Topic: [ANN] cudaMiner & ccMiner CUDA based mining applications [Windows/Linux/MacOSX] - page 356. (Read 3426938 times)
June 13, 2014, 08:00:01 PM
Any chance for some windows release?
June 13, 2014, 07:58:49 PM
thx sir
Posted from Bitcointa.lk - #9ZVzh3DcIJnT6NKU
Posted from Bitcointa.lk - #9ZVzh3DcIJnT6NKU
June 13, 2014, 07:49:18 PM
So, does this seem correct for one 750ti?:
-X11:2.6MH/s
-Jackpot:5.5MH/s
-NIST5: 8.4MH/s
-Quark: 4.7MH/s
-Groestl: 8MH/s (!!!!!! was 3.1MH/s before !!!!!)
I used these numbers to generate following list for one 750ti: http://pastebin.com/xcFqtX9E
Still old hashrates for X13. Want to see Christian's magic first
Again, this list is only a snapshot for this moment, with current prices and difficulties, so lower your own estimates.
-X11:2.6MH/s
-Jackpot:5.5MH/s
-NIST5: 8.4MH/s
-Quark: 4.7MH/s
-Groestl: 8MH/s (!!!!!! was 3.1MH/s before !!!!!)
I used these numbers to generate following list for one 750ti: http://pastebin.com/xcFqtX9E
Still old hashrates for X13. Want to see Christian's magic first
Again, this list is only a snapshot for this moment, with current prices and difficulties, so lower your own estimates.
June 13, 2014, 07:43:52 PM
June 13, 2014, 07:43:41 PM
interesting:
780ti on jpc is doing now 11.5MHash and on nist5 17MHash/s
750ti 5.5MHash 8.2MHash/s
Actually the gap between the 2 cards also decreases (before it was around 2.5 now it is getting close to 2)
780ti on jpc is doing now 11.5MHash and on nist5 17MHash/s
750ti 5.5MHash 8.2MHash/s
Actually the gap between the 2 cards also decreases (before it was around 2.5 now it is getting close to 2)
June 13, 2014, 07:36:41 PM
Jackpotcoin with 5x750Ti Linux, stock clocks:
old ccminer: 19.5MH @ 310W
new ccminer: 27.5MH @ 350W
so it's 41% more hashpower for 13% power usage - nice
old ccminer: 19.5MH @ 310W
new ccminer: 27.5MH @ 350W
so it's 41% more hashpower for 13% power usage - nice
June 13, 2014, 07:36:16 PM
2 770's and an 670...
Groestl:
Before: 17MH
After: 24MH
NIST5:
Before: 20MH
After: 24.5MH
Groestl:
Before: 17MH
After: 24MH
NIST5:
Before: 20MH
After: 24.5MH
June 13, 2014, 07:28:55 PM
Getting an extra ~1Mhash/s per 780ti card mining x11 (from 4.5Mhash/s to 5.5Mhash/s) with v1.1. A nice little bump
June 13, 2014, 07:27:34 PM
so I guess now jpc algo is using a lot more groestl
Previously Groestl was also eliminated in the 1st and 2nd rounds, leaving only 1/4th
of the hashes to be evaluated.
Now with the faster Groestl I made it run Groestl in the 2nd round too. So half the
hashes are fully evaluated now. It's faster in the end - this is what counts.
Christian
I'm <3'ing it, thank you so much, Christian! You're god
June 13, 2014, 07:26:47 PM
And I have it compiled and ready for everyone to try out: https://mega.co.nz/#!FFU2DYbZ!UT3LyvrLblyA81SZmtZmMv3vwelOe70JCTPrzNfBAoc
Please, still note, this is an unofficial compile, don't blame me for bugs
I'm collecting hashrates and will do one of my profit calc lists
Please, still note, this is an unofficial compile, don't blame me for bugs
so I guess now jpc algo is using a lot more groestl
What about Talkcoin he I'm collecting hashrates and will do one of my profit calc lists
https://www.virustotal.com/en/file/9d232b4bf2659251312c4ba1d182ec93d43435e6de49dbda045d68a735c15704/analysis/1402705533/
1 false positive so this is a good one.
Thanks!
June 13, 2014, 07:26:39 PM
What about Talkcoin he
ah yes, NIST5 includes Groestl, too. I amended my posting.
June 13, 2014, 07:21:36 PM
so I guess now jpc algo is using a lot more groestl
Previously Groestl was also eliminated in the 1st and 2nd rounds, leaving only 1/4th
of the hashes to be evaluated.
