CCminer(SP-MOD) Modded GPU kernels.

[chrysophylax]

Wich algorithm are you mining?  76020 KH/s is high. What is the temps on the card? (use gpuz)

Haha that's not the real hash (I wish!!) - it seems the last figure it spits out is an error - it hashes OC around 10MHz+ - but then falls over. I can see quite a few people here are getting the same error. Temps fine just before it crashes - it's a hefty system custom built nothing goes over 71C during gaming or normal mining - and it's currently 35C in Aus and humid so temps not the issue.

it maybe your clocks ... lower them a little at a time and see if that helps ...

im in adelaide and it gets very hot here most days ( apart from the last few ) and i had major issues recently with the other cards i had ( amd cards ) with overclocking and crashing ...

lowered the clocks and they all worked like a charm - and have since replaced everything with 750ti oc - stock clocks ...

they run much more smooth than the amd's ...

btw - thats with most algo's ... some cards just refuse to start if the clocking is too high ...

#crysx

[1714688174]

1714688174

[1714688174]

1714688174

[1714688174]

1714688174

[bluejam00]

ok will try 10-20 increments with and also with the -i setting - and see how we go. Actually never thought this - will try say 10MHz+ on mem and gpu - if that crashes then we should know even the smallest OC on a stock cards kills it? Never thought of doing that first!

I also heard some stock cards (mines the Geforce) - stops any overclocking so you need to hack the gpu bios first.

Someone else here was talking about games using P1 or P2 setting but the miner only accessing P0 so you need to up this rather than how you would OC a card for gaming. I need to research more on this as I'm noob on maxwell GPU's to be fair!

Wich algorithm are you mining?  76020 KH/s is high. What is the temps on the card? (use gpuz)

Haha that's not the real hash (I wish!!) - it seems the last figure it spits out is an error - it hashes OC around 10MHz+ - but then falls over. I can see quite a few people here are getting the same error. Temps fine just before it crashes - it's a hefty system custom built nothing goes over 71C during gaming or normal mining - and it's currently 35C in Aus and humid so temps not the issue.

it maybe your clocks ... lower them a little at a time and see if that helps ...

im in adelaide and it gets very hot here most days ( apart from the last few ) and i had major issues recently with the other cards i had ( amd cards ) with overclocking and crashing ...

lowered the clocks and they all worked like a charm - and have since replaced everything with 750ti oc - stock clocks ...

they run much more smooth than the amd's ...

btw - thats with most algo's ... some cards just refuse to start if the clocking is too high ...

#crysx

[chrysophylax]

ok will try 10-20 increments with and also with the -i setting - and see how we go. Actually never thought this - will try say 10MHz+ on mem and gpu - if that crashes then we should know even the smallest OC on a stock cards kills it? Never thought of doing that first!

I also heard some stock cards (mines the Geforce) - stops any overclocking so you need to hack the gpu bios first.

Someone else here was talking about games using P1 or P2 setting but the miner only accessing P0 so you need to up this rather than how you would OC a card for gaming. I need to research more on this as I'm noob on maxwell GPU's to be fair!

Wich algorithm are you mining?  76020 KH/s is high. What is the temps on the card? (use gpuz)

Haha that's not the real hash (I wish!!) - it seems the last figure it spits out is an error - it hashes OC around 10MHz+ - but then falls over. I can see quite a few people here are getting the same error. Temps fine just before it crashes - it's a hefty system custom built nothing goes over 71C during gaming or normal mining - and it's currently 35C in Aus and humid so temps not the issue.

it maybe your clocks ... lower them a little at a time and see if that helps ...

im in adelaide and it gets very hot here most days ( apart from the last few ) and i had major issues recently with the other cards i had ( amd cards ) with overclocking and crashing ...

lowered the clocks and they all worked like a charm - and have since replaced everything with 750ti oc - stock clocks ...

they run much more smooth than the amd's ...

btw - thats with most algo's ... some cards just refuse to start if the clocking is too high ...

#crysx

with this - its about trial and error and patience ... all cards work differently due to the state of the manufacturing of the cards ...

not all cards are the same and they certainly dont get manufactured with EXACTLY the same 'bits' ... this means that some cards can handle more than others ...

with different chipsets and BIOS and RAM - there are a number of things that will make the cards react differently to different situations ...

i also use linux ( fedora 19 x64 ) and to the best of my knowledge - there is no real way of overclocking via console / os level apart from tweaking the card via BIOS and other reprogrammed firmware ...

as easy as it is to do this with some cards - i dont play with that side of it ...

my agenda is very simple - and thats keeping the cards as stock with stock clocks and running inside systems so they can be resold when another upgrade becomes necessary - or i fold the whole project and rebuild these machines into personal computers ( yes - thats why i use fedora ) ... but thats another story ...

