CCminer(SP-MOD) Modded NVIDIA Maxwell / Pascal kernels. - page 1238.

djm34

legendary

Activity: 1400

Merit: 1050

Quote from: sp_ on October 20, 2014, 03:48:02 PM

Today I have integrated the changes from Schleicher that he posted on the ccminer thread into my mod:

Quote from: Schleicher on October 19, 2014, 05:07:14 PM
I managed to increase the quark and nist5 speed a little bit.
Source code:
https://github.com/KlausT/ccminer

simd512 seems to be 10% faster.
My stock 750TI is approaching 2700 without overclock. and 3000 with oclock.

This is without the faster Keccak (created by nvidia) in cudaminer.

These are the optimized numbers so far.

Blake xxx (not done)
skein    1.5%
BMW     60%
jh512    4.5%
keccac    1%
cubehash: 7.5%
shavite: 3.6%
simd512: 9,2%
fuge:   4,70%
hamsi: 6.97%
shabal: 22%
wirlpool:1.87%
echo: 5.5%
luffa: 0.4%

3 coders have contributed to the new speedup.

The sourcecode will be checked into the blakecoin fork by Epsylon3

https://bitcointalksearch.org/topic/ann-ccminer-23-opensource-gpl-tpruvot-770064

That uint2 keccak was done by mtrlt, if I'm not mistaken.

actually using it makes no difference on x11 it is already the fastest kernel of the bunch by far, also for some strange reason it does not work with compute 5.2 (registers gets confused Grin

had some weird issues when testing where some variables weren't updated at all...)

Amph

legendary

Activity: 3206

Merit: 1069

Quote from: SS2006 on October 19, 2014, 08:43:42 PM

Very nice! I'm now seeing 6600-6700 KH/S on my GTX 970 OC'd by 200 on clock on X11. (Original was 6100-6200), so an extra 500 KH/S
Def the best ccminer I've seen for X11 (i only mine X11)
Thanks! I hope there is more improvement, donations coming soon

I should add the hash rate is not too stable at times, it does bounce around a lot from 6300 to 6700

hey that's good, can you tell me the consumption?

sp_

legendary

Activity: 2884

Merit: 1087

Team Black developer

Today I have integrated the changes from Schleicher that he posted on the ccminer thread into my mod:

Quote from: Schleicher on October 19, 2014, 05:07:14 PM
I managed to increase the quark and nist5 speed a little bit.
Source code:
https://github.com/KlausT/ccminer

simd512 seems to be 10% faster.
My stock 750TI is approaching 2700 without overclock. and 3000 with oclock.

This is without the faster Keccak (created by nvidia) in cudaminer.

These are the optimized numbers so far.

Blake xxx (not done)
skein    1.5%
BMW    60%
jh512    4.5%
keccac    1%
cubehash: 7.5%
shavite: 3.6%
simd512: 9,2%
fuge:   4,70%
hamsi: 6.97%
shabal: 22%
wirlpool:1.87%
echo: 5.5%
luffa: 0.4%

3 coders have contributed to the new speedup.

The sourcecode will be checked into the blakecoin fork by Epsylon3

https://bitcointalksearch.org/topic/ann-ccminer-23-opensource-gpl-tpruvot-770064

go6ooo1212

legendary

Activity: 1512

Merit: 1000

quarkchain.io

@sp_
I've tested the new reliese all day on x13.
970 went up to 5050 kH/s ; 980 reaches 5960 kH/s
The losses are about 2.4-2.5%

sp_

legendary

Activity: 2884

Merit: 1087

Team Black developer

Quote from: ?? on ??

I don't know if the 980 will ever beat a 290X in raw hashrate; power consumption, certainly, but not plain performance. I've got 8.2MH/s on low clocks now - getting a better card in a couple of days.

Still alot than can be done. Half of the Hashfunctions are still not modified in my mod. My compute 5.2 build is close to 8 MHASH with heavy overclock. 10MHASH is only 25% faster.

25% faster but 100% more profit for the miners.

happy hashing

sp_

legendary

Activity: 2884

Merit: 1087

Team Black developer

Quote from: SS2006 on October 20, 2014, 02:54:04 AM

the difficulty fixed it. I always thought the faster i saw yay the better. Even tho now I see consistent high rates (6700), the accepts are much slower, is that OK, am I still getting paid the same/more on the pool side?
Thanks man

Yes, you will be payed the same. Probobly abit more because higher stable hashrates.

SS2006

sr. member

Activity: 285

Merit: 250

Quote from: sp_ on October 20, 2014, 02:12:12 AM

Quote from: SS2006 on October 19, 2014, 08:43:42 PM

Very nice! I'm now seeing 6600-6700 KH/S on my GTX 970 OC'd by 200 on clock on X11. (Original was 6100-6200), so an extra 500 KH/S
Def the best ccminer I've seen for X11 (i only mine X11)
Thanks! I hope there is more improvement, donations coming soon

I should add the hash rate is not too stable at times, it does bounce around a lot from 6300 to 6700

Try to mine with a higher diff. Each time the miner is finding a nounce the hashrate is dropping. You can alse see this in GPU-Z. I think the author of the fork tried to fix this. Thread issue... I am focusing on the kernals right now. I beleive I can push the 980 above 10MHASH. NVIDIA will sell a shitload of cards with my improved miner, ask them to donate a card for me. Wink

the difficulty fixed it. I always thought the faster i saw yay the better. Even tho now I see consistent high rates (6700), the accepts are much slower, is that OK, am I still getting paid the same/more on the pool side?
Thanks man

SS2006

sr. member

Activity: 285

Merit: 250

heck if you can push the 970 that far relatively might get you card myself Tongue

sp_

legendary

Activity: 2884

Merit: 1087

Team Black developer

Quote from: SS2006 on October 19, 2014, 08:43:42 PM

Very nice! I'm now seeing 6600-6700 KH/S on my GTX 970 OC'd by 200 on clock on X11. (Original was 6100-6200), so an extra 500 KH/S
Def the best ccminer I've seen for X11 (i only mine X11)
Thanks! I hope there is more improvement, donations coming soon

I should add the hash rate is not too stable at times, it does bounce around a lot from 6300 to 6700

Try to mine with a higher diff. Each time the miner is finding a nounce the hashrate is dropping. You can alse see this in GPU-Z. I think the author of the fork tried to fix this. Thread issue... I am focusing on the kernals right now. I beleive I can push the 980 above 10MHASH. NVIDIA will sell a shitload of cards with my improved miner, ask them to donate a card for me. Wink

jjjordan

sr. member

Activity: 271

Merit: 251

EVGA GTX 970 SC ACX1 (stock)
X11 5900-6000
X13 4800-4900

jpouza

legendary

Activity: 2660

Merit: 1106

Monster!

share image

jpouza

legendary

Activity: 2660

Merit: 1106

Quote from: sp_ on October 19, 2014, 08:22:05 PM

I have emailed some of you a version of the miner with compute 5.2

Please test and report in the thread.

If I forgot anyone please resend your email on pm.

JESUS!!!!WTF!!!
Almost 8MH/s@X11 on my 980 cards. WOW!!!
Keep up the awesome job man!

SS2006

sr. member

Activity: 285

Merit: 250

Very nice! I'm now seeing 6600-6700 KH/S on my GTX 970 OC'd by 200 on clock on X11. (Original was 6100-6200), so an extra 500 KH/S
Def the best ccminer I've seen for X11 (i only mine X11)
Thanks! I hope there is more improvement, donations coming soon

I should add the hash rate is not too stable at times, it does bounce around a lot from 6300 to 6700

sp_

legendary

Activity: 2884

Merit: 1087

Team Black developer

I have emailed some of you a version of the miner with compute 5.2

Please test and report in the thread.

If I forgot anyone please resend your email on pm.

sp_

legendary

Activity: 2884

Merit: 1087

Team Black developer

Works. thanks

in jh512 replace the swap implementation with these: A small boost.

#define SWAP4(x,y)\
      y = (x & 0xf0f0f0f0UL); \
      x = (x ^ y); \
      y = (y >> 4); \
      x = (x << 4); \
      x= x | y;

#define SWAP2(x,y)\
      y = (x & 0xccccccccUL); \
      x = (x ^ y); \
      y = (y >> 2); \
      x = (x << 2); \
      x= x | y;

#define SWAP1(x,y)\
      y = (x & 0xaaaaaaaaUL); \
      x = (x ^ y); \
      y = (y >> 1); \
      x = x + x; \
      x= x | y;

sp_

legendary

Activity: 2884

Merit: 1087

Team Black developer

Quote from: ?? on ??

Try Hamsi. Easy pickings.

0^m0=m0 ? Trying it now

sp_

legendary

Activity: 2884

Merit: 1087

Team Black developer

A new build is soon ready for the betatesters. This version includes compute5_2 compilation.

On the TI

(x11,x13,x15,x17) Keccak +1%
(x14,x15,x17) shabal: +22%

Jackpotcoin and quark still have issues, the fixes I did yesterday didn't work.

If you like more hashpower, please donate.

I got some free beers from some of you this weekend and it was ‪appriciated.

jpouza

legendary

Activity: 2660

Merit: 1106

Keep up the good work:

subir foto

sp_

legendary

Activity: 2884

Merit: 1087

Team Black developer

Quote from: ?? on ??

Probably a little more; and X15 can be improved a lot.

I managed to optimize shabal 22% faster. (x15)
from 104800 MHASH
to 128500 MHASH.

But wirlpool is much slower, so not so much increase in the total hash. x15@ 1800-1850 standard clocked 750ti

jjjordan

sr. member

Activity: 271

Merit: 251

I am doing now 5700 and more, stock.
Replaced the MB and installed Windows 10.
I love Win10 so far! It looks great

Topic: CCminer(SP-MOD) Modded NVIDIA Maxwell / Pascal kernels. - page 1238. (Read 2347426 times)