CCminer(SP-MOD) Modded NVIDIA Maxwell / Pascal kernels. - page 1023.

Slava_K

hero member

Activity: 677

Merit: 500

Quote from: sp_ on July 29, 2015, 04:36:10 PM

if you know how to code, you can change a few lines and make a bitcoin in a couple of days with few rigs.
With some more workl the gpu miner can be 20 times faster than the cpu miner for an average cpu. That's why somebody is selling private kernals for 3 BTC++
Donate some Beers, and I will finish the kernal.

i am not a programmer... but some btc for beer? tomorrow in evening i send to you 0.045 btc (3 days of my mining)

hashbrown9000

sr. member

Activity: 427

Merit: 250

Quote

if you know how to code, you can change a few lines and make a bitcoin in a couple of days with few rigs.

HA! Here I am with a couple rigs, barely scraping by at about 1/10 that rate. wish i knew what i was looking at when i open up some of this code.

Grim

sr. member

Activity: 506

Merit: 252

Quote from: ?? on ??

Quote from: sp_ on July 29, 2015, 04:39:07 PM

Quote from: ?? on ??

Quote from: sp_ on July 29, 2015, 04:27:35 PM

The gtx 970 has a tdp of 150W but the windforce has a 4pin and a 3pin power cable. It uses around 180W to produce 16.4 MHASH. (0,091 MHASH per watt)
the R9 Fury has a tdp of 275W.

Measurements from the wall? Because mine's in a machine with overvolted cards, and an FX-8370...

yes. I use 1x risers and have 1x enabled in the bios for all cards. standard voltage and just factory overclock. Other 970 cards use less power and produce less hash, but 2x 970 seems to cost the same, and produce more hash per watt. But nice job on the kernals.

Ah, okay. I need to drop volts on the other cards and see how it does, then.

@Wolf0 you need to find yourself a memory bandwidth limited algo. That is the only place a fury can shine at.
A gcn 1.2 core is only slightly more efficient than a gcn 1.0 core so no wonder your hash per watt numbers on compute intensive algos is as bad as on any 280x.

Hint! Axiom Hint!

hashbrown9000

sr. member

Activity: 427

Merit: 250

damn. flaxscript really dumped hard today

sp_

legendary

Activity: 2926

Merit: 1087

Team Black developer

Quote from: ?? on ??

Quote from: sp_ on July 29, 2015, 04:27:35 PM

The gtx 970 has a tdp of 150W but the windforce has a 4pin and a 3pin power cable. It uses around 180W to produce 16.4 MHASH. (0,091 MHASH per watt)
the R9 Fury has a tdp of 275W.

Measurements from the wall? Because mine's in a machine with overvolted cards, and an FX-8370...

yes. I use 1x risers and have 1x enabled in the bios for all cards. standard voltage and just factory overclock. Other 970 cards use less power and produce less hash, but 2x 970 seems to cost the same, and produce more hash per watt. But nice job on the kernals.

sp_

legendary

Activity: 2926

Merit: 1087

Team Black developer

Quote from: Slava_K on July 29, 2015, 03:18:14 PM

Doesnot submit hashes (or even found any). GPU load very small...
GREAT WORK SP_!!!

if you know how to code, you can change a few lines and make a bitcoin in a couple of days with few rigs.
With some more workl the gpu miner can be 20 times faster than the cpu miner for an average cpu. That's why somebody is selling private kernals for 3 BTC++
Donate some Beers, and I will finish the kernal.

sp_

legendary

Activity: 2926

Merit: 1087

Team Black developer

The gtx 970 has a tdp of 150W but the windforce has a 4pin and a 3pin power cable. It uses around 180W to produce 16.4 MHASH. (0,091 MHASH per watt)
the R9 Fury has a tdp of 275W.

Slava_K

hero member

Activity: 677

Merit: 500

Quote from: ?? on ??

The R9 Fury (NOT Fury X, less CUs), does 20.5MH/s on Quark.

But hash per watt?

dga

hero member

Activity: 737

Merit: 511

Quote from: sp_ on July 29, 2015, 03:42:27 PM

axioncoin
@github. Some work left. (will not work without fixing the bugs)

It's not finished. Don't use it. I am sharing so somebody else can fix it..

Submitted a speedup in quark and x11.. Precalculated 1/32 of blake from cpu->const mem

quark is now hashing at 16.4 MHASH on the gigabyte windforce 970 factory clocks.

Typo in Makefile.am --

Code:

@@ -54,7 +55,7 @@ ccminer_SOURCES = elist.h miner.h compat.h \
   x11/cuda_x11_luffa512_Cubehash.cu \
   x13/x13.cu x13/cuda_x13_hamsi512.cu x13/cuda_x13_fugue
   x15/x14.cu x15/x15.cu x15/cuda_x14_shabal512.cu x15/cu
- x15/whirlpool.cu x15\cuda_axiom.cu x15\axiom.cu \
+ x15/whirlpool.cu x15/cuda_axiom.cu x15/axiom.cu \

(backslash instead of forward slash)

Also, for a bit, I thought that Wolf was bragging about the R9 Furry. I'm glad I misread.

sp_

legendary

Activity: 2926

Merit: 1087

Team Black developer

axioncoin
@github. Some work left. (will not work without fixing the bugs)

It's not finished. Don't use it. I am sharing so somebody else can fix it..

Submitted a speedup in quark and x11.. Precalculated 1/32 of blake from cpu->const mem

quark is now hashing at 16.4 MHASH on the gigabyte windforce 970 factory clocks.

Epsylon3

legendary

Activity: 1484

Merit: 1082

ccminer/cpuminer developer

dunno what to answer... but the coin dev will not be happy about that... and that draft on your main branch

Slava_K

hero member

Activity: 677

Merit: 500

Quote from: sp_ on July 29, 2015, 02:26:51 PM

@github. Some work left. (will not work without fixing the bugs)

The padding bytes are incorrect.
I think the temp buffer needs to be moved from the stack out to globalmem etc..

Perhaps we should add cpu mining first to verfiy the methods.

Doesnot submit hashes (or even found any). GPU load very small...
GREAT WORK SP_!!!

sp_

legendary

Activity: 2926

Merit: 1087

Team Black developer

@github. Some work left. (will not work without fixing the bugs)

The padding bytes are incorrect.
I think the temp buffer needs to be moved from the stack out to globalmem etc..

Perhaps we should add cpu mining first to verfiy the methods.

pallas

legendary

Activity: 2716

Merit: 1094

Black Belt Developer

Push it and I'll have a look :-D

sp_

legendary

Activity: 2926

Merit: 1087

Team Black developer

Release\cuda_axion.cu.obj "C:\code\ccminer\x15\cuda_axion.cu"
1> ptxas info : Overriding global maxrregcount 80 with entry-specific value 255 computed using thread count
1> ptxas info : 0 bytes gmem, 80 bytes cmem[3]
1> ptxas info : Compiling entry function '_Z17axiom_gpu_hash_80jjPj' for 'sm_50'
1> ptxas info : Function properties for _Z17axiom_gpu_hash_80jjPj
1> 2097152 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads
1> ptxas info : Used 123 registers, 332 bytes cmem[0]
1> ptxas info : Overriding global maxrregcount 80 with entry-specific value 255 computed using thread count
1> ptxas info : 0 bytes gmem, 80 bytes cmem[3]
1> ptxas info : Compiling entry function '_Z17axiom_gpu_hash_80jjPj' for 'sm_52'
1> ptxas info : Function properties for _Z17axiom_gpu_hash_80jjPj
1> 2097152 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads
1> ptxas info : Used 123 registers, 332 bytes cmem[0]
1> cuda_axion.cu

there has to bether way to do this