Now with the faster Groestl I made it run Groestl in the 2nd round too. So half the
hashes are fully evaluated now. It's faster in the end - this is what counts.
Christian
June 13, 2014, 07:19:25 PM
And I have it compiled and ready for everyone to try out: https://mega.co.nz/#!FFU2DYbZ!UT3LyvrLblyA81SZmtZmMv3vwelOe70JCTPrzNfBAoc
Please, still note, this is an unofficial compile, don't blame me for bugs
I'm collecting hashrates and will do one of my profit calc lists
Please, still note, this is an unofficial compile, don't blame me for bugs
so I guess now jpc algo is using a lot more groestl
What about Talkcoin he I'm collecting hashrates and will do one of my profit calc lists
June 13, 2014, 07:16:30 PM
Hello, I was trying of Groestlcoin with ccminer50.exe on my 750ti, I looked at the configs for 750ti on cudminer.cc
ambiohazardccminer - v1.02groestlccminer50.exe -q -s 1 -a groestlStock | Stock | Stock8800kh but I am getting no where near that hashrate , is this correct or am I doing something wrong.
Posted from Bitcointa.lk - #YahgoukViw0GWS07
ambiohazardccminer - v1.02groestlccminer50.exe -q -s 1 -a groestlStock | Stock | Stock8800kh but I am getting no where near that hashrate , is this correct or am I doing something wrong.
Posted from Bitcointa.lk - #YahgoukViw0GWS07
June 13, 2014, 07:15:51 PM
so I guess now jpc algo is using a lot more groestl
June 13, 2014, 07:07:38 PM
Looking pretty nice
June 13, 2014, 07:07:03 PM
Seems like you've already got rid of the MAXWELL_OR_FERMI constant, superb 
yes, the table based Groestl is history, and so are the architecture dependent optimizations we made for it.
Bitslice, baby! That's what it's doing now. More use of ALUs, much less use of the memory controller.
Getting this to run was about 2 weeks of research and implementation. And then some more making the
Compute 3.0 optimized Quad version you're enjoying now.
June 13, 2014, 07:05:32 PM
Congratz to all who made these new improvements... 
June 13, 2014, 07:02:19 PM
Aaand it's out: https://github.com/tsiv/ccminer
Should be an easy modification, right now my X13 is using the same code for the first 11 algos as the original X11. At the simplest it just takes a recompile after throwing in the new groestl version.
I did end up messing with the original echo hash, it seemed to have an extraneous xor that made the full 512 bit hash come out wrong. Then using that result for the additional hamsi+fugue hashes kinda ends badly. The funny thing is that it worked fine for X11 because it was the last hashing algo in the chain and the X11 algo only takes the first 256 bits of the hash, discarding the rest and the flipped bit ended up in the discarded part. Was a bitch to track down for sure.
Should be an easy modification, right now my X13 is using the same code for the first 11 algos as the original X11. At the simplest it just takes a recompile after throwing in the new groestl version.
I did end up messing with the original echo hash, it seemed to have an extraneous xor that made the full 512 bit hash come out wrong. Then using that result for the additional hamsi+fugue hashes kinda ends badly. The funny thing is that it worked fine for X11 because it was the last hashing algo in the chain and the X11 algo only takes the first 256 bits of the hash, discarding the rest and the flipped bit ended up in the discarded part. Was a bitch to track down for sure.
Hi, let me take the time to congratulate you for this excellent piece of work. The fact that you were able to debug the Echo hash proves that you are developing absolutely at the same level as the C&C Hash factory at its best.
Which is now unfortunately the "Cash factory" because Chris H (C. H.) just recently dropped out so he could focus more on his business.
I salute you, noble contributor!!!
Now back to making a killing with XMR and BBR , sorry
June 13, 2014, 07:00:52 PM
Killer hangover is cured, so I added Killer Groestl in the quad version. This cudaminer revision is
source code only and called V1.1
the following algos should now mine faster
-quark
-anime
-x11
-jackpot
-groestl
-myr-gr
Tomorrow I will try to get the X13 algo from the forked version back into my code and call it V1.2,
including a binary release.
Would you be so kind to test this V1.1 release out and report the speed-ups you're getting?
I've only tried Windows compilation - will the Linux version still build fine?
What other patches would you guys like to have included in V1.2 or V1.3?
Christian
Jump to: