - You may need to add at least the 20k-25k record check list on gapcoin.org.  That's easily reachable on shift 1024.

[enerbyte]

@j0nn9 thanks for the new gapminer.

[pdazzl]

@j0nn9 thanks for the new gapminer.

I second that, kudos to j0nn9.


Here are my thoughts from initially testing out the new crt miner.  What I think is my ideal performance setup is setting roughly half the total threads to Fermat threads.  My sieve-primes # is what I then fine tune on a particular shift size(near the end of fine tuning only adjusting 1000 on sieve-primes at a time then re-running for a minute to observe).  The ideal behavior is the gaplist size generally(most of the time) stays above 100.  It'll creep up to a 2000-3000 then oscillate back down to 100 and then continues in this see-saw fashion.  The Fermat test speeds seem to kick in faster when the list grows and tamps down the list.

Those are my findings after only 24 hours.  Two shift 512's so far.  I parked my beefier server on shift 1024 for most of the day but gave up after the probability reached 170% with no solved block.

[vidarn]

New experimental GapMiner release (CPU only)

source code: https://github.com/gapcoin/GapMiner/archive/crt-rev5.zip

Hi j0nn9,
In crt-rev5.zip I can only find the original source code (minus the POW) from October last year. Did you mix up some batches of code? (or am I doing something wrong?)

[j0nn9]

New experimental GapMiner release (CPU only)

source code: https://github.com/gapcoin/GapMiner/archive/crt-rev5.zip

Hi j0nn9,
In crt-rev5.zip I can only find the original source code (minus the POW) from October last year. Did you mix up some batches of code? (or am I doing something wrong?)

My mistake, wrong branch. Links are fixed.
Also POW is submodule which must be downloaded separately:
https://github.com/gapcoin/Gapcoin-PoWCore/archive/master.zip


You can also use git:

Code:

git clone https://github.com/gapcoin/GapMiner -b gpu-miner
cd GapMiner
git submodule update --init

[pdazzl]

Gapcoin's first record in the 15000-19998 range with a shift 700
https://bchain.info/GAP/block/125025

@j0nn9 - Looks like gapcoin.org thinks shift 700 is shift 594?  Also, curious if there are any plans for a CRT on GPU?  My machine with faster RAM seems to have gotten a better benefit to this new algo, would be interesting to see how a GPU would handle it.


Edit: Looks like the shift 700 blew the doors off the bchain.info calculator, it concatenated some of the last digits on the start number.  Calculating by hand the two ends are primes per factordb.

[vidarn]

My mistake, wrong branch. Links are fixed.
Also POW is submodule which must be downloaded separately:
https://github.com/gapcoin/Gapcoin-PoWCore/archive/master.zip

You can also use git:

Code:

git clone https://github.com/gapcoin/GapMiner -b gpu-miner
cd GapMiner
git submodule update --init

Thanks!

[j0nn9]

Gapcoin's first record in the 15000-19998 range with a shift 700
https://bchain.info/GAP/block/125025

@j0nn9 - Looks like gapcoin.org thinks shift 700 is shift 594?  Also, curious if there are any plans for a CRT on GPU?  My machine with faster RAM seems to have gotten a better benefit to this new algo, would be interesting to see how a GPU would handle it.


Edit: Looks like the shift 700 blew the doors off the bchain.info calculator, it concatenated some of the last digits on the start number.  Calculating by hand the two ends are primes per factordb.

Thanks for pointing that out!

It was a database error on gapcoin.org. For the gap start and end there were only 256 Digits available, so that the last (decimal) digits of the number weren't saved.
bchain.info is propably suffering from a similar issue.

The database is currently recreated, gapcoin.org should be fixed within a few hours.

[pdazzl]

Gapcoin's first record in the 20000-24998 range, shift 920
https://bchain.info/GAP/block/125746

@j0nn9 - Is it possible to get 20000-24998 added to the gapcoin.org record range?
EDIT: Nevermind, looks like it works!

Also for anyone interested in hunting for larger overall length gaps, here are some settings that have worked for me (your mileage may vary).  As time goes on I'm starting to think in some ways larger shifts can be tuned for larger performance (notice I set the fermat thread to total thread ratio very high on these).

Shift 702 on an i7 (8 cores)
./gapminer-cpu -o 127.0.0.1 -p 31397 -u ***** -x ***** -t 7 --shift 702 --crt crt/crt-22m-704s.txt --fermat-threads 5 --sieve-primes 450000

Shift 920 on a 4 xeon server (16 cores)
./gapminer-cpu -o 127.0.0.1 -p 31397 -u ***** -x ***** -t 18 --shift 920 --crt crt/crt-22m-928s.txt --fermat-threads 15 --sieve-primes 300000