[ANN] sph-sgminer: multi-coin multi-algorithm GPU miner | added MaruCoin - page 27.

gpuminer

full member

Activity: 210

Merit: 100

Quote from: pallas on April 11, 2014, 09:00:18 AM

The cuda guys are saying that optimizing groestl will mean much faster X11 and quark.
I don't think so: I did some tests on X11 and groestl takes about 20% of the time; so even if you double its performance, globally you will just get 10% improvement.
Granted, 20% is much more than what is taken for example by keccak (just a couple percent).

maybe he got something else in addition to groestl. He was saying cuda gonna own groestl, x11 and quark. He seems serious and do not want to release the source code.

pallas

legendary

Activity: 2716

Merit: 1094

Black Belt Developer

The cuda guys are saying that optimizing groestl will mean much faster X11 and quark.
I don't think so: I did some tests on X11 and groestl takes about 20% of the time; so even if you double its performance, globally you will just get 10% improvement.
Granted, 20% is much more than what is taken for example by keccak (just a couple percent).

xfelinho

member

Activity: 68

Merit: 10

current groestlcoin performance on 3 GTX 780Ti (only mild overclock)

Code:

[2014-04-11 01:54:41] accepted: 111/111 (100.00%), 29119 khash/s (yay!!!)
[2014-04-11 01:54:46] accepted: 112/112 (100.00%), 29190 khash/s (yay!!!)
[2014-04-11 01:54:50] accepted: 113/113 (100.00%), 29117 khash/s (yay!!!)
[2014-04-11 01:54:54] accepted: 114/114 (100.00%), 29186 khash/s (yay!!!)

I am sure we can do better Wink

Christian

gpuminer

full member

Activity: 210

Merit: 100

Quote from: ol92 on April 11, 2014, 07:50:21 AM

Quote from: xfelinho on April 11, 2014, 07:33:03 AM

Groestl is useless atm, cause Cuda Miner is 2-3x faster Undecided

no :
780ti : 11Mh/s (don't know if oc or not)
R9 290 : 9Mh/s (gpu@1100)

R9 290 9.9 Mh/s (gpu @1100)

amd sph miner beaten but not so much

how about power consumption ? If put that into consideration , which is better ?

gpuminer

full member

Activity: 210

Merit: 100

Quote from: xfelinho on April 11, 2014, 07:33:03 AM

Groestl is useless atm, cause Cuda Miner is 2-3x faster Undecided

how about x11 ? Does Cuda Miner perform better than sph miner ?

ol92

sr. member

Activity: 445

Merit: 255

Quote from: xfelinho on April 11, 2014, 07:33:03 AM

Groestl is useless atm, cause Cuda Miner is 2-3x faster Undecided

no :
780ti : 11Mh/s (don't know if oc or not)
R9 290 : 9Mh/s (gpu@1100)

R9 290 9.9 Mh/s (gpu @1100)

amd sph miner beaten but not so much

djm34

legendary

Activity: 1400

Merit: 1050

Quote from: xfelinho on April 11, 2014, 07:33:03 AM

Groestl is useless atm, cause Cuda Miner is 2-3x faster Undecided

yep useless... Roll Eyes

you should try to mine ~bcx~ instead, very rare coin and nvidia can't even get any....

xfelinho

member

Activity: 68

Merit: 10

Groestl is useless atm, cause Cuda Miner is 2-3x faster Undecided

restless

legendary

Activity: 1151

Merit: 1001

https://github.com/badman74/Binaries/archive/master.zip - latest binary from Badman74

loonb

full member

Activity: 193

Merit: 100

Any working url to last ver?

Judaz49

full member

Activity: 135

Merit: 100

did anyone find a solution to 10% hashrate drop after 10mins of mining darkcoin with r9 290s?

restless

legendary

Activity: 1151

Merit: 1001

1 more X11 coin - hashcoin https://bitcointalksearch.org/topic/ann-hashcoin-hash-x11-dgw-v2-automatic-checkpoints-instant-confirms-560955 Wink

istvandv

sr. member

Activity: 342

Merit: 250

Quote from: badman74 on April 08, 2014, 06:13:17 AM

Quote from: restless on April 08, 2014, 02:52:24 AM

Maybe someone will add support for Aiden coin? https://bitcointalksearch.org/topic/ann-aiden-1st-to-use-scrypt-og-gpu-optimized-v01022-released-71015-558414
Its supposed to be scrypt with nfactor 6, but the sgminer build they offer doesn't work for me on 5850 and 6970
And my attempts to use it badman builds are unsuccessful

PS:
-k zuikkis --nfactor 6
doesn't work, maybe i'm missing something? It seems there are hundreds of sgminer builds and all use slightly different syntax and/or features Sad

try that on the build from my github

i have the exact same problem, im running 7850 on linux and ive tried fresh compilations of:

veox 4.1.242 - gpu 0 kh/s and after quitting sgminer goes defuct and using 100% cpu
zuikkis 4.1.153 - segmentation fault core dumped
troky 4.1.153 - gpu 0 kh/s and after quitting sgminer goes defuct and using 100% cpu
badman74 4.1.153 - gpu 0 kh/s and after quitting sgminer goes defuct and using 100% cpu
yacminer 3.5.0-yac2 - share above target

it doesnt seem to like older AMD cards
running them all with nfactor 6

cgminer-o

sr. member

Activity: 350

Merit: 250

1$ =1 MEC 2013 (2,56 $) 5$ =1MEC 2014 ?

Orrechorre

sr. member

Activity: 569

Merit: 250

how do I raise the diff so i can mine quark on a pool with gpu?

coda

full member

Activity: 212

Merit: 100

hi

just to say there's a new one in the X11 familly that looks neat: Limecoin
https://bitcointalksearch.org/topic/annlimlimecoin-x11-algo-obsolete-move-to-limecoinx-please-559534

xfelinho

member

Activity: 68

Merit: 10

Optimization for Groestl is possible

Quote from: cbuchner1 on April 05, 2014, 04:55:02 PM

Quote

we're working on a KILLER groestl, requiring a deeper understanding of the hash algo.
When done, everything else will be toast.

This would turn quark, X11, groestl, myriad-groestl into CUDA coins.

open sourcing it would be a problem. Last time we did this (heavycoin) the AMD guys caught up within a few days...
Christian

AlexGR

legendary

Activity: 1708

Merit: 1049

Hmm... but how much time is spent in Groestl?

Theoretically if one disables hash-by-hash, one can get a breakdown of how much time is consumed in each in order to see where the potential for speedup is.

restless

legendary

Activity: 1151

Merit: 1001

The real optimisation is possible for Groestl, atm its implementation is at least 10x slower than whats possible

pallas

legendary

Activity: 2716

Merit: 1094

Black Belt Developer

Quote from: AlexGR on April 09, 2014, 08:11:50 AM

Yes, I've seen bogus fluctuations of 1%, but the 4.25% for #pragma unroll 1 on the 8th "for i" is repeatable.

939kh is the baseline, I did #pragma unroll 1 in all "for" loops. I only gained with the 4th and 8th, but interestingly when doing them both I went <939. So I kept the 8th #pragma unroll 1 which is the fastest of all.

I had previously tried #pragma unroll but didn't find any speed. I also hardcoded easily computed values in simd.cl instead of having then been computed, and still didn't get any speedup.

I believe all this has to do with compiler optimizations, which usually can figure out how to best unroll or pre-compute values automatically, but sometimes it's better to do them manually. And results may vary by card, for example a simple optimization may work for a 7950 but lead to worse results on a r9 290.

Topic: [ANN] sph-sgminer: multi-coin multi-algorithm GPU miner | added MaruCoin - page 27. (Read 515718 times)