windows is a different story altogether - as msi afterburner is one of the pieces of software that is build for tweaking cards amongst others ...

always start from the bottom and work your way up - stepping each overclock in small increments and letting it run ( if in fact it does run ) for at least 15mins initially to see if its stable at those clocks ... the moment it crashes - fall back and then run for a much longer period to test for stability / heat / hashrate continuity ...

sp has built quite a solid product out of ccminer - and is always improving it ...

noob? ... hehehe - this whole industry are noobs ... crypto hasnt been around long mate - and honestly - there arent any 'real' veterans out there - unless satoshi comes out of the dark again ...

dont get me wrong - the industry is full of very talented people ... look at most of the guys here ...

but at only a few years old since its inception - we all have a lot of developing to do ...

and i dont mean just the software ...

#crysx

[Zels]

My 750ti looks to be stable with Oc of +200core / +250Mem

Running since 30hours and still didn't crash 
x11", "hashrate": 3200
x13", "hashrate": 2540
x14", "hashrate": 2480
x15", "hashrate": 2200
nist5", "hashrate": 9650
quark", "hashrate": 6050

No drop in hashrate from +400mem to +250mem, just that it doesn't crash driver for now

[sp_]

Checked in groestl speedup.

Faster groestl part 1. quark+150KHASH, x11 +60KHASH (750ti)

I managed to shrink the method "to_bitslice_quad" from around 800 asm instructions to around 80.

and "from_bitslice_quad" from around 400 instructions to around 200 instructions. With some more work, I will shrink this to 80 as well.

Instead of calculating one bit at a time I use the whole register in the cpu. Similar to a chunky2planar convertion.

[sp_]

Added my first binary to github

1.5.30(sp-MOD) is available here: (15-jan-2015)

https://github.com/sp-hash/ccminer/releases/tag/1.5.30

The sourcecode is available here:

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

[sp_]

I will be taking a break now and work on the private spreadcoin miner. If you want more hash, please donate some BTC thanks.

[tbearhere]

Checked in groestl speedup.

Faster groestl part 1. quark+250KHASH, x11 +50KHASH (750ti)

I managed to shrink the method "to_bitslice_quad" from around 800 asm instructions to around 80.

and "from_bitslice_quad" from around 400 instructions to around 200 instructions. With some more work, I will shrink this to 80 as well.

Instead of calculating one bit at a time I use the whole register in the cpu. Similar to a chunky2planar convertion.

Good one sp..getting about 100+ to 150+ on quark...don't have time yet to check other algo's. 

[flipclip]

Quark:
v28= ~11,160 kh/s
v30= ~11,441 kh/s

x11:
v28= ~5,730 kh/s
v30= ~5,850 kh/s

lyra2:
v28= ~1,350 kh/s
v30= ~1,350 kh/s

(2 750Ti's, no overclock)

[chrysophylax]

I will be taking a break now and work on the private spreadcoin miner. If you want more hash, please donate some BTC thanks.

tanx sp ...

im pulling the latest from git - compiling - and then as of tomorrow ( adelaide australia time ) ill mine your address as donation for 48 hours with the upgraded miners on ccminer ...

any algo you want to be mining with? on yaamp? ...

btw - how do can i be included in the private project for spreadcoin - even if its just for testing? or at all? ...

i would really like to see how it runs so far ...

tanx ...

#crysx

[flipclip]

... ill mine your address as donation for 48 hours with the upgraded miners on ccminer ...

any algo you want to be mining with? on yaamp? ...

#crysx

I'd hazard a guess, the most profitable  

[tbearhere]

Looks like pool rejects are higher making #30 less efficient then #29. quark

wrong its good 

[jpouza]

I will be taking a break now and work on the private spreadcoin miner. If you want more hash, please donate some BTC thanks.

Nice, I've sent you a PM.

Cheers

[flipclip]

Looks like pool rejects are higher making #30 less efficient then #29. quark

how many rejects and which pool?  I was on yaamp for 1.5 hours, with one reject ("reject reason: Job not found" right after a block change), which seemed fine to me.

[sp_]

Quark:
v28= ~11,160 kh/s
v30= ~11,441 kh/s
x11:
v28= ~5,730 kh/s
v30= ~5,850 kh/s
lyra2:
v28= ~1,350 kh/s
v30= ~1,350 kh/s
(2 750Ti's, no overclock)

Try groestl or diamondgroestl.  

You can see on the commit on github that this is groestl speedup part 1. I have part 2 soon ready for checkin, but it is currently mixing the bits, and produce wrong results. Lyra is not using the bitslice groestl (killer groestl)
so my improvements will not have an effect. But I guess if I swap the implementation, Lyra2 will get a boost as well. Did you try that DJM34?

About the spreadcoinminer:

I will integrate and optimize TSIV's spreadcoin implementation into the latest fork of ccminer. I will send out beta versions for a fee of 0.1 BTC. The beta will be a windows executable.
I might send out more than one exe in the testing phase, if I manage to optimize more.
I will publish the sourcecode after one month of betatesting. (When I publish the sourcecode, the spreadcoin will spread bether, and secure the coin)
 
The current speed will be announced when the exefile is released. Hopefully next weekend 25-26 january.

The estimated speed increase is 30-40% on the 980 cards

If you want a early seat, you can start donating to my BTC  or DRK adress in my signature.The current speed will be announced when the exefile is released. Hopefully next weekend 25-26 january.

[tbearhere]

Looks like pool rejects are higher making #30 less efficient then #29. quark

how many rejects and which pool?  I was on yaamp for 1.5 hours, with one reject ("reject reason: Job not found" right after a block change), which seemed fine to me.

wrong its good ....great   sorry

[chrysophylax]

Quark:
v28= ~11,160 kh/s
v30= ~11,441 kh/s
x11:
v28= ~5,730 kh/s
v30= ~5,850 kh/s
lyra2:
v28= ~1,350 kh/s
v30= ~1,350 kh/s
(2 750Ti's, no overclock)

Try groestl or diamondgroestl.  

You can see on the commit on github that this is groestl speedup part 1. I have part 2 soon ready for checkin, but it is currently mixing the bits, and produce wrong results. Lyra is not using the bitslice groestl (killer groestl)
so my improvements will not have an effect. But I guess if I swap the implementation, Lyra2 will get a boost as well. Did you try that DJM34?

About the spreadcoinminer:

I will integrate and optimize TSIV's spreadcoin implementation into the latest fork of ccminer. I will send out beta versions for a fee of 0.1 BTC. The beta will be a windows executable.
I might send out more than one exe in the testing phase, if I manage to optimize more.
I will publish the sourcecode after one month of betatesting. (When I publish the sourcecode, the spreadcoin will spread bether, and secure the coin)
 
The current speed will be announced when the exefile is released. Hopefully next weekend 25-26 january.

The estimated speed increase is 30-40% on the 980 cards

If you want a early seat, you can start donating to my BTC  or DRK adress in my signature.The current speed will be announced when the exefile is released. Hopefully next weekend 25-26 january.

any possibility of a linux x64 version for beta as well sp? ...

ill donate ( and i assume most will ) for the cause of improving the spreadcoin miner ... but a windows version is useless to me ...

anyway around that - or are you just keeping for the windows based systems? ...

it just simply means that 30 days of no testing on my end ...

#crysx

[sp_]

The new gtx 960 might fail on the default settings because the intensity is set to high for compute 5.2 devices. How much memory are they planning to ship with the new cards?

The fix will be to set the intensity manually with the -i parameter. f.eks -i 19

[djm34]

Quark:
v28= ~11,160 kh/s
v30= ~11,441 kh/s
x11:
v28= ~5,730 kh/s
v30= ~5,850 kh/s
lyra2:
v28= ~1,350 kh/s
v30= ~1,350 kh/s
(2 750Ti's, no overclock)

Try groestl or diamondgroestl.  

You can see on the commit on github that this is groestl speedup part 1. I have part 2 soon ready for checkin, but it is currently mixing the bits, and produce wrong results. Lyra is not using the bitslice groestl (killer groestl)
so my improvements will not have an effect. But I guess if I swap the implementation, Lyra2 will get a boost as well. Did you try that DJM34?

I tried a bit (I was on a rather tight schedule) but without success, the problem the quad implementation was written for groestl512 and using it for groestl256 isn't really straight forward...

[flipclip]

Quark:
v28= ~11,160 kh/s
v30= ~11,441 kh/s
x11:
v28= ~5,730 kh/s
v30= ~5,850 kh/s
lyra2:
v28= ~1,350 kh/s
v30= ~1,350 kh/s
(2 750Ti's, no overclock)

Try groestl or diamondgroestl.  

You can see on the commit on github that this is groestl speedup part 1. I have part 2 soon ready for checkin, but it is currently mixing the bits, and produce wrong results. Lyra is not using the bitslice groestl (killer groestl)
so my improvements will not have an effect. But I guess if I swap the implementation, Lyra2 will get a boost as well. Did you try that DJM34?

I realized Lyra2 wasn't part of the speed up, it just happened that the algo was profitable for like five minutes on yaamp, and I happen to be in front of my computer while it happened.  Since I hadn't posted any Lyra2 rates in a while (ever?) thought I would do it, just in case someone was interested.