Just a note:
ccminer-52.exe to run with 900 series by tpruvot 1.4.6 => fastest to 900 series till now (8.2-8.3MH/s X11 per card)
ccminermod.exe by sp_ => fastest to run with 750Ti => 2.85MH/s X11 per card
Cheers
Topic: CCminer(SP-MOD) Modded NVIDIA Maxwell / Pascal kernels. - page 1235. (Read 2347601 times)
Good improovement , Ill test 900s later 
980 doing 8.2MH/s X11.
But miner crashes when closing.
But miner crashes when closing.
before you take all the credits, i released a new version 1.4.6 :
https://github.com/tpruvot/ccminer/releases/
https://github.com/tpruvot/ccminer/releases/
From https://github.com/tsiv/ccminer/releases s3 algo ~9mh/s 750ti overclocked some
EDIT: at 60 watts per
before you take all the credits, i released a new version 1.4.6 :
https://github.com/tpruvot/ccminer/releases/
https://github.com/tpruvot/ccminer/releases/
I have hit 7000 KH/S with a 970!! Using 250 clock OC with 37 mv overvoltage
This is using SP's first release, I feel when I have his latest release it can go even higher
price per performance ratio for the 970 is unmatched right now
Good work!
This is using SP's first release, I feel when I have his latest release it can go even higher
price per performance ratio for the 970 is unmatched right now
Good work!
I have the 970, posted the numbers
Would be very interested in this, but I'd be going to buy some 970s from Micro Center to do so. I checked a few pages back and didn't see any hard numbers on the 970s (only 980s) in terms of kh/s and wattage. I have a bunch of rigs with various 270Xs but I'm looking to downscale and convert to nvidia cards due to their increased efficiency.
The sharemem tweaks was a dead end. But managed to squeze another 1.5% faster than the Schleicher implementation in the echo hash.
This code can be optimized:
uint32_t t;
t = ((ab & 0x80808080) >> 7);
uint32_t abx = t<<4 ^ t<<3 ^ t<<1 ^ t;
t = ((bc & 0x80808080) >> 7);
uint32_t bcx = t<<4 ^ t<<3 ^ t<<1 ^ t;
t = ((cd & 0x80808080) >> 7);
uint32_t cdx = t<<4 ^ t<<3 ^ t<<1 ^ t;
abx ^= ((ab & 0x7F7F7F7F) << 1);
bcx ^= ((bc & 0x7F7F7F7F) << 1);
cdx ^= ((cd & 0x7F7F7F7F) << 1);
because
(ab & 0x7F7F7F7F)=ab^((ab & 0x80808080)
saves a register/moves/deadcode, and with the proper configuration 1.5% more hash in the ECHO.
This code can be optimized:
uint32_t t;
t = ((ab & 0x80808080) >> 7);
uint32_t abx = t<<4 ^ t<<3 ^ t<<1 ^ t;
t = ((bc & 0x80808080) >> 7);
uint32_t bcx = t<<4 ^ t<<3 ^ t<<1 ^ t;
t = ((cd & 0x80808080) >> 7);
uint32_t cdx = t<<4 ^ t<<3 ^ t<<1 ^ t;
abx ^= ((ab & 0x7F7F7F7F) << 1);
bcx ^= ((bc & 0x7F7F7F7F) << 1);
cdx ^= ((cd & 0x7F7F7F7F) << 1);
because
(ab & 0x7F7F7F7F)=ab^((ab & 0x80808080)
saves a register/moves/deadcode, and with the proper configuration 1.5% more hash in the ECHO.
x11: 2680khash (ccminer-djm34)
x13: 2030khash (ccminer-djm34)
x15: 1800khash (ccminer-djm34)
zotac 750ti stock
x11: 2820khash (ccminer-sp)
x13: 2090khash (ccminer-sp)
x15: 1880khash (ccminer-sp)
I will test 970 and 980 tomorrow.
Sadly my Suarez lost the game in his first show.
, upset
x13: 2030khash (ccminer-djm34)
x15: 1800khash (ccminer-djm34)
zotac 750ti stock
x11: 2820khash (ccminer-sp)
x13: 2090khash (ccminer-sp)
x15: 1880khash (ccminer-sp)
I will test 970 and 980 tomorrow.
