CCminer(SP-MOD) Modded GPU kernels.

[chrysophylax]

With BTC price = trash, nothing is profitable anymore.

This is correct

its THIS that keeps it dead ...

there was a time when 'the market' ( ie 'the people' ) actually believed that btc was its OWN monetary unit and NOT a commodity for the government controlled fiat currencies ...

a time when btc was its OWN market - its OWN system - its OWN way of dealing selling buying and profiting ...

oh - hang on ... that time is still nigh ...

as i posted on @cryptobucket ...

no bloody wonder satoshi went underground ...

people have just 'lost' the whole ideal and intention that btc was built in the first place ...

ho hum ... the governments have won ...

for now ...

#crysx

[chrysophylax]

Above 0,2 BTC per day for 1 gigahash.

you only need 60 980 cards (10 rigs), or 180 750TI Cards(30 rigs) to reach 1 gigahash.

$)

ok - 30 rigs - here we come ...

#crysx

[chrysophylax]

I was only getting about 8 Mh/s effective hashrate at the pool with r24 as opposed to 10.5 Mh/s with r23.  I'll be sticking with r23 for now.

those hashrates are superb ... but as sp works on the miner and optimizations - there is bound to be some mishaps along the way ...

im still runing the latest git pull - and will stay on it for now - to test stability also ...

i too have issues with extranonce2 errors though ... westhash is the main one ... yaamp is down at the moment - but was not getting too many errors there ...

have you overclocked the cards to get those hashrates? ... if so - how? ... assuming you are using linux too as per your post ...

tanx ...

#crysx

Those hashrates are for two 750Ti's.. sorry if I wasn't clear about that!
10.5M/2 = 5250 khash/s effective at the pool (YAAMP) for quark. I get about 5550 +/- 50 khash/s reported from the miner itself.

still very nice ...

sp - you are doing an awesome job mate ...

#crysx

[chrysophylax]

I say effective hashrate as opposed to "actual" hashrate.  Rejects are low.  It's just a matter of how hashrates are calculated by the pool.  Multipool explains it well: "The hashrate your miner displays is its own internal hashrate for calculating difficulty 1 shares, and thus is fairly meaningless to the pool. The pool displays an average hashrate calculated based on your valid, accepted shares per minute."
5250/5550 = ~95% which isn't bad considering it's connected through a stratum proxy for leasing.

I would say that is acceptable. If the network connection is slow on the pool, the numbers get lower. And you get some Job not found rejects.
I found a problem in the code now when the hashcounter sum get's higher than 2^32. for a 16000MHASH rig, the rig will hash fine for 5 mins, and then might run into trouble. Testing now at suchpool.pw.

agreed - acceptable in my eyes also ...

the issue doesnt seem to be a problem here ... maybe im not connected long enough to notice it for any period of time? ...

#crysx

[chrysophylax]

fixed it now

No GPU codechanges in this version from build 24

- fixed command output in benchmark mode
- fixed hashrate on some pools. coinmine.pl, suchpool.pw etc.

The 25th public beta(windows executable) is available here: (05-jan-2015)

http://www.filedropper.com/release25

The sourcecode is available here:

https://github.com/sp-hash/ccminer

pull - compile - test ...

see how we go ...

#crysx

[scryptr]

SP_hash v25--

I carefully cloned from git on both my 750ti rigs.  My faster rig has been running version v24 with no problem, no crashes.  I reverted my slower rig to tpruvot v1.5.2-git after several crashes, and it has run over 24 hours without rebooting or restarting. 

I restarted both rigs with v25 a few minutes ago.  The slower rig clocks in initially at 18,045kh/s, and the fast rig clocks in initially at 18,445kh/s.  However, while I was writing this, the slower rig dropped to ~300kh/s.  I will again revert this rig to  tpruvot v1.5.2-git, but will wipe the disk and reload the OS tomorrow.  The faster rig is clocking along nicely as before.

I am considering upgrading the CPU on the slower rig.  The systems are nearly identical, except for the card models and AMD CPU models.

Also, the 10 second delay on shutdown (thread closure) was noticable when I switched between ccminer versions.

--scryptr

[tbearhere]

My rig went down total 30kh/s quark but my 3 cards on this computer went up 30kh/s.

[sp_]

My rig went down total 30kh/s quark but my 3 cards on this computer went up 30kh/s.

Down on the pool? try to restart the miner. The result vil vary abit for every run. I think it has to do with alligment of buffers but I am not sure. 10/5500 is 0.02%

[tbearhere]

My rig went down total 30kh/s quark but my 3 cards on this computer went up 30kh/s.

