CCminer(SP-MOD) Modded GPU kernels.

[rednoW]

Neoscrypt works well for me in release 54.

Yes, git compiled from this commit https://github.com/sp-hash/ccminer/commit/4958b610a77e5a54d5c9cf1e2d18ca0b48513b9c neoscrypt works just fine at nicehash and hamsterpool

[sp_]

-Quark +200khash on the gtx 970 +300khash on the gtx 980
-Lyra2 fixed on the gtx960
-Fixed bufferoverflow(stack) in neoscrypt
-Can be compiled with cuda 7.5 but cuda 6.5 is still the fastest.

1.5.55(sp-MOD) is available here: (14-07-2015)

https://github.com/sp-hash/ccminer/releases/tag/1.5.55

The sourcecode is available here:

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

[johnambush]

If you have flooded the pool with rejected shares the pool might ban your ip adress. Change the ip

It may be the reason. I forgot about that. I can't do it right now but I will report you if it was or not soon as I check it. Thank you. Two heads are often smarter than one
EDIT: But I also tried to mine on Suprnova (for the first time with neoscrypt) and had same result. I doubt it banned my ip  

Look's like neoscrypt is broken in release 54.
I suspect there are some errors in the vector implementation done in the lyra2re rewrite since they share some code.

if there are errors they are not from me...  
(and the current public version of neoscrypt use different verctor types...)

Obviously my ip was banned. I relocated miner to another router and it established connection with pool immediately. But reason for ban ip may be flooding pool with bad shares due to broken neoscrypt. Works fine now

[Slava_K]

-Quark +200khash on the gtx 970 +300khash on the gtx 980
-Lyra2 fixed on the gtx960
-Fixed bufferoverflow(stack) in neoscrypt
-Can be compiled with cuda 7.5 but cuda 6.5 is still the fastest.

1.5.55(sp-MOD) is available here: (14-07-2015)

https://github.com/sp-hash/ccminer/releases/tag/1.5.55

The sourcecode is available here:

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

Can`t compile last commit - errores after this line in cuda_noescrypt.cu //   neoscrypt_gpu_hash_orig << <grid, block >> >(threads, startNounce, d_NNonce[thr_id]); (expectet an expression)
PS. Problem is solved. My PSU is not enough for 2x980 and 960 + 100% load for CPU... :,(

[pallas]

-Quark +200khash on the gtx 970 +300khash on the gtx 980
-Lyra2 fixed on the gtx960
-Fixed bufferoverflow(stack) in neoscrypt
-Can be compiled with cuda 7.5 but cuda 6.5 is still the fastest.

1.5.55(sp-MOD) is available here: (14-07-2015)

https://github.com/sp-hash/ccminer/releases/tag/1.5.55

The sourcecode is available here:

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

Thanks! It is indeed faster on 970, but lyra doesn't work, it doesn't validate on CPU. It also segfaults at times when exiting. Both problems where present on the previous version as well.
Need some debug info?

[Slava_K]

Thanks! It is indeed faster on 970, but lyra doesn't work, it doesn't validate on CPU. It also segfaults at times when exiting. Both problems where present on the previous version as well.
Need some debug info?

Set -i 17.5

[pallas]

Thanks! It is indeed faster on 970, but lyra doesn't work, it doesn't validate on CPU. It also segfaults at times when exiting. Both problems where present on the previous version as well.
Need some debug info?

Set -i 17.5

Thanks but same validation error.
Besides, the segfault is on all algos.

[Slava_K]

Thanks! It is indeed faster on 970, but lyra doesn't work, it doesn't validate on CPU. It also segfaults at times when exiting. Both problems where present on the previous version as well.
Need some debug info?

Set -i 17.5

Thanks but same validation error.
Besides, the segfault is on all algos.

what os and drivers?

[pallas]

Thanks! It is indeed faster on 970, but lyra doesn't work, it doesn't validate on CPU. It also segfaults at times when exiting. Both problems where present on the previous version as well.
Need some debug info?

Set -i 17.5

Thanks but same validation error.
Besides, the segfault is on all algos.

what os and drivers?

Ubuntu 15.04 with distribution cuda toolkit (6.5 and compute 5.0)

[Slava_K]

Thanks! It is indeed faster on 970, but lyra doesn't work, it doesn't validate on CPU. It also segfaults at times when exiting. Both problems where present on the previous version as well.
Need some debug info?

Set -i 17.5

Thanks but same validation error.
Besides, the segfault is on all algos.

what os and drivers?

Ubuntu 15.04 with distribution cuda toolkit (6.5 and compute 5.0)

Nv have a patch for 6.5 for 5.2 - gtx970 is 5.2 card (may be problem is here)

[joblo]

Thanks! It is indeed faster on 970, but lyra doesn't work, it doesn't validate on CPU. It also segfaults at times when exiting. Both problems where present on the previous version as well.
Need some debug info?

Set -i 17.5

Thanks but same validation error.
Besides, the segfault is on all algos.

Ubuntu 15.04 with distribution cuda toolkit (6.5 and compute 5.0)

Lyra2 works on a 970 for me (compiled for 5.2). Segfaults on ^C still occur but don't appear to have any
residual effect.

I've noticed the default intensity is too high for some cards/algos. Ive also noticed some odd behaviour
with intensity. Quark would initially run with -i 23.5. The next time i started it it failed at -i 23.5 but
ran at 23. Then 23 failed and I lowered it to 22.5.  970 + 780ti.

[hashbrown9000]

@sp_ , what is the default intensity for neoscrypt algo on a 750ti?  do you have any optimal settings? one of my cards crashed @ +80 MHz GPU clock

[chrysophylax]

Thanks! It is indeed faster on 970, but lyra doesn't work, it doesn't validate on CPU. It also segfaults at times when exiting. Both problems where present on the previous version as well.
Need some debug info?

Set -i 17.5

Thanks but same validation error.
Besides, the segfault is on all algos.

what os and drivers?

Ubuntu 15.04 with distribution cuda toolkit (6.5 and compute 5.0)

hi pallas ...

this may help with all future cards ...

https://developer.nvidia.com/cuda-gpus/ ...

#crysx

[bensam1231]

Hmmm Quark hit 70GH...

[sp_]

Ubuntu 15.04 with distribution cuda toolkit (6.5 and compute 5.0)

You need to build for compute 5.2. 5.0 builds won't work on gtx 970 etc.

[chrysophylax]

aaand - there it is ...

http://www.abc.net.au/news/2015-07-15/cern-scientists-discover-new-kind-of-particle-the-pentaquark/6620308 ...

a new algo maybe? ...

...

#crysx

[pallas]

Ubuntu 15.04 with distribution cuda toolkit (6.5 and compute 5.0)

You need to build for compute 5.2. 5.0 builds won't work on gtx 970 etc.

Is this for lyra only? The other algos appear to be working fine. Currently hashing quark at 15 Mh/s stock clocks.

[sp_]

I get 16.2 MHASH@quark with compute 5.2 build for x86 in the opensource version. (gigabyte windforce 970)

Compute 5.2 has bether launchconfigurations.

[sp_]

In the code you have different launch configurations for different compute versions like here in skein:

__global__
#if __CUDA_ARCH__ > 500
__launch_bounds__(480, 3)
#else
__launch_bounds__(240, 6)
#endif
void quark_skein512_gpu_hash_64(uint32_t threads, uint32_t startNounce, uint64_t * const __restrict__ g_hash, const uint32_t *

On the 970/980/980ti I run 480 gputhreads per block, but on 750ti only 240 threads per block. More threads use more power and generates more heat, but sometimes also more hash.


etc.

[sp_]

Is this for lyra only? The other algos appear to be working fine. Currently hashing quark at 15 Mh/s stock clocks.

It should be fixed, because I think the compute5.0 builds will be bether in the summertime. (less power, less heat and lower hashrates)