Topic: CCminer(SP-MOD) Modded NVIDIA Maxwell / Pascal kernels. - page 1165. (Read 2347599 times)

I saw same error on X15 today with r38

Was?  Ins algos x11,x13,x14, und fresh.  Vielleicht mehr...

Aber, es tut mir leid, ins Englisch, die worte "which" ein "h" haben.       --scryptr

Wich algos?

I am getting many "result does not validate on cpu" errors in various algos.  However, shares are still accepted.  In setting up auto-switching yaamp by using the "-r 0" round-robin flag on my Windows commandline, I have been seeing results on more algos than I usually mine.  The version running is v38, on a GTX 960.           --scryptr   

yes. for compute 5.0 (750,750ti) the speed is the same, but for compute 5.2 much faster. release37 had a drop in the speed for 960,970,980 cards.

compiles and runs really well sp ...

will test on the rest of the farm tomorrow ...

tanx ...

#crysx

Thanx for new release! It works. quark performance is nearly the same with r36 abd r37.
One moment - i don't now if it worth to point at - x13 don't work on default settings on Hamsterpool.
It works fine on Yaamp and Nicehash though ... )))

1.5.38(sp-MOD) is available here: (8-feb-2015)

https://github.com/sp-hash/ccminer/releases/tag/1.5.38

The sourcecode is available here:

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

Get the latest. Build 38 should work now.

Groestl is used in quark and x11 etc.

this commit "restore performance quark" https://github.com/sp-hash/ccminer/commit/9b94e71d6764cf8176d5b5321585cc28a55ce22e
seems to broke quark on gtx750

just for groestl? this is what im getting for x11 also mate ...

*** ccminer 1.5.37-git(SP-MOD) for nVidia GPUs by sp-hash@github ***
        Built with the 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.

[2015-02-08 18:46:03] Starting Stratum on stratum+tcp://192.168.0.205:3340/
[2015-02-08 18:46:03] GPU monitoring is not available.
[2015-02-08 18:46:03] Binding thread 0 to cpu 0 (mask 1)
0
[2015-02-08 18:46:03] Binding thread 2 to cpu 2 (mask 4)
2
[2015-02-08 18:46:03] Binding thread 1 to cpu 1 (mask 2)
1
[2015-02-08 18:46:03] Binding thread 3 to cpu 3 (mask
3
[2015-02-08 18:46:03] 5 miner threads started, using 'x11' algorithm.
[2015-02-08 18:46:03] Binding thread 4 to cpu 0 (mask 1)
0
[2015-02-08 18:46:03] Stratum difficulty set to 0.03125
[2015-02-08 18:46:06] GPU #4: result for ccee969d does not validate on CPU!
[2015-02-08 18:46:11] GPU #2: result for 6752aecd does not validate on CPU!
[2015-02-08 18:46:22] GPU #0: GeForce GTX 750 Ti, 2753 kH/s
[2015-02-08 18:46:22] GPU #3: GeForce GTX 750 Ti, 2770 kH/s
[2015-02-08 18:46:27] GPU #4: result for d07c0ff1 does not validate on CPU!
[2015-02-08 18:46:31] GPU #4: GeForce GTX 750 Ti, 2799 kH/s
[2015-02-08 18:46:34] GPU #2: GeForce GTX 750 Ti, 2750 kH/s
[2015-02-08 18:46:34] GPU #1: GeForce GTX 750 Ti, 2690 kH/s
[2015-02-08 18:46:45] GPU #3: result for 9d4ab3ae does not validate on CPU!

so maybe broke it for all?

#crysx

looks like the last groestl change broke the hash. will submit a fix.

tanx sp ...

#crysx

But release 36 is bugged. There was a bug in echo has that will reduce the number of yay!'s

I fixed the compiler errors. The change in skein hash reduced the performance in quark. I will modify it.

Confirm R37 fixed Qubit diff. No need -f 0.008 anymore. TQ.

getting compile errors sp - for the first time ...

edit - i should be more specific - sorry ...

when it gets to the groestl compile ...

bitslice_transformations_quad.cu(141): error: more than one instance of overloaded function "__shfl" matches the argument list:
            function "__shfl(int, int, int)"
            function "__shfl(float, int, int)"
            argument types are: (uint32_t, unsigned int, int)

bitslice_transformations_quad.cu(143): error: more than one instance of overloaded function "__shfl" matches the argument list:
            function "__shfl(int, int, int)"
            function "__shfl(float, int, int)"
            argument types are: (uint32_t, unsigned int, int)

bitslice_transformations_quad.cu(144): error: more than one instance of overloaded function "__shfl" matches the argument list:
            function "__shfl(int, int, int)"
            function "__shfl(float, int, int)"
            argument types are: (uint32_t, unsigned int, int)

20 errors detected in the compilation of "/tmp/tmpxft_00000e15_00000000-6_cuda_myriadgroestl.cpp1.ii".
20 errors detected in the compilation of "/tmp/tmpxft_00000e0e_00000000-6_cuda_groestlcoin.cpp1.ii".
make[2]: *** [cuda_myriadgroestl.o] Error 2
make[2]: *** Waiting for unfinished jobs....
make[2]: *** [cuda_groestlcoin.o] Error 2

#crysx

Speed is the same on X15...

-fixed echo hash
-fixed qubit default diff
-faster skein on compute 5.2
-faster groestl


1.5.37(sp-MOD) is available here: (7-feb-2015)

https://github.com/sp-hash/ccminer/releases/tag/1.5.37

The sourcecode is available here:

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

Yes.