CCminer(SP-MOD) Modded GPU kernels.

[chrysophylax]

nice...
I can't get more than 6 working in win 8.1 (the same mobo had 7 amd cards running)

You probobly need a modded driver. I usually have 2-3 amd cards, and fill the rest of the slots with nvidia cards. Good for the powersupply as well.

ya, only thing that amd is left good at ...

This looks interesting. 16 GPU's on one motherboard using 4 PCI-e slots.

http://amfeltec.com/products/gpu-oriented-cluster/

But Might have to modify the driver...

EDIT: just checked the price.. £140 for pciE splitter 1->4 slots. Bether to buy another motherboard. Not worth it.

Yes...they look good. I posted that in the cuda forum a while back and I also contacted the seller. He wants a high price for them..but worth it if you have a work horse like the 980. But not worth it for the 750ti...

... and they wont work if your motherbard bios only allows a maximum of 8 gpus ...

clustering gpus only works with modded bios ... i have contacted them regarding this also - and the response is the same all round ... really gets up me that there is that limitation ...

IF you can get a high end motherboard ( like some of the top shelf supermicro motherboards ) that are designed for it - you can get 12 gpus on the one board ...

the units WILL connect and the os ( only linux as far as i know ) will show all the gpus - but they will not all mine unless you know how to mod ... it will top out at 8 gpus if you have standard motherboards ...

#crysx

[Dotcommie]

You should reflash the bios to the black edition. Then you will get 200+ KHASH with stock clocks.

you know sp - i think ill try that with one of the cards ...

you cant flash the lp ones can you ( low profile - non-powered ) ...

do you have a link I can get the bios from? ... ill check it out ...

that way - if it dies - its only one card :| ...

#crysx

Did you buy any blacks? You can get it off one of them if so.  I spent a few hours the other week looking for one but should have probably just pm'd someone who owned one.  I'm not used to not finding what I'm looking for on the all powerful google after that long so i just went to sleep instead 

[antonio8]

First feedback of GTX960 on Qubit - 7.6- 7.7 MH/s with Palit GTX960 Super Jetsteam...

x11?

Here you go: http://cryptomining-blog.com/4219-the-new-nvidia-geforce-gtx-960-mining-performance/

[Bombadil]

Sp, can you take a look at tpruvots latest additions? (KlausT has them too already)
Setting --cpu-priority 5 (sets it to realtime from command line whoohoo) gives a nice extra Even "highest" priority doesn't get this boost on windows.

[sp_]

just ordered a 960 card. Will probobly increase the performance on this with some finetuning..

[chrysophylax]

You should reflash the bios to the black edition. Then you will get 200+ KHASH with stock clocks.

you know sp - i think ill try that with one of the cards ...

you cant flash the lp ones can you ( low profile - non-powered ) ...

do you have a link I can get the bios from? ... ill check it out ...

that way - if it dies - its only one card :| ...

#crysx

Did you buy any blacks? You can get it off one of them if so.  I spent a few hours the other week looking for one but should have probably just pm'd someone who owned one.  I'm not used to not finding what I'm looking for on the all powerful google after that long so i just went to sleep instead 

im not sure if any of them are black edition ...

i do know they are all oc edition - even the non-powered ones ...

i will have a look tomorrow through all the boxes and see what they say ... sleep time now for me

if i dont - how can i make them without the blackbios? where do i get it from? ...

#crysx

[Epsylon3]

ordered a MSI 960 too (twin frozr 5), i sold my last "normal" 750 Ti (gigabyte) monday at 120€

[tbearhere]

release 31 is here:

- improved groestl512 part2. faster x11,x13,x14,x15,nist5,quark etc.
- merged branch with klaus_t.  Faster quark + other fixes.
- fixed crash on 960 cards. (default intensity)

1.5.31(sp-MOD) is available here: (18-jan-2015)

https://github.com/sp-hash/ccminer/releases/tag/1.5.31

The sourcecode is available here:

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

sp...do you have #31 with the fix... compiled? I get errors with the source code.  thx

[go6ooo1212]

First feedback of GTX960 on Qubit - 7.6- 7.7 MH/s with Palit GTX960 Super Jetsteam...

x11?

Ill test the other algos today or tomorrow and will give a feedback...

[flipclip]

release 31 is here:

- improved groestl512 part2. faster x11,x13,x14,x15,nist5,quark etc.
- merged branch with klaus_t.  Faster quark + other fixes.
- fixed crash on 960 cards. (default intensity)

1.5.31(sp-MOD) is available here: (18-jan-2015)

https://github.com/sp-hash/ccminer/releases/tag/1.5.31

The sourcecode is available here:

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

sp...do you have #31 with the fix... compiled? I get errors with the source code.  thx

I assume you are compiling for Linux?  The exact release 31 won't compile, you would have to pull from commit a855bbaf5d6c4c75cd06688a00171ca49bce6bd1, or edit the miner.h file manually from release 31.

[tbearhere]

release 31 is here:

- improved groestl512 part2. faster x11,x13,x14,x15,nist5,quark etc.
- merged branch with klaus_t.  Faster quark + other fixes.
- fixed crash on 960 cards. (default intensity)

1.5.31(sp-MOD) is available here: (18-jan-2015)

https://github.com/sp-hash/ccminer/releases/tag/1.5.31

The sourcecode is available here:

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

sp...do you have #31 with the fix... compiled? I get errors with the source code.  thx

I assume you are compiling for Linux?  The exact release 31 won't compile, you would have to pull from commit a855bbaf5d6c4c75cd06688a00171ca49bce6bd1, or edit the miner.h file manually from release 31.

No windows

[flipclip]

release 31 is here:

- improved groestl512 part2. faster x11,x13,x14,x15,nist5,quark etc.
- merged branch with klaus_t.  Faster quark + other fixes.
- fixed crash on 960 cards. (default intensity)

1.5.31(sp-MOD) is available here: (18-jan-2015)

https://github.com/sp-hash/ccminer/releases/tag/1.5.31

The sourcecode is available here:

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

sp...do you have #31 with the fix... compiled? I get errors with the source code.  thx

I assume you are compiling for Linux?  The exact release 31 won't compile, you would have to pull from commit a855bbaf5d6c4c75cd06688a00171ca49bce6bd1, or edit the miner.h file manually from release 31.

No windows

Oh, sorry about that.  Hadn't seen anyone report window compile errors, so I just assumed.

[tbearhere]

release 31 is here:

- improved groestl512 part2. faster x11,x13,x14,x15,nist5,quark etc.
- merged branch with klaus_t.  Faster quark + other fixes.
- fixed crash on 960 cards. (default intensity)

1.5.31(sp-MOD) is available here: (18-jan-2015)

https://github.com/sp-hash/ccminer/releases/tag/1.5.31

The sourcecode is available here:

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

sp...do you have #31 with the fix... compiled? I get errors with the source code.  thx

I assume you are compiling for Linux?  The exact release 31 won't compile, you would have to pull from commit a855bbaf5d6c4c75cd06688a00171ca49bce6bd1, or edit the miner.h file manually from release 31.

No windows

Oh, sorry about that.  Hadn't seen anyone report window compile errors, so I just assumed.

Thx anyhow ... sp usually gives the compiled for windows release.

[sp_]

soon time to build number 32.

[Bombadil]

soon time to build number 32.

I saw that you hardcoded the realtime priority. I'd advice against it: if for some reason the process loops endlessly, the complete rig will freeze up. Don't do realtime if you're not watching your rigs (and there are many like that around..)
So that's where the command line, like with KlausT and tpruvot comes in handy: --cpu-priority and --cpu-affinity

[sp_]

Submitted another groestl speedup@github. the 980 is up 200KHASH in quark.

[sp_]

-Less cpu usage, more GPU. x11 down from 20% cpu to 1%
-faster groestl(quark, x11, x13...), mostly on 970/980

nist5 is broken

1.5.32(sp-MOD) is available here: (23-jan-2015)

https://github.com/sp-hash/ccminer/releases/tag/1.5.32

The sourcecode is available here:

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

[tbearhere]

soon time to build number 32.

I saw that you hardcoded the realtime priority. I'd advice against it: if for some reason the process loops endlessly, the complete rig will freeze up. Don't do realtime if you're not watching your rigs (and there are many like that around..)
So that's where the command line, like with KlausT and tpruvot comes in handy: --cpu-priority and --cpu-affinity

The commands    --cpu-priority and --cpu-affinity  .... what exactly do they do? It looks self explanatory, but?  thx

[Bombadil]

soon time to build number 32.

I saw that you hardcoded the realtime priority. I'd advice against it: if for some reason the process loops endlessly, the complete rig will freeze up. Don't do realtime if you're not watching your rigs (and there are many like that around..)
So that's where the command line, like with KlausT and tpruvot comes in handy: --cpu-priority and --cpu-affinity

The commands    --cpu-priority and --cpu-affinity  .... what exactly do they do? It looks self explanatory, but?  thx

They set the process priority and affinity Realtime priority (5) will grant full priority of your miner over any other process on your windows. This boosts your speed a lot, I've seen quark go up by +200kh by using realtime priority.
Setting affinity will lock it to certain core(s), so the threads don't hop around too much. If you grant realtime priority to a process, it could lock/freeze up your system (nothing else can kill it), but if you assign it to only 1 out of 4 cores, other processes (like taskmgr) can run on the other 3 cores and kill it, if necessary.
Also, if you run cpuminers next to ccminer, or something else that's cpu intensive like av scan, and they run on the same core(s), they'll interfere with eachother. You don't want to lose 200kh quark per 750ti because some funny CPU only coin you're mining next to it
I've also seen on some cpuminers/algos that setting up a different process per CPU core, and locking them to an assigned core, they perform much better. This doesn't mean it will necessarily boost ccminer speed (at the moment), but it doesn't hurt it at all to say the least. I'd like it mostly because it's much easier to manage your low-level-hardware miners.

[chrysophylax]

soon time to build number 32.

I saw that you hardcoded the realtime priority. I'd advice against it: if for some reason the process loops endlessly, the complete rig will freeze up. Don't do realtime if you're not watching your rigs (and there are many like that around..)
So that's where the command line, like with KlausT and tpruvot comes in handy: --cpu-priority and --cpu-affinity

The commands    --cpu-priority and --cpu-affinity  .... what exactly do they do? It looks self explanatory, but?  thx

yes please - im interested in excatly what they do also ...

#crysx