Down on the pool? try to restart the miner. The result vil vary abit for every run. I think it has to do with alligment of buffers but I am not sure. 10/5500 is 0.02%

I tried X11 on my rig asrock 81 got 18,406 kh/s
The pool is down for quark. Will try again later.
EDIT: also quark algo does jump around a little.
Off topic but 2 months ago they were doing 300 btc a day 1-1-15 they started pumping out 4,000 a day 2 a minute ...the big drop in btc price maybe and panic sale.

[jpouza]

fixed it now

No GPU codechanges in this version from build 24

- fixed command output in benchmark mode
- fixed hashrate on some pools. coinmine.pl, suchpool.pw etc.

The 25th public beta(windows executable) is available here: (05-jan-2015)

http://www.filedropper.com/release25

The sourcecode is available here:

https://github.com/sp-hash/ccminer

I will test it as soon I get home.

Cheers

[jpouza]

With BTC price = trash, nothing is profitable anymore.

This is correct

its THIS that keeps it dead ...

there was a time when 'the market' ( ie 'the people' ) actually believed that btc was its OWN monetary unit and NOT a commodity for the government controlled fiat currencies ...

a time when btc was its OWN market - its OWN system - its OWN way of dealing selling buying and profiting ...

oh - hang on ... that time is still nigh ...

as i posted on @cryptobucket ...

no bloody wonder satoshi went underground ...

people have just 'lost' the whole ideal and intention that btc was built in the first place ...

ho hum ... the governments have won ...

for now ...

#crysx

I´m so disapointed, hopeless, now that Bitstamp get "hacked" as they say (Mt.Gox #2?), BTC price crashed, not in a good wave in 2015.

I think sooner or later things will change, to mature or simple to disappear for BTC.

[djm34]

With BTC price = trash, nothing is profitable anymore.

This is correct

its THIS that keeps it dead ...

there was a time when 'the market' ( ie 'the people' ) actually believed that btc was its OWN monetary unit and NOT a commodity for the government controlled fiat currencies ...

a time when btc was its OWN market - its OWN system - its OWN way of dealing selling buying and profiting ...

oh - hang on ... that time is still nigh ...

as i posted on @cryptobucket ...

no bloody wonder satoshi went underground ...

people have just 'lost' the whole ideal and intention that btc was built in the first place ...

ho hum ... the governments have won ...

for now ...

#crysx

I´m so dissapointed, hopeless, now that Bitstamp get "hacked" as they say (Mt.Gox #2?), BTC price crashed, not in a good wave in 2015.

I think sooner or later things will change, to mature or simple to disappear for BTC.

they also said, they will reimburse back... (and btc went down before the announcement)

[sp_]

found an issue in build 25 when blocks changes often.. More rejects. New fix comming today.

[Handofmidas]

Tested release25 now, I think we're getting there! Hashrate is now recognized correctly as far as I can tell, fluctuates around the real hashrate as it should. Awesome job SP, here's the graph:

[sp_]

More hashrate is comming. I have optimized blake 25% faster.

[sp_]

Submitted the optimized blake kernals to gihub.(25% faster)

I had to split the kernals and implement 2 seperate implementations. one for the x series. x11,x13,x14,x15,x17 and one for the others, quark, nist5, jackpotcoin etc. In the x'series implementation I moved some precalculation  from const mem into the instructioncache. I also improved the uint2 implementation.


Building release 26

results on 750ti: (up from release 25)

x11 +30-50 KHASH
quark +50-100 KHASH


970 and 980 untested. Please test if you have time

[sp_]

The 26th public beta(windows executable) is available here: (07-jan-2015)

http://www.filedropper.com/release26

The sourcecode is available here:

https://github.com/sp-hash/ccminer

[squadbox]

The 26th public beta(windows executable) is available here: (07-jan-2015)

http://www.filedropper.com/release26

The sourcecode is available here:

https://github.com/sp-hash/ccminer

I'm not sure if you caught my earlier message about releases.  When you made the r26 commit (86dd73f) and updated the version numbers in cpuminer-config.h it only reflects this change for windows builds.  The first line of configure.ac needs to also be changed for linux builds.

[Dotcommie]

Submitted the optimized blake kernals to gihub.(25% faster)

I had to split the kernals and implement 2 seperate implementations. one for the x series. x11,x13,x14,x15,x17 and one for the others, quark, nist5, jackpotcoin etc. In the x'series implementation I moved some precalculation  from const mem into the instructioncache. I also improved the uint2 implementation.


Building release 26

results on 750ti: (up from release 25)

x11 +30-50 KHASH
quark +50-100 KHASH


970 and 980 untested. Please test if you have time

Didn't test much, but I think my 2 970s were getting about 13000 each.  Doesn't seem like they're doing as well as they could, but then again I haven't really mined quark on them before so i don't have previous numbers to compare them to.

[sp_]

I tested version 23 on the 980 and it made 15.5 MHASH on stock clocks. but build 26 should be faster.
13MHASH@970 is very good.

Sgminer does 5.6MHASH on the AMD 290x