CCminer(SP-MOD) Modded NVIDIA Maxwell / Pascal kernels. - page 1184.

Dotcommie

sr. member

Activity: 476

Merit: 250

Quote from: scryptr on January 17, 2015, 01:12:50 PM

Quote from: sp_ on January 17, 2015, 05:32:10 AM

I have now fixed the groestl speedup 2 on github. Klaus_t has been optimizing quark, so I am waiting for a pull request, and then build release 31.

SP_ BUILD v30a/b --

Version 30b is faster yet. Both my rigs see an increase of about 1200-1300kh/s mining quark. This difference is from v29 to v30b.

However, in the long-haul, 24hrs or so, build 30a/b is not stable on my slower rig. Again, it is crashing.

I hope v31 comes soon, and with smooth running code. My fast rig is now at 37.5mh/s, my slow rig at 36.2mh/s, and my fastest 750ti is at 6340kh/s (quark). --scryptr

EDIT: I've reverted my slower rig to v29 after a third crash. --scryptr

If you're getting that on even your slow rig I'm guessing you're overclocking them too much. My 6xEVGA FTW cards can't be OC'd more than 10mhz over stock OC without crashing eventually and it's been that way since the beginning. They get 36.55mh so I think you're pushing them too much. I clocked the memory to -150 since it really has no effect on a lot of algos besides wasting power. Try dropping mem clocks to stock maybe?

My slow rig of Gigabyte 750ti's only get 33mh and it's a clone of the other rig with an even better i3 cpu, so I'm not sure why your numbers are quite a bit more than mine besides maybe OC. Haven't tried much OC on the Gigabyte on quark yet. I left it at +25mhz core and 100mhz memory I think since too high and other algos have a problem when it autoswitches to them. Rather have them work consistently 24hrs than have one algo crash them and lose half a day or more.

AdamWhite

hero member

Activity: 605

Merit: 500

Quote from: ?? on ??

Quote from: AdamWhite on January 18, 2015, 08:11:42 AM

Hi, is there a possibility to improve the 750ti performance on CryptoNight algorithm?

Sorry if this has been discussed already.. thanks for sharing your work!

Yes.

Thanks. I'm using tsiv's September 2014 release.. is there another (public) fork being developed?

djm34

legendary

Activity: 1400

Merit: 1050

Quote from: sp_ on January 17, 2015, 03:09:19 PM

NVIDIA domination:

The people who are running ccminer/2014.06.15 could get a 40% boost if they switch to 1.5.30-git(SP-MOD). I guess someone has rented some tesla hashpower, and they are not smart enough to compile my fork for compute 3.0, 3.5.

lol, if you still expecting crypto people to be smart...

AdamWhite

hero member

Activity: 605

Merit: 500

Hi, is there a possibility to improve the 750ti performance on CryptoNight algorithm?

Sorry if this has been discussed already.. thanks for sharing your work!

tbearhere

legendary

Activity: 3164

Merit: 1003

Quote from: chrysophylax on January 17, 2015, 09:26:47 PM

Quote from: scryptr on January 17, 2015, 01:12:50 PM

Quote from: sp_ on January 17, 2015, 05:32:10 AM

I have now fixed the groestl speedup 2 on github. Klaus_t has been optimizing quark, so I am waiting for a pull request, and then build release 31.

SP_ BUILD v30a/b --

Version 30b is faster yet. Both my rigs see an increase of about 1200-1300kh/s mining quark. This difference is from v29 to v30b.

However, in the long-haul, 24hrs or so, build 30a/b is not stable on my slower rig. Again, it is crashing.

I hope v31 comes soon, and with smooth running code. My fast rig is now at 37.5mh/s, my slow rig at 36.2mh/s, and my fastest 750ti is at 3275kh/s. --scryptr

EDIT: I've reverted my slower rig to v29 after a third crash. --scryptr

scryptr - have you thought about a different ( possibly underlying ) issue with the slower rig?

i mean - some of my rigs would crash if i used the compiled versions of ccminer on other rigs ... when i compiled ccminer of the rig itself - it worked flawlessly ... no crashes - no issues - same version - just compiled locally ...

hence the reason why im 'standardizing' the farm with all the same components ... getting rid of the older motherboards ( which are nice giagbyte ones ) and rebuilding them into computers and rebuilding the farm with asrock h81-pro-btc motherboards with the same cpu and ram ...

i use the same version build of ccminer on ALL the machines - but some of them have locally compiled ones - as i get very similar issues to your slower machies if i dont compile locally - and some of them refuse to work at all ...

#crysx

If you have the cards all synced together ....try unsyncing them. I cant run my cards synced.

chrysophylax

legendary

Activity: 2912

Merit: 1091

--- ChainWorks Industries ---

Quote from: scryptr on January 17, 2015, 01:12:50 PM

Quote from: sp_ on January 17, 2015, 05:32:10 AM

I have now fixed the groestl speedup 2 on github. Klaus_t has been optimizing quark, so I am waiting for a pull request, and then build release 31.

SP_ BUILD v30a/b --

Version 30b is faster yet. Both my rigs see an increase of about 1200-1300kh/s mining quark. This difference is from v29 to v30b.

However, in the long-haul, 24hrs or so, build 30a/b is not stable on my slower rig. Again, it is crashing.

I hope v31 comes soon, and with smooth running code. My fast rig is now at 37.5mh/s, my slow rig at 36.2mh/s, and my fastest 750ti is at 3275kh/s. --scryptr

EDIT: I've reverted my slower rig to v29 after a third crash. --scryptr

scryptr - have you thought about a different ( possibly underlying ) issue with the slower rig?

i mean - some of my rigs would crash if i used the compiled versions of ccminer on other rigs ... when i compiled ccminer of the rig itself - it worked flawlessly ... no crashes - no issues - same version - just compiled locally ...

hence the reason why im 'standardizing' the farm with all the same components ... getting rid of the older motherboards ( which are nice giagbyte ones ) and rebuilding them into computers and rebuilding the farm with asrock h81-pro-btc motherboards with the same cpu and ram ...

i use the same version build of ccminer on ALL the machines - but some of them have locally compiled ones - as i get very similar issues to your slower machies if i dont compile locally - and some of them refuse to work at all ...

#crysx

scryptr

legendary

Activity: 1797

Merit: 1028

Quote from: sp_ on January 17, 2015, 02:43:15 PM

Quote from: scryptr on January 17, 2015, 01:12:50 PM

Quote from: sp_ on January 17, 2015, 05:32:10 AM

I have now fixed the groestl speedup 2 on github. Klaus_t has been optimizing quark, so I am waiting for a pull request, and then build release 31.

SP_ BUILD v30a/b --
Version 30b is faster yet. Both my rigs see an increase of about 1200-1300kh/s mining quark. This difference is from v29 to v30b.
However, in the long-haul, 24hrs or so, build 30a/b is not stable on my slower rig. Again, it is crashing.
I hope v31 comes soon, and with smooth running code. My fast rig is now at 37.5mh/s, my slow rig at 36.2mh/s, and my fastest 750ti is at 3275kh/s. --scryptr
EDIT: I've reverted my slower rig to v29 after a third crash. --scryptr

Release 30 is running stable here. Could be that your cards get too hot? old driver? one sick card?

Thanks--

I am running nVidia drivers 346.35, the latest for Linux. My cards are in the mid 50 degree range. This rig has the Sempron 145 unlocked to 2 cores. v29 just crashed on me, also. Since v28, stability has again been an off-and-on issue with this rig. I suspect the CPU-workload, but there are too many loose variables.

I hope release 31 runs better. --scryptr

exemplaar

hero member

Activity: 978

Merit: 506

Sp, did you tried new drivers 347.09?

Any improvements in hashrate towards 344.75?

Appreciate your work and dedication.

sp_

legendary

Activity: 2926

Merit: 1087

Team Black developer

NVIDIA supports the demoscene:

http://nv.scene.org/2015/

The demoscene support NVIDIA

www.pouet.net

sp_

legendary

Activity: 2926

Merit: 1087

Team Black developer

NVIDIA domination:

The people who are running ccminer/2014.06.15 could get a 40% boost if they switch to 1.5.30-git(SP-MOD). I guess someone has rented some tesla hashpower, and they are not smart enough to compile my fork for compute 3.0, 3.5.

sp_

legendary

Activity: 2926

Merit: 1087

Team Black developer

Quote from: scryptr on January 17, 2015, 01:12:50 PM

Quote from: sp_ on January 17, 2015, 05:32:10 AM

I have now fixed the groestl speedup 2 on github. Klaus_t has been optimizing quark, so I am waiting for a pull request, and then build release 31.

SP_ BUILD v30a/b --
Version 30b is faster yet. Both my rigs see an increase of about 1200-1300kh/s mining quark. This difference is from v29 to v30b.
However, in the long-haul, 24hrs or so, build 30a/b is not stable on my slower rig. Again, it is crashing.
I hope v31 comes soon, and with smooth running code. My fast rig is now at 37.5mh/s, my slow rig at 36.2mh/s, and my fastest 750ti is at 3275kh/s. --scryptr
EDIT: I've reverted my slower rig to v29 after a third crash. --scryptr

Release 30 is running stable here. Could be that your cards get too hot? old driver? one sick card?

sp_

legendary

Activity: 2926

Merit: 1087

Team Black developer

Just removed 1296 assembly instructions from fugue hash, but something wrong with the byteorder of the registers. Boost in x13,x14,x15 comming soon

scryptr

legendary

Activity: 1797

Merit: 1028

Quote from: sp_ on January 17, 2015, 05:32:10 AM

I have now fixed the groestl speedup 2 on github. Klaus_t has been optimizing quark, so I am waiting for a pull request, and then build release 31.

SP_ BUILD v30a/b --

Version 30b is faster yet. Both my rigs see an increase of about 1200-1300kh/s mining quark. This difference is from v29 to v30b.

However, in the long-haul, 24hrs or so, build 30a/b is not stable on my slower rig. Again, it is crashing.

I hope v31 comes soon, and with smooth running code. My fast rig is now at 37.5mh/s, my slow rig at 36.2mh/s, and my fastest 750ti is at 6340kh/s (quark). --scryptr

EDIT: I've reverted my slower rig to v29 after a third crash. --scryptr

Bombadil

hero member

Activity: 644

Merit: 500

Quote from: crzyblue on January 17, 2015, 10:49:11 AM

[2015-01-17 10:44:32] NVAPI GPU monitoring enabled.
[2015-01-17 10:44:32] 1 miner thread started, using 'x11' algorithm.
[2015-01-17 10:44:32] Stratum difficulty set to 0.01
Cuda error in func 'x11_simd512_cpu_init' at line 641 : invalid texture referenc
e.

I get this error every time I try to use release 30

Thought so already Wink

IIRC, you're skipping compute3.0 since release 10-ish?
Been using the tpruvot branch for my 760, is also a bit more up to date with api etc Tongue

Quote from: crzyblue on January 17, 2015, 11:23:54 AM

GTX 650 too old?

Yep, seems like it. You can't blame sp_, the improvements he makes, make use of features not existing in cards with a compute lower than 5.0, so just keep on using the best older one for those cards.

crzyblue

newbie

Activity: 2

Merit: 0

GTX 650 too old?

sp_

legendary

Activity: 2926

Merit: 1087

Team Black developer

Your card is too old. Release30 is compiled for compute 5.0 and 5.2 only

crzyblue

newbie

Activity: 2

Merit: 0

[2015-01-17 10:44:32] NVAPI GPU monitoring enabled.
[2015-01-17 10:44:32] 1 miner thread started, using 'x11' algorithm.
[2015-01-17 10:44:32] Stratum difficulty set to 0.01
Cuda error in func 'x11_simd512_cpu_init' at line 641 : invalid texture referenc
e.

I get this error every time I try to use release 30

chrysophylax

legendary

Activity: 2912

Merit: 1091

--- ChainWorks Industries ---

Quote from: sp_ on January 17, 2015, 09:10:07 AM

Quote from: Bombadil on January 17, 2015, 05:55:51 AM

Quote from: sp_ on January 17, 2015, 05:32:10 AM

I have now fixed the groestl speedup 2 on github. Klaus_t has been optimizing quark, so I am waiting for a pull request, and then build release 31.

Didn't you say release 30 was the last one? J/k, not complaining ^^"
Know the feel too of not being able to let a project go, still so much to do Cheesy

No, still more hash to make. Found some pickings in the x13 algo(fugue) but will have to wait.
I am moving the spreadcoin fork into my ccminer fork.

yay yay yay ...

ok - im calm ... im calm ...

Wink

#crysx

sp_

legendary

Activity: 2926

Merit: 1087

Team Black developer

Quote from: Bombadil on January 17, 2015, 05:55:51 AM

Quote from: sp_ on January 17, 2015, 05:32:10 AM

I have now fixed the groestl speedup 2 on github. Klaus_t has been optimizing quark, so I am waiting for a pull request, and then build release 31.

Didn't you say release 30 was the last one? J/k, not complaining ^^"
Know the feel too of not being able to let a project go, still so much to do Cheesy

No, still more hash to make. Found some pickings in the x13 algo(fugue) but will have to wait.
I am moving the spreadcoin fork into my ccminer fork.

Bombadil

hero member

Activity: 644

Merit: 500

Quote from: sp_ on January 17, 2015, 05:32:10 AM

I have now fixed the groestl speedup 2 on github. Klaus_t has been optimizing quark, so I am waiting for a pull request, and then build release 31.

Didn't you say release 30 was the last one? J/k, not complaining ^^"
Know the feel too of not being able to let a project go, still so much to do Cheesy

Topic: CCminer(SP-MOD) Modded NVIDIA Maxwell / Pascal kernels. - page 1184. (Read 2347597 times)