CCminer(SP-MOD) Modded GPU kernels.

[tsiv]

New build from github

http://www.filedropper.com/release10

Contains 2 bugfixes from the tpruvot (stratum disconnects)
X13 and x15 is a bit faster. (hamsi+8,3%)
fixed linuxbuild (untested)

The sourcecode is available here:

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

http://pastebin.com/sV7C2TWR

Try that for Hamsi. It sure is not pretty but it's a whole lot faster than the one I originally left X13 with. My notes say +59% on the 750 Ti

[SavellM]

Hey sp_ quick question.

I've just downloaded your Release10 (my first download of yours).
What is the difference between ccminer and ccminer12?

Also it says that its missing msvcr120.dll

I managed to copy this from tpruvot.

Is this correct?
Or do you have a better one to use?

[sp_]

ccminer12 is for compute 5.2 cards.  970/980. Since this is just a build from the github it doesn't contain the dependency files. You can download the files in the latest tvpovot release.

[SS2006]

what is the range of intensity parameter we can set?

[sp_]

New build from github
http://www.filedropper.com/release10
Contains 2 bugfixes from the tpruvot (stratum disconnects)
X13 and x15 is a bit faster. (hamsi+8,3%)
fixed linuxbuild (untested)
The sourcecode is available here:
https://github.com/sp-hash/ccminer

http://pastebin.com/sV7C2TWR
Try that for Hamsi. It sure is not pretty but it's a whole lot faster than the one I originally left X13 with. My notes say +59% on the 750 Ti

Yes, faster indeed! Thanks, I will add it.
Tested it to be 14.34% faster than my last version. (750ti)

[sp_]

what is the range of intensity parameter we can set?

try. max is 30, but the miner will probobly crash and spit out wild hashes. on the 970 20 or 21 @x11?

Default is 19

[sp_]

Release 10 works perfectly. No problems whatsoever.
Hashrates (750Ti O/C)
X11: 3000 kH/s
X13: 2320 kH/s
X15: 2000 kH/s

Congratulations, sp_. You have beaten the code I did for CUDA... I think a couple months ago now, on X11. I think all three.

You should help me push it up to 3,5 MHASH. Still some easy pickings.

[ryen123]

Release 10 works perfectly. No problems whatsoever.

Hashrates (750Ti O/C)
X11: 3000 kH/s
X13: 2320 kH/s
X15: 2000 kH/s

What are your clocks with the 750Ti?

I can get only 2950-2970KH/s with 1246MHz Core / 1580MHz Memory

1210MHz Core / 1600MHz Memory

@jpouza I set -i 20 and managed to get the figures above.

[sp_]

Yes, faster indeed! Thanks, I will add it.
Tested it to be 14.34% faster than my last version. (750ti)

Actually it was even faster with 64 registers But I guess your default already was 64 regs..
My default is 80, so I had to add  launchbounds to force 64 registers, and running only with 32 threads.

Then I get it to be 20% faster.

[sp_]

Commited tsivs hamsi, and removed memoryleaks in x11 and x13. Also merged in tvprovot's stratum bugfix.

[sp_]

@jpouza I set -i 20 and managed to get the figures above.

The default is set to -i=19 because -i=20 will cause out of memory on the 1GB cards (750,750ti) On 970/980 -i should be set to 20 or higher.

[Epsylon3]

http://pastebin.com/sV7C2TWR

Try that for Hamsi. It sure is not pretty but it's a whole lot faster than the one I originally left X13 with. My notes say +59% on the 750 Ti

Yep, i get much more with your code i keep it tx.

2250KHs on a 750 Ti, vs 2150 before changes, and slower with sp commit (tried differents maxregcount on linux)

edit: i used 72 registers, 64 is too low

[jpouza]

@jpouza I set -i 20 and managed to get the figures above.

The default is set to -i=19 because -i=20 will cause out of memory on the 1GB cards (750,750ti) On 970/980 -i should be set to 20 or higher.

-i 21 on 980 boosted almost 200KH/s!

-i 20 on 750Ti 2GB nothing changed, will try to increase memory clock.

EDIT: nothing, still can't reach 3MH/s with 750 Ti, even with 1600MHz mem clock and -i 20.
980 just amazing, 9,8 MH/s achieved with -i 21

WOW


subefotos

[sp_]

The best setting for the TI is betweeen 19 and 20, so I changed the default. Preparing a new build now. #11.
We should add xintensity to ccminer for optimal performance.

[sp_]

http://pastebin.com/sV7C2TWR
Try that for Hamsi. It sure is not pretty but it's a whole lot faster than the one I originally left X13 with. My notes say +59% on the 750 Ti

Yep, i get much more with your code i keep it tx.
2250KHs on a 750 Ti, vs 2150 before changes, and slower with sp commit (tried differents maxregcount on linux)
edit: i used 72 registers, 64 is too low

I use launchbound 256,4 and only 32 threads. Seems to be the fastest on windows. I get no memory spills with 64 registers for compute 5.0, but for compute 3.0 there are spills (because of the missing funnelshift)

[sp_]

Eplyson, the API seems to fail when you run two instances of ccminer. Probobly because the default port is used.

[sp_]

I reset the default throughput to be the same as in build 8, it increases the x11 hash with around 30KHASH (750ti). Now trying out 72 registers and no launchbound as suggested @x13. But so slow to build...

[Epsylon3]

I reset the default throughput to be the same as in build 8, it increases the x11 hash with around 30KHASH (750ti). Now trying out 72 registers and no launchbound as suggested @x13. But so slow to build...

I why i prefer to do tests on linux when i can... I work over nfs on windows and build in putty

Increasing too much intensity will make Windows very slow, its why i dont use the max possible as default. But i understand your ".5" intensity, could be done like that rounded to the nearest multiple of 256

for the api its normal... "-b 0" to fully disable it, or set a custom port if required... There is only 4 tries to bind the first minute, then stops. Nice also to know there is another instance running

[sp_]

yes, this is how the X-intensity is implemented in sgminer. Just Multiply the factor by 256.

-xi 128 = gives throughput of 128*256

or perhaps the constant is more than 256. Don't remember. But the idea is to multiply the constant with the inputparameter.

[sp_]

Buildt a new version with tsivs x13 improvements.
Some bugfixes, and bether default intensity for 750ti (I get +30 KHASH)
version 10 had a memory leak. If the miner was running for a long time, the performance went down.

The 11th public beta(windows executable) is available here: (11-25-2014)

http://www.filedropper.com/release11

The sourcecode is available here:

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