Sadly my Suarez lost the game in his first show.
, upset
Not Skein, I ment Shavite. Was mixing the X'es here. They both share the implementation in cuda_x11_aes.cu. The method aes_round()
I want to do to 2 table reads into one. but this will require 256kb of shared mem. so I need to split the bits and do seperate code for the upper bits combinations.
sharedMemory[__byte_perm(x0, 0, 0x4440)]^sharedMemory[__byte_perm(x1, 0, 0x4441) + 256],
I want to do to 2 table reads into one. but this will require 256kb of shared mem. so I need to split the bits and do seperate code for the upper bits combinations.
sharedMemory[__byte_perm(x0, 0, 0x4440)]^sharedMemory[__byte_perm(x1, 0, 0x4441) + 256],
I re-wrote two new hashing algorithms yesterday, it has some bugs so the numbers are not ready yet, but it looks like 3,5 MHASH on the 750ti on x11 with 38 watt in the wall per card. 9.2 watt per MHASH.
Impressive ! But it I cant't get it to work properly:(. The idea was to reduce the number of sharemem access in echo and skein (ahs). ccminer is currently it is doing a lookup for each byte, but i want to use the increased and improved sharemem in the maxwell to lookup more bits. The 1kb table can become 32kb or 48kb. Block latency should stay low as the probabillity of hitting 2 equal adresses is lower with 48kb combinations.
Strangely, I tried it while working on skein-1024 and the performance were just terrible and it was a major slow down compared to the version not using it...
I re-wrote two new hashing algorithms yesterday, it has some bugs so the numbers are not ready yet, but it looks like 3,5 MHASH on the 750ti on x11 with 38 watt in the wall per card. 9.2 watt per MHASH.
Impressive ! But it I cant't get it to work properly:(. The idea was to reduce the number of sharemem access in echo and skein (ahs). ccminer is currently it is doing a lookup for each byte, but i want to use the increased and improved sharemem in the maxwell to lookup more bits. The 1kb table can become 32kb or 48kb. Block latency should stay low as the probabillity of hitting 2 equal adresses is lower with 48kb combinations.
You can send me a pm with your email adress and I will send you the betaminer. (sundays build).
nice work , I PMd you already I re-wrote two new hashing algorithms yesterday, it has some bugs so the numbers are not ready yet, but it looks like 3,5 MHASH on the 750ti on x11 with 38 watt in the wall per card. 9.2 watt per MHASH.
Impressive !
You can send me a pm with your email adress and I will send you the betaminer. (sundays build).
Can someone give the download url for miner?
I re-wrote two new hashing algorithms yesterday, it has some bugs so the numbers are not ready yet, but it looks like 3,5 MHASH on the 750ti on x11 with 38 watt in the wall per card. 9.2 watt per MHASH.
Beautiful, just adding to the post, your latest miner don´t crash as the old ones when closing the app, now I can close it without problems.
I re-wrote two new hashing algorithms yesterday, it has some bugs so the numbers are not ready yet, but it looks like 3,5 MHASH on the 750ti on x11 with 38 watt in the wall per card. 9.2 watt per MHASH.
Hit 9MH/s on my new 290X, clocks 1110/1625.
That is very good. I hope you don't release it. 90% of the GPU mining rigs are from AMD.
I wil try to push the 980 above 10MHASH, and then keep it private if the speed get's bether.9.45MH/s on 290X - race you to 10MH!
What´s the Power consumption?
Last I checked, over 270X, 280X, 290X, and 7950 all undervolted, I got 22.13MH/s for 610W.
EDIT: I'll go test high performance numbers against the stock miner with wattages in a bit.
Wow, what a mix of cards! Nice results.
I´m getting 490-500W for 3x980 + 1 750Ti stock voltages and Overclocked for 25,3MH/s@X11.
I built her for testing - Pitcairn, Tahiti, and Hawaii - just need the R9 285 for Tonga.
Cool, testing in dif. chips.
Can´t wait to see 980 mining at 10MH/s!
Jump to: