Open Source XPM Pool + GPU Miner (aka. madPrimeMiner)

[eXtremal]

I can't optimize miner for 270/280X, because I don't have them.
Version 9.1: https://www.dropbox.com/s/auba1v24elkz23y/xpmclient_v9.1.tar.gz?dl=0

- All OpenCL kernel parameters (chain length target, sieve size, etc) moved to config.txt file. After changing kernel parameters you must delete old kernel.bin file before miner launching
- Full 11ch target support (set target = "11"; in config.txt when difficulty reach ~10.988-10.99).

For 270 and 280X cards you can try:
- primorials 12, 13 and 14
- "sieveSize" values - 210, 420, 630 and 840
- different "weaveDepth" values (must be a multiple of 256)

[1715480621]

1715480621

[ap1]

I can't optimize miner for 270/280X, because I don't have them.
Version 9.1: https://www.dropbox.com/s/auba1v24elkz23y/xpmclient_v9.1.tar.gz?dl=0

- All OpenCL kernel parameters (chain length target, sieve size, etc) moved to config.txt file. After changing kernel parameters you must delete old kernel.bin file before miner launching
- Full 11ch target support (set target = "11"; in config.txt when difficulty reach ~10.988-10.99).

For 270 and 280X cards you can try:
- primorials 12, 13 and 14
- "sieveSize" values - 210, 420, 630 and 840
- different "weaveDepth" values (must be a multiple of 256)

Sure no problem and I don't expect that you will optimize it for all cards as you are planning to support old ATi cards and even Nvidia Maxwell.
Just give you the information that old/slower cards probably need much different tuning parameters. 

[eXtremal]

My experiments with R9 280X (1100/1250) and version 9,1:

primorial 14

[GPU 0] T=64C A=97% E=0 primes=0.104240 fermat=709830/sec cpd=9.29/day
(ST/INV/DUP): 1x 9ch(0/0/0)
Work received: height=736596 diff=10.976544 latency=169ms
Work received: height=736597 diff=10.976542 latency=167ms
[GPU 0] T=65C A=97% E=0 primes=0.104230 fermat=700775/sec cpd=9.16/day
(ST/INV/DUP): 1x 9ch(0/0/0)
Work received: height=736598 diff=10.976527 latency=167ms
Work received: height=736599 diff=10.976546 latency=170ms
[GPU 0] T=64C A=95% E=0 primes=0.104253 fermat=708750/sec cpd=9.29/day
(ST/INV/DUP): 1x 9ch(0/0/0)

primorial 13

[GPU 0] T=64C A=95% E=0 primes=0.105844 fermat=610418/sec cpd=9.31/day
(ST/INV/DUP): 1x 10ch(0/0/0)
[GPU 0] T=65C A=97% E=0 primes=0.105879 fermat=622893/sec cpd=9.53/day
(ST/INV/DUP): 1x 10ch(0/0/0)
[GPU 0] T=65C A=97% E=0 primes=0.105916 fermat=617762/sec cpd=9.48/day
(ST/INV/DUP): 1x 10ch(0/0/0)
Work received: height=736534 diff=10.976459 latency=172ms
Work received: height=736535 diff=10.976453 latency=172ms

primorial 12

Work received: height=736542 diff=10.976462 latency=173ms
[GPU 0] T=65C A=95% E=0 primes=0.107466 fermat=497988/sec cpd=8.84/day
(ST/INV/DUP): none
Work received: height=736543 diff=10.976461 latency=172ms
Work received: height=736544 diff=10.976463 latency=175ms
[GPU 0] T=65C A=94% E=0 primes=0.107517 fermat=493357/sec cpd=8.80/day
(ST/INV/DUP): none

primorial 13 is optimal, trying change weave depth to 38912:

[GPU 0] T=65C A=97% E=0 primes=0.105512 fermat=650201/sec cpd=9.61/day
(ST/INV/DUP): 3x 9ch(0/0/0)
Work received: height=736833 diff=10.976756 latency=175ms
Work received: height=736834 diff=10.976754 latency=172ms
[GPU 0] T=65C A=97% E=0 primes=0.105512 fermat=650914/sec cpd=9.62/day
(ST/INV/DUP): 3x 9ch(0/0/0)
Work received: height=736835 diff=10.976758 latency=176ms
[GPU 0] T=65C A=97% E=0 primes=0.105516 fermat=651608/sec cpd=9.63/day
(ST/INV/DUP): 3x 9ch(0/0/0)

Memory overclocking to 1500 test:

[GPU 0] T=66C A=97% E=0 primes=0.105486 fermat=662313/sec cpd=9.76/day
(ST/INV/DUP): none
[GPU 0] T=66C A=97% E=0 primes=0.105513 fermat=661911/sec cpd=9.78/day
(ST/INV/DUP): none

You can reach ~9,8CPD on R9 280X @ 1100/1500 with primorial = 13 and weave depth = 38912

[ivanlabrie]

Amazing results, props for the great work you've put out.

[ap1]

You can reach ~9,8CPD on R9 280X @ 1100/1500 with primorial = 13 and weave depth = 38912

Thanks for your effort. I have tried v9.1 and it is quite funny that it seems default weave depth is faster this time...  I need to test more to confirm it.

Are you going to update your git source after you get to a stable release? (I know the CL code for kernel is in the tar.gz already.)

[eXtremal]

I have tried v9.1 and it is quite funny that it seems default weave depth is faster this time...  I need to test more to confirm it.

I run 280X with default weave depth again, waiting...

Are you going to update your git source after you get to a stable release? (I know the CL code for kernel is in the tar.gz already.)

Latest sources already in repository https://github.com/eXtremal-ik7/xpmclient
CL code sources need for recompile after kernel parameters (sieve size, weave depth, width and target) changed.

[eXtremal]

it seems default weave depth is faster this time

Not faster, same speed ~9,7x on R9 280X 1100/1500

And some results for 11ch target. Primorial 13:

[GPU 0] T=68C A=96% E=0 primes=0.105659 fermat=221072/sec cpd=0.35/day
(ST/INV/DUP): 3x 9ch(0/0/0)
Work received: height=740488 diff=10.979589 latency=171ms
[GPU 0] T=68C A=96% E=0 primes=0.105674 fermat=224389/sec cpd=0.36/day
(ST/INV/DUP): 3x 9ch(0/0/0)
Work received: height=740489 diff=10.979591 latency=169ms
Work received: height=740490 diff=10.979592 latency=170ms
Work received: height=740491 diff=10.979594 latency=170ms
[GPU 0] T=68C A=94% E=0 primes=0.105695 fermat=219435/sec cpd=0.35/day

Primorial 14:

GPU 0: core=1100MHz mem=1500MHz powertune=20 fanspeed=70
(ST/INV/DUP): 7x 9ch(0/0/0) 1x 10ch(0/0/0)
[GPU 0] T=69C A=96% E=0 primes=0.103969 fermat=284947/sec cpd=0.38/day
(ST/INV/DUP): 7x 9ch(0/0/0) 1x 10ch(0/0/0)
[GPU 0] T=70C A=96% E=0 primes=0.103970 fermat=284381/sec cpd=0.38/day
(ST/INV/DUP): 7x 9ch(0/0/0) 1x 10ch(0/0/0)

For target 11, primorial 14 is optimal.

[kev7112001]

it seems default weave depth is faster this time

Not faster, same speed ~9,7x on R9 280X 1100/1500

And some results for 11ch target. Primorial 13:

[GPU 0] T=68C A=96% E=0 primes=0.105659 fermat=221072/sec cpd=0.35/day
(ST/INV/DUP): 3x 9ch(0/0/0)
Work received: height=740488 diff=10.979589 latency=171ms
[GPU 0] T=68C A=96% E=0 primes=0.105674 fermat=224389/sec cpd=0.36/day
(ST/INV/DUP): 3x 9ch(0/0/0)
Work received: height=740489 diff=10.979591 latency=169ms
Work received: height=740490 diff=10.979592 latency=170ms
Work received: height=740491 diff=10.979594 latency=170ms
[GPU 0] T=68C A=94% E=0 primes=0.105695 fermat=219435/sec cpd=0.35/day

Primorial 14:

GPU 0: core=1100MHz mem=1500MHz powertune=20 fanspeed=70
(ST/INV/DUP): 7x 9ch(0/0/0) 1x 10ch(0/0/0)
[GPU 0] T=69C A=96% E=0 primes=0.103969 fermat=284947/sec cpd=0.38/day
(ST/INV/DUP): 7x 9ch(0/0/0) 1x 10ch(0/0/0)
[GPU 0] T=70C A=96% E=0 primes=0.103970 fermat=284381/sec cpd=0.38/day
(ST/INV/DUP): 7x 9ch(0/0/0) 1x 10ch(0/0/0)

For target 11, primorial 14 is optimal.

yeah i was about to ask if the cpd was correct for 11ch i got the same. also how do u figure the weave depth any calcs on that

[eXtremal]

also how do u figure the weave depth any calcs on that

Weave depth is a number of primes using by sieve weaving... more primes - less candidates on sieve output but with more prime probability.

