CCminer(SP-MOD) Modded GPU kernels.

[tbearhere]

On the 750ti 50kh more x11. 

[Epsylon3]

Linux profile of your repo, indeed big difference :

Code:

sp - before echo (linux x64)
==11174== Profiling result:
Time(%)      Time     Calls       Avg       Min       Max  Name
 20.76%  2.87625s        53  54.269ms  54.098ms  55.278ms  x11_echo512_gpu_hash_64(int, unsigned int, unsigned long*, unsigned int*)
 18.83%  2.60877s        54  48.311ms  48.168ms  53.868ms  quark_groestl512_gpu_hash_64_quad(int, unsigned int, unsigned int*, unsigned int*)
 13.02%  1.80384s        54  33.404ms  32.752ms  37.241ms  x11_shavite512_gpu_hash_64(int, unsigned int, unsigned long*, unsigned int*)
 11.04%  1.52931s        53  28.855ms  28.780ms  30.472ms  x11_simd512_gpu_expand_64(int, unsigned int, unsigned long*, unsigned int*, uint4*)
  7.25%  1.00414s        54  18.595ms  18.548ms  20.737ms  x11_cubehash512_gpu_hash_64(int, unsigned int, unsigned long*, unsigned int*)
  5.32%  737.65ms        54  13.660ms  13.589ms  15.234ms  quark_jh512_gpu_hash_64(int, unsigned int, unsigned long*, unsigned int*)
  5.03%  697.42ms        54  12.915ms  12.778ms  14.462ms  x11_luffa512_gpu_hash_64(int, unsigned int, unsigned long*, unsigned int*)
  3.05%  422.23ms        53  7.9665ms  7.8972ms  8.0252ms  x11_simd512_gpu_compress2_64(int, unsigned int, unsigned long*, unsigned int*, uint4*, int*)
  2.90%  401.89ms        54  7.4425ms  6.9065ms  8.3138ms  quark_bmw512_gpu_hash_64(int, unsigned int, unsigned long*, unsigned int*)
  2.90%  401.74ms        54  7.4396ms  7.4077ms  8.2859ms  quark_blake512_gpu_hash_80(int, unsigned int, void*)
  2.77%  383.50ms        53  7.2358ms  7.1146ms  7.3789ms  x11_simd512_gpu_compress1_64(int, unsigned int, unsigned long*, unsigned int*, uint4*, int*)
  2.69%  373.04ms        54  6.9082ms  6.8450ms  7.7322ms  quark_skein512_gpu_hash_64(int, unsigned int, unsigned long*, unsigned int*)
  2.55%  353.48ms        54  6.5459ms  6.5278ms  7.2944ms  quark_keccak512_gpu_hash_64(int, unsigned int, unsigned long*, unsigned int*)
  1.60%  221.22ms        53  4.1741ms  4.1419ms  4.2535ms  x11_simd512_gpu_final_64(int, unsigned int, unsigned long*, unsigned int*, uint4*, int*)

sp - 12d436ae1ecdc5e647a6a1576b98c4803510b13f
==25578== Profiling result:
Time(%)      Time     Calls       Avg       Min       Max  Name
 20.56%  6.72060s       127  *52.918ms  52.822ms  53.985ms  x11_echo512_gpu_hash_64(int, unsigned int, unsigned long*, unsigned int*)
 18.89%  6.17511s       128  48.243ms  48.147ms  53.860ms  quark_groestl512_gpu_hash_64_quad(int, unsigned int, unsigned int*, unsigned int*)
 12.65%  4.13517s       128  *32.306ms  32.181ms  36.017ms  x11_shavite512_gpu_hash_64(int, unsigned int, unsigned long*, unsigned int*)
 11.29%  3.69123s       128  28.838ms  28.787ms  30.680ms  x11_simd512_gpu_expand_64(int, unsigned int, unsigned long*, unsigned int*, uint4*)
  7.27%  2.37746s       128  18.574ms  18.547ms  20.732ms  x11_cubehash512_gpu_hash_64(int, unsigned int, unsigned long*, unsigned int*)
  5.35%  1.74723s       128  13.650ms  13.589ms  15.257ms  quark_jh512_gpu_hash_64(int, unsigned int, unsigned long*, unsigned int*)
  5.05%  1.65134s       128  12.901ms  12.699ms  14.372ms  x11_luffa512_gpu_hash_64(int, unsigned int, unsigned long*, unsigned int*)
  3.12%  1.01978s       128  7.9670ms  7.9183ms  8.0247ms  x11_simd512_gpu_compress2_64(int, unsigned int, unsigned long*, unsigned int*, uint4*, int*)
  2.91%  951.09ms       128  7.4304ms  7.4097ms  8.2771ms  quark_blake512_gpu_hash_80(int, unsigned int, void*)
  2.88%  941.80ms       128  7.3578ms  6.9981ms  8.3027ms  quark_bmw512_gpu_hash_64(int, unsigned int, unsigned long*, unsigned int*)
  2.83%  926.04ms       128  7.2347ms  7.0956ms  7.3374ms  x11_simd512_gpu_compress1_64(int, unsigned int, unsigned long*, unsigned int*, uint4*, int*)
  2.72%  887.67ms       128  6.9349ms  6.8876ms  7.7173ms  quark_skein512_gpu_hash_64(int, unsigned int, unsigned long*, unsigned int*)
  2.56%  836.91ms       128  6.5384ms  6.5282ms  7.2936ms  quark_keccak512_gpu_hash_64(int, unsigned int, unsigned long*, unsigned int*)
  1.63%  533.83ms       128  4.1705ms  4.1391ms  4.2481ms  x11_simd512_gpu_final_64(int, unsigned int, unsigned long*, unsigned int*, uint4*, int*)

Are you testing on linux too ? or just in windows ?

Still trying to get the same gains on windows... but that take a lof of time

[sp_]

I am only testing on windows. There was a small bug in the exe file I sendt out. Exe 7. I have fixed it, and noe I am preparing another checkin later today. Next kernal to be checked in is blake.

[sp_]

Faster blake (nist5, quark,x11 etc.).

http://www.filedropper.com/release8

fixed bug in release7 (unvalid nounces) (the bug was only present in the exe, because an old file was linked in instead of the latest)

source:

https://github.com/sp-hash/ccminer

[Epsylon3]

wow good game, didnt check blake512 for the moment, trying to fix x13 weird behavior during benchmark

but i was able to see the improvements on windows too with the previous commit

EDIT: + 10KH also with blake on the 750 ti

Code:

 
 20.56%  4.55172s        86  52.927ms  52.850ms  53.968ms  x11_echo512_gpu_hash_64(int, unsigned int, unsigned long*, unsigned int*)
 18.97%  4.19956s        87  48.271ms  48.164ms  53.870ms  quark_groestl512_gpu_hash_64_quad(int, unsigned int, unsigned int*, unsigned int*)
 12.70%  2.81199s        87  32.322ms  32.149ms  36.061ms  x11_shavite512_gpu_hash_64(int, unsigned int, unsigned long*, unsigned int*)
 11.33%  2.50956s        87  28.846ms  28.786ms  30.704ms  x11_simd512_gpu_expand_64(int, unsigned int, unsigned long*, unsigned int*, uint4*)
  7.30%  1.61676s        87  18.584ms  18.549ms  20.739ms  x11_cubehash512_gpu_hash_64(int, unsigned int, unsigned long*, unsigned int*)
  5.36%  1.18770s        87  13.652ms  13.590ms  15.225ms  quark_jh512_gpu_hash_64(int, unsigned int, unsigned long*, unsigned int*)
  5.08%  1.12451s        87  12.925ms  12.721ms  14.430ms  x11_luffa512_gpu_hash_64(int, unsigned int, unsigned long*, unsigned int*)
  3.09%  685.04ms        86  7.9656ms  7.9084ms  8.0212ms  x11_simd512_gpu_compress2_64(int, unsigned int, unsigned long*, unsigned int*, uint4*, int*)
  2.93%  648.66ms        87  7.4559ms  7.1070ms  8.3455ms  quark_bmw512_gpu_hash_64(int, unsigned int, unsigned long*, unsigned int*)
  2.84%  629.44ms        87  7.2350ms  7.1123ms  7.3753ms  x11_simd512_gpu_compress1_64(int, unsigned int, unsigned long*, unsigned int*, uint4*, int*)
  2.73%  604.12ms        87  6.9439ms  6.8900ms  7.7449ms  quark_skein512_gpu_hash_64(int, unsigned int, unsigned long*, unsigned int*)
  2.62%  579.68ms        87  *6.6630ms  6.6329ms  7.4384ms  quark_blake512_gpu_hash_80(int, unsigned int, void*)
  2.57%  569.19ms        87  6.5424ms  6.5284ms  7.2974ms  quark_keccak512_gpu_hash_64(int, unsigned int, unsigned long*, unsigned int*)
  1.62%  358.63ms        86  4.1702ms  4.1305ms  4.2341ms  x11_simd512_gpu_final_64(int, unsigned int, unsigned long*, unsigned int*, uint4*, int*)

[jpouza]

Targeting 10MH/s X11, keep pushing  

Disabling SLI things go higher, will print screen trying to hit 10MH/s.


hosting imagenes

[jpouza]

Maximum on the 980, limited by 1.2500v, voltage limitations of the reference cards


subefotos

[th00ber]

good job !

[sp_]

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

Bitslice it if you must. That will help you remove the memory issue.

Checked in a small boost by using the perm  instruction in whirlpool, but I think I have to rewrite the shared mem part to get 1/8th the memory reads.

[polanskiman]

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

Bitslice it if you must. That will help you remove the memory issue.

Checked in a small boost by using the perm  instruction in whirlpool, but I think I have to rewrite the shared mem part to get 1/8th the memory reads.

Bitslice would mean NO memory reads.

For 750ti:

With ccminer v7 from DJm34 I have an average of:
x11 = 2605 Khash

With ccminer by SP_ release 8 I have an average of:
X11 = 2800 Khash

That's a 195 Khash average difference. It wont make me any richer but it is always welcome

ccminer by SP_ release 8 seems to have a few bugs though. When you proceed with ctrl+c, most of the times ccminer will crash. Also when you are asked to terminate the batch job whether you say Y or N yields the same result: ccminer is closed.

[polanskiman]

By the way, any intentions on including and perhaps improving m7 algo in ccminer by SP_  releases?

[sp_]

Some of the bugs have been removed in the the Tvpouvet release that I forked. I will probobly refork. My focus is on the kernals, and only 50% of the kernals of x11 have been modded in the opensource.

I just recompiled with yesterdays NVIDIA driver (344.75) There seems to be a hashincrease on the 750ti of around 30KHASH.

[kingscrown]

this mod looks SICK!

[sp_]

EDIT: Just looked at Echo again. Quite well done, I tip my hat.

Instead of doing 10 rounds of echo, I do 9.25 rounds. This is because most of the first round is done on constant input. But someparts of round2 can also be precalculated... More boost is expected.
To be continued.

[sp_]

With ccminer v7 from DJm34 I have an average of:
x11 = 2605 Khash

With ccminer by SP_ release 8 I have an average of:
X11 = 2800 Khash

That's a 195 Khash average difference. It wont make me any richer but it is always welcome

You should try the 980. Up from 7MHASH to 9. 30% faster.

[jjjordan]

No improvement with GTX 970 whatsoever over 1.4.9.
Or maybe there was no such intended?

[sp_]

Kernals are beeing merged up to the head branch by tvpouvet. 1.4.9 and this mod based on 1.4.6 share the same optimalizations.
There is a small difference though. 1.4.9 has improved the throughput settings in the kernals. I need to do a refork, and redo launchbounds tweaks for 1.4.9.

[Bombadil]

Kernals are beeing merged up to the head branch by tvpouvet. 1.4.9 and this mod based on 1.4.6 share the same optimalizations.
There is a small difference though. 1.4.9 has improved the throughput settings in the kernals. I need to do a refork, and redo launchbounds tweaks for 1.4.9.

So it means you'll make use of his API too? I'm busy writing a .NET app that monitors your cudarigs with his API, with detailed stats etc, so that would be nice for comparisons

[sp_]

yes, When I do a refork, only some of the kernals will be different. On github alot of good changes and bugfixes have been done in 1.4.9. API support as well.
Makes it worth the upgrade

[polanskiman]

Some of the bugs have been removed in the the Tvpouvet release that I forked. I will probobly refork. My focus is on the kernals, and only 50% of the kernals of x11 have been modded in the opensource.

I just recompiled with yesterdays NVIDIA driver (344.75) There seems to be a hashincrease on the 750ti of around 30KHASH.

You are talking about this: https://github.com/tpruvot/ccminer ?