NSGminer v0.9.4: The Fastest NeoScrypt GPU Miner

[wreckingball]

thank you very much, I actually got it working using just the place you mentioned.  I just assumed that scrypt-n was neoscrypt because the miner at the beginning of the thread has that as the only option other than scrypt and sha-256, anyway, the miner at the beginning of the thread doesn't work.  Thanks again, much obliged. 

[tmuir12]

We also have a GPU miner too, its still Beta and not yet optimised but it does work.

You can download it from the links in this thread.
https://bitcointalk.org/index.php?topic=821738.0

[ghostlander]

I just assumed that scrypt-n was neoscrypt because the miner at the beginning of the thread has that as the only option other than scrypt and sha-256, anyway, the miner at the beginning of the thread doesn't work.  Thanks again, much obliged. 

Maybe you could start with the built-in help?

Code:

$ minerd --help
Usage: minerd [OPTIONS]
Options:
  -a, --algo=ALGO       specify the algorithm to use
      neoscrypt  NeoScrypt(128, 2, 1) with Salsa20/20 and ChaCha20/20 (default)
      altscrypt  Scrypt(1024, 1, 1) with Salsa20/8 through NeoScrypt
      scrypt     Scrypt(1024, 1, 1) with Salsa20/8
      sha256d    SHA-256d

And yes, it does work.

[pravin]

i request FTC to create C# implementation of neoscrypt.

[Conqueror]

Why?

[BitAvrillis]

Heyo,
the Idea is interesting and I like it  but I need to configure all my miners new and the settings dont wanna work like before .... I use cgminer 3.7.7b neoscrypt and it is working but I have only 40kh/s instead of 650khs/s like before. Can anyone help to find the optimal settings for the R9 280X GV-R928XOC-3GD-GA   (-I 13 -g 2 -w 256 --thread-concurrency 8192 --gpu --powertune 20     isnt working that good anymore )

[tmuir12]

Heyo,
the Idea is interesting and I like it  but I need to configure all my miners new and the settings dont wanna work like before .... I use cgminer 3.7.7b neoscrypt and it is working but I have only 40kh/s instead of 650khs/s like before. Can anyone help to find the optimal settings for the R9 280X GV-R928XOC-3GD-GA   (-I 13 -g 2 -w 256 --thread-concurrency 8192 --gpu --powertune 20     isnt working that good anymore )

You will probably need to lower the intescity slightly from what you used for Scrypt, but the Hash rate is considerably lower for Neoscrypt than Scrypt.
For exaple my GPU gave me 370kh/s on Scrypt but on 29 to 30kh/s for NeoScrypt.
The fact it is lower is not an issue because it is lower for everyone, your GPU will also run cooler too.
I'm sure over the coming months there will be optimisation made to the mining software and the value will go up slowly over time, but highly unlikely ever back to what you got with Scrypt.

If you haven't already I suggest joining the feathercoin forum and have a read through the threads there as people are just starting to post their optimal settings for Neoscrypt.
https://forum.feathercoin.com/

[alt-ru-ist]

Heyo,
the Idea is interesting and I like it  but I need to configure all my miners new and the settings dont wanna work like before .... I use cgminer 3.7.7b neoscrypt and it is working but I have only 40kh/s instead of 650khs/s like before. Can anyone help to find the optimal settings for the R9 280X GV-R928XOC-3GD-GA   (-I 13 -g 2 -w 256 --thread-concurrency 8192 --gpu --powertune 20     isnt working that good anymore )

-w48, not -w256

[MaxDZ8]

Explain me this worksize 48 thing.
I'm very surprised. AMD Radeon has native hardware work size 64, wsize 48 implies 25% performance trashed... is this supposed to happen?

[bigtop1967]

w48 causes lots of HW errors for me

w64 and I=12 gets me about 200kh - but still with hw errors

Saphire 280X Toxic

[chris190]

With a dual processor intel xeon e5 2680v2 @ 2.80 GHz (20c 40t), I can get a good result?

I'm undecided at which cryptocurrencies can i mine! (Pts, gap, xpt, mmc, riecoin, yac ...)

[Conqueror]

With a dual processor intel xeon e5 2680v2 @ 2.80 GHz (20c 40t), I can get a good result?

I'm undecided at which cryptocurrencies can i mine! (Pts, gap, xpt, mmc, riecoin, yac ...)

I can suggest PXC or FTC.

[chris190]

With a dual processor intel xeon e5 2680v2 @ 2.80 GHz (20c 40t), I can get a good result?

I'm undecided at which cryptocurrencies can i mine! (Pts, gap, xpt, mmc, riecoin, yac ...)

I can suggest PXC or FTC.

thanks! 

[ghostlander]

neoscrypt_vliw.cl v3

HD6970: 19.5KH/s to 50.5KH/s and no HW errors!

Thanks to a much better global memory management with burst writes. Also tried a trick with copying a workgroup buffer to local memory and writing back asynchronously to global memory through async_work_group_copy(), but it didn't work out (36KH/s and 15% of HW errors). A 2.5x speed increase is something to celebrate anyway. Anyone cares to donate towards drinks, whores and blackjack?

[jkister]

HD6970: 19.5KH/s to 50.5KH/s and no HW errors!

hmm - i replaced my neoscrypt.cl with yours then deleted my bin file.  when sgminer goes to recreate it:

[01:20:47] Building binary neoscryptCaymangw128l8tc7000.bin
[01:20:47] Error -11: Building Program (clBuildProgram)
 [01:20:47] "/tmp/OCL18375T14.cl", line 596: error: identifier "WORKGROUPSIZE" is undefined
  __kernel __attribute__((reqd_work_group_size(WORKGROUPSIZE, 1, 1)))
                                               ^

1 error detected in the compilation of "/tmp/OCL18375T14.cl".

Frontend phase failed compilation.

any idea what i've done wrong?  not sure if 'workgroupsize' has anything to do with 'worksize' but ive tried it at 128, 256, and undefined.

[nuxxer]

isnt scrypt algo will be well done with hardware ?

[Conqueror]

isnt scrypt algo will be well done with hardware ?

Not exactly sure what you ask.
Anyway, Neoscrypt is technicaly superior to scrypt in all aspects (security, speed, temperature...) at CPU/GPU domain of course...

[ghostlander]

HD6970: 19.5KH/s to 50.5KH/s and no HW errors!

hmm - i replaced my neoscrypt.cl with yours then deleted my bin file.  when sgminer goes to recreate it:

[01:20:47] Building binary neoscryptCaymangw128l8tc7000.bin
[01:20:47] Error -11: Building Program (clBuildProgram)
 [01:20:47] "/tmp/OCL18375T14.cl", line 596: error: identifier "WORKGROUPSIZE" is undefined
  __kernel __attribute__((reqd_work_group_size(WORKGROUPSIZE, 1, 1)))
                                               ^

1 error detected in the compilation of "/tmp/OCL18375T14.cl".

Frontend phase failed compilation.

any idea what i've done wrong?  not sure if 'workgroupsize' has anything to do with 'worksize' but ive tried it at 128, 256, and undefined.

SGminer should define it as WORKSIZE. If even this fails, try a safe way.

Replace the following in the kernel:

Code:

/* NeoScrypt core engine:
 * N = 128, r = 2, p = 1, salt = password */
__kernel __attribute__((vec_type_hint(uint4)))
__kernel __attribute__((reqd_work_group_size(WORKGROUPSIZE, 1, 1)))
__kernel void search(__global const uint4* restrict password,
  volatile __global uint* restrict output, __global uint16* globalcache,
  const uint target) {
    uint glbid = get_global_id(0);
    uint grpid = get_group_id(0);
    uint lsize = WORKGROUPSIZE;
    uint lclid = glbid & (WORKGROUPSIZE - 1);

    __global uint16 *G = &globalcache[(grpid * WORKGROUPSIZE) << 9];

with this:

Code:

/* NeoScrypt core engine:
 * N = 128, r = 2, p = 1, salt = password */
__kernel __attribute__((vec_type_hint(uint4)))
__kernel void search(__global const uint4* restrict password,
  volatile __global uint* restrict output, __global uint16* globalcache,
  const uint target) {
    uint glbid = get_global_id(0);
    uint grpid = get_group_id(0);
    uint lsize = get_local_size(0);
    uint lclid = glbid & (lsize - 1);

    __global uint16 *G = &globalcache[(grpid * lsize) << 9];

It's going to reduce performance by about 1% though.

In addition, don't try to set thread concurrency to anything other than power of 2, i.e. 1024, 2048, 4096... It will only produce HW errors otherwise.

[jkister]

SGminer should define it as WORKSIZE. If even this fails, try a safe way.

Perfect.  Changing to WORKSIZE did fix the problem - but the GPUs were going sick/dead very quickly.  I ended up having to run with i12 -- i13 or above and the gpus get unhappy.

In addition, don't try to set thread concurrency to anything other than power of 2, i.e. 1024, 2048, 4096... It will only produce HW errors otherwise.

Oh, okay.  I set tc4096 while doing the above -

[nuxxer]

isnt scrypt algo will be well done with hardware ?

Not exactly sure what you ask.
Anyway, Neoscrypt is technicaly superior to scrypt in all aspects (security, speed, temperature...) at CPU/GPU domain of course...

i mean, to get big gain in scrypt algo, only using hardware than software ?