Difficulty stop growing, 11ch target not needed again

[kev7112001]

also how do u figure the weave depth any calcs on that

Weave depth is a number of primes using by sieve weaving... more primes - less candidates on sieve output but with more prime probability.

Difficulty stop growing, 11ch target not needed again

yeah i know what it is but i was just wondering how to figure out the sizes i should u with other settings

also u see that shieldcoin guess its using prime too looks like a moded primeminer

[eXtremal]

Version 9.2: https://www.dropbox.com/s/3pk4c6l7g8lpfsv/xpmclient_v9.2.tar.gz?dl=0
- +3% on R9 280X, stable ~10CPD with clocks 1100/1500:

[GPU 0] T=70C A=99% E=0 primes=0.105558 fermat=676272/sec cpd=10.04/day
GPU 0: core=1100MHz mem=1500MHz powertune=20 fanspeed=70
(ST/INV/DUP): none
[GPU 0] T=70C A=99% E=0 primes=0.105597 fermat=672678/sec cpd=10.02/day
(ST/INV/DUP): none
Work received: height=751324 diff=10.974172 latency=165ms

Stock R9 290X shows up to 14.1-14.2CPD (catalyst 14.7, primorial 13 used)

- Sieve benchmark added (for benchmarking, run ./xpmclient -b)

square 320 bits: 20.752ms (6467.701M ops/sec)
multiply 320 bits: 27.231ms (4928.858M ops/sec)
square 352 bits: 25.454ms (5272.952M ops/sec)
multiply 352 bits: 32.821ms (4089.386M ops/sec)
Fermat tests 320 bits: 36.403ms (3.601M ops/sec)
Fermat tests 352 bits: 47.552ms (2.756M ops/sec)

 *** hashmod benchmark ***
 MHash per second: 592.928
 Hash per iteration: 32.484 (0.000387 %)
 Hashes by multipliers count:
   * [8] 31.062 (95.623%)
   * [9] 1.344 (4.137%)
   * [10] 0.078 (0.241%)


 *** sieve (check) benchmark ***
 * [OK] found candidates by CPU: 5456 by GPU: 5461
 * [OK] invalid candidates: 0
 * [OK] CPU/GPU candidates difference: 0

 *** sieve (performance) benchmark ***
 * scan speed: 94.472 G
 * iteration time: 5.827ms
 * candidates per second: 1062827.414
 * candidates per iteration: 6193.26 (3228.46 320bit, 2964.81 352bit)
 * 320bit/352bit ratio: 1.089/1

- Reboot system after GPU crash (properly untested feature)

yeah i know what it is but i was just wondering how to figure out the sizes i should u with other settings

If I understand right.. only experiments helps you. For 10ch target default value optimal, for 9ch and 11ch may be another value of weave depth better.

also u see that shieldcoin guess its using prime too looks like a moded primeminer

You about this https://bitcointalk.org/index.php?topic=681570.0 ? It's X15.

[kev7112001]

https://bitcointalk.org/index.php?topic=657601.0 coinshield not shieldcoin my bad

[eXtremal]

https://bitcointalk.org/index.php?topic=657601.0 coinshield not shieldcoin my bad

Coin uses Riecoin algorithm

First block with 13ch share found by this miner
https://coinplorer.com/XPM/Blocks/47707ad92585d6f73afcabab31f409b34c2c6178ec02791d8cd289ab729f1643

[kev7112001]

i noticed this new version gives errors if the cpu is being tasked before it would just lower speed some

[LongAndShort]

https://bitcointalk.org/index.php?topic=657601.0 coinshield not shieldcoin my bad

Coin uses Riecoin algorithm

First block with 13ch share found by this miner
https://coinplorer.com/XPM/Blocks/47707ad92585d6f73afcabab31f409b34c2c6178ec02791d8cd289ab729f1643

Awesome i also found one ages ago with your epic optimisations in the miner

Thanks for 9.2 you really are a baws. Please give me your xpm address again please i want to donate some more to you

[kev7112001]

i keep getting random errors every once in a while even with rigs that arent using any cpu usage just xpm mining only happens with 9.2

[kev7112001]

were at a all time high again

[LongAndShort]

were at a all time high again

Yup  diff=10.980315 pretty high indeed looks like other gpu coins are not as profitable. I guess and people are looking to return to or diversify.

[seansplayin]

I'm trying to compile the primeserver but I'm getting  /usr/bin/ld: cannot find -lboost_random
/usr/bin/ld: cannot find -lpq
collect2: ld returned 1 exit status
make: *** [primecoind] Error 1

I have not Set the include paths etc. in makefile.unix, could this be what's giving me the error?

Ubuntu 12.04

[eXtremal]

Ubuntu 12.04

sudo apt-get install libboost-all-dev libpq-dev