Author

Topic: CCminer(SP-MOD) Modded NVIDIA Maxwell / Pascal kernels. - page 1150. (Read 2347599 times)

hero member
Activity: 644
Merit: 500
I'm going to bangkok and spend some satoshis. Wink


Shall we call you Mrs. Sp_ afterwards? Tongue
J/k, have a great one Cheesy

That's MY dream vacation!

lets all make a btc pool and all go on a tour for a few weeks Wink ...

#crysx

God yes!

Count me in  Cool

Lol.
* Bombadil already sees "Cudaminer Conference, Bangkok" in the upcoming headlines
legendary
Activity: 1050
Merit: 1000
I'm going to bangkok and spend some satoshis. Wink


That's MY dream vacation!

lets all make a btc pool and all go on a tour for a few weeks Wink ...

#crysx

God yes!

Count me in  Cool
sr. member
Activity: 271
Merit: 251
release 42, trying whirlpoolx:
Code:
C:\_Mining\ccminer-sp>ccminer -a whirlpoolx -o stratum+tcp://stratum.westhash.com:3343 -R 5 -u 1xxxxxxxxxxxxxxxxxxxxxx -p d=0.02 -d 0
*** ccminer 1.5.41-git(SP-MOD) for nVidia GPUs by sp-hash@github ***
        Built with VC++ 2013 and nVidia CUDA SDK 6.5

  Based on pooler cpuminer 2.3.2 and the tpruvot@github fork
   CUDA support by Christian Buchner, Christian H. and DJM34
  Includes optimizations implemented by sp , klaust, tpruvot and tsiv.

[2015-03-09 22:33:33] Starting Stratum on stratum+tcp://stratum.westhash.com:3343
[2015-03-09 22:33:33] NVAPI GPU monitoring enabled.
[2015-03-09 22:33:33] 1 miner thread started, using 'whirlpoolx' algorithm.
[2015-03-09 22:33:34] Stratum difficulty set to 0.04
[2015-03-09 22:33:34] stratum.westhash.com:3343 whirlpoolx block 72182
[2015-03-09 22:33:37] stratum time is at least 34s in the future
[2015-03-09 22:33:37] stratum.westhash.com:3343 whirlpoolx block 72183
[2015-03-09 22:33:59] stratum.westhash.com:3343 whirlpoolx block 72183
[2015-03-09 22:34:32] stratum.westhash.com:3343 whirlpoolx block 72184
[2015-03-09 22:35:27] stratum.westhash.com:3343 whirlpoolx block 72184

what am I missing? I have GTX 970s
legendary
Activity: 2912
Merit: 1091
--- ChainWorks Industries ---
Rough idea - use lookup tables to pick up only linear independent bits and regenerate the rest by boolean matrix multiplication.
Small R and E lookup tables, the only source of non-linearity in whirlpool, take only 4*16 bits and could be implemented via uint64_t shift, without any memory access. Yeah, chatting with myself here Smiley

its all a good thing smolen ...

small improvement is still improvement ...

always watching ...

Smiley ...

#crysx
hero member
Activity: 524
Merit: 500
Rough idea - use lookup tables to pick up only linear independent bits and regenerate the rest by boolean matrix multiplication.
Small R and E lookup tables, the only source of non-linearity in whirlpool, take only 4*16 bits and could be implemented via uint64_t shift, without any memory access. Yeah, chatting with myself here Smiley
legendary
Activity: 2912
Merit: 1091
--- ChainWorks Industries ---
I'm going to bangkok and spend some satoshis. Wink


That's MY dream vacation!

lets all make a btc pool and all go on a tour for a few weeks Wink ...

#crysx

How could a bitcoin pool hold[water?  Wouldn't the tourists wade in and grab them?       --scryptr

not if you have btc sharks ...

there are PLENTY of those around Wink ... hehehe ...

#crysx
legendary
Activity: 1797
Merit: 1028
I'm going to bangkok and spend some satoshis. Wink


That's MY dream vacation!

lets all make a btc pool and all go on a tour for a few weeks Wink ...

#crysx

How could a bitcoin pool hold[water?  Wouldn't the tourists wade in and grab them?       --scryptr
legendary
Activity: 2912
Merit: 1091
--- ChainWorks Industries ---
I'm going to bangkok and spend some satoshis. Wink


That's MY dream vacation!

lets all make a btc pool and all go on a tour for a few weeks Wink ...

#crysx
newbie
Activity: 14
Merit: 0
latest source on git compiles and runs fine here
changed x15/whirlpoolx.cu line 47 to:   uint32_t throughput = pow(2,23);
23 was the highest that works on my GTX850M
was getting about 63Mh
overclocked GPU +105 memory +800

optirun nvidia-settings -c :8 -a [gpu:0]/GPUGraphicsClockOffset[2]=105 -a [gpu:0]/GPUMemoryTransferRateOffset[2]=800

[2015-03-09 17:28:29] GPU #0: GeForce GTX 850M, 69457 kH/s
[2015-03-09 17:28:29] accepted: 907/908 (99.89%), 69475 khash/s yay!!!
[2015-03-09 17:28:42] GPU #0: GeForce GTX 850M, 69404 kH/s
[2015-03-09 17:28:42] accepted: 908/909 (99.89%), 69474 khash/s yay!!!

zero rejects and card is running quite cool 64C
legendary
Activity: 2002
Merit: 1051
ICO? Not even once.
I'm going to bangkok and spend some satoshis. Wink


Have fun!
legendary
Activity: 2912
Merit: 1091
--- ChainWorks Industries ---
I'm going to bangkok and spend some satoshis. Wink


that sounds like an awesome idea ...

njoi your time and come back refreshed mate ...

i envy you at the moment Tongue ...

#crysx
sp_
legendary
Activity: 2926
Merit: 1087
Team Black developer
I'm going to bangkok and spend some satoshis. Wink
legendary
Activity: 2912
Merit: 1091
--- ChainWorks Industries ---
Some nice work there done in the other forks. I Will be away for a while so will take some time before I test and
merge. I see some more possible speedups in whirlpoolx as well.



all good sp ...

if its a holiday - njoi mate Smiley ...

#crysx
sp_
legendary
Activity: 2926
Merit: 1087
Team Black developer
Some nice work there done in the other forks. I Will be away for a while so will take some time before I test and
merge. I see some more possible speedups in whirlpoolx as well.

legendary
Activity: 2912
Merit: 1091
--- ChainWorks Industries ---
guys ... sp ccminer works perfectly with all workers - and all workers have a minimum of 5 cards in them ...
cards are gigabyte 750ti oc lp ...
if you guys are overclocking - then the clocks are too high for one of the cards to handle ...
#crysx

thanks - but we talk exclusive about the new -a whirlpoolX algo.
this is broken on linux for multi GPU systems. anything else works.

tpruvot ccminer works with multiple GPUs under Linux.

yep. tpruvot saw the problem quite early and fixed it yesterday.
I was already away from the computer - otherwise I would gave you a hint.

o ok ... thats all good then ...

#crysx
sr. member
Activity: 330
Merit: 252
guys ... sp ccminer works perfectly with all workers - and all workers have a minimum of 5 cards in them ...
cards are gigabyte 750ti oc lp ...
if you guys are overclocking - then the clocks are too high for one of the cards to handle ...
#crysx

thanks - but we talk exclusive about the new -a whirlpoolX algo.
this is broken on linux for multi GPU systems. anything else works.

tpruvot ccminer works with multiple GPUs under Linux.

yep. tpruvot saw the problem quite early and fixed it yesterday.
I was already away from the computer - otherwise I would gave you a hint.
hero member
Activity: 524
Merit: 500
That is really good. my 980 is around 300MHASH with overclocking. I tried to rewrite to uint4 but it went slower. What did you do? did you remove the lookuptables?
I have, but it was slower - that implementation should be AWESOME on FPGA, though.
Rough idea - use lookup tables to pick up only linear independent bits and regenerate the rest by boolean matrix multiplication.
member
Activity: 111
Merit: 10
... but I've encountered another issue.  When I hash with both GPUs (both 750Ti) one GPU gives me a crazy hashing number, for .

brother. i confirm. i have the same issue with every source out there (linux version!).

tpruvot ccminer works with multiple GPUs under Linux.

guys ... sp ccminer works perfectly with all workers - and all workers have a minimum of 5 cards in them ...

cards are gigabyte 750ti oc lp ...

if you guys are overclocking - then the clocks are too high for one of the cards to handle ...

#crysx

no overclock on any of my cards.  cards work fine individually, but not together.  I only have two cards.  can you run whirlpoolx using -d 0,1 (so you'd only run two cards), and does that work? thanks

what version of ccminer-sp are you running? ...

ill test when I get into the office in a couple of hours ...

what is the commandline parameters you use so that I can duplicate your setup? ...

I can test it easily as I have a worker with only two cards in it also ...

#crysx

I'm not using any parameters, just --algo, --url, --user, --pass (without any difficulty passing), and --api-bind.  No need to test though, I just cp tpruvot's x15/cuda_whirlpoolx.cu and x15/whirlpoolx.cu over sp-hash's x15/cuda_whirlpoolx.cu and x15/whirlpoolx.cu, compiled and works with my two 750Ti's in linux  Grin.
legendary
Activity: 2912
Merit: 1091
--- ChainWorks Industries ---
... but I've encountered another issue.  When I hash with both GPUs (both 750Ti) one GPU gives me a crazy hashing number, for .

brother. i confirm. i have the same issue with every source out there (linux version!).

tpruvot ccminer works with multiple GPUs under Linux.

guys ... sp ccminer works perfectly with all workers - and all workers have a minimum of 5 cards in them ...

cards are gigabyte 750ti oc lp ...

if you guys are overclocking - then the clocks are too high for one of the cards to handle ...

#crysx

no overclock on any of my cards.  cards work fine individually, but not together.  I only have two cards.  can you run whirlpoolx using -d 0,1 (so you'd only run two cards), and does that work? thanks

what version of ccminer-sp are you running? ...

ill test when I get into the office in a couple of hours ...

what is the commandline parameters you use so that I can duplicate your setup? ...

I can test it easily as I have a worker with only two cards in it also ...

#crysx
legendary
Activity: 1797
Merit: 1028
... but I've encountered another issue.  When I hash with both GPUs (both 750Ti) one GPU gives me a crazy hashing number, for .

brother. i confirm. i have the same issue with every source out there (linux version!).

tpruvot ccminer works with multiple GPUs under Linux.

guys ... sp ccminer works perfectly with all workers - and all workers have a minimum of 5 cards in them ...

cards are gigabyte 750ti oc lp ...

if you guys are overclocking - then the clocks are too high for one of the cards to handle ...

#crysx

no overclock on any of my cards.  cards work fine individually, but not together.  I only have two cards.  can you run whirlpoolx using -d 0,1 (so you'd only run two cards), and does that work? thanks


LINUX and SP_ v42--

SP_'s build errored out under Linux on my 6x750ti FTW rig.  I got the "out of memory/scanhash" error reported a page back.  No cards would hash.  Using tpruvot's v1.6 rc, it compiled with no errors.  It runs smoothly at 458MH/s on the rig, about 76MH/s per card, for whirlpoolx.  This is the same rate as SP_ and KlausT are getting.       --scryptr
Jump to: