CCminer(SP-MOD) Modded GPU kernels.

[Ayers]

X11:

My 1.5.74 does 14.5MHASH on the 980ti with OC

My private does around 15MHASH on the same clocks.(0.1BTC)

I am working to reach 15.5MHASH kernal in the sp-mod ultimate(simd rewrite).. Much more interesting than the stupid blake algos.. a 980ti@15.5 MHASH makes $4.6 a day on nicehash mining x11 with todays prices..


Sp-mod ultimate will contain these modded kernals:

1. x11,x13,x14,x15,x17
2. quark,nist5m
3. lyra2re,lyra2v2
4. Neoscrypt
5. Decred, Blake2s, Vcash, Pentablake

Refork of 1.7.5 with all the bugfixes and improvements by tvpruvot.

not really an improvement, since my 970 can do half of that (8m) on x11, and it is known that the 980 is double of that

[sp_]

not really an improvement, since my 970 can do half of that (8m) on x11, and it is known that the 980 is double of that

You are mixing the algos. a 980 cannot do 16MHASH@x11 with the public kernals. 1.5.74 (sp-mod release 74 cuda 6.5) can do close to 11MHASH with OC. SP-mod private #5 can do 11.5 MHASH with oc.

The gtx 970 has 1664 shaders. The gtx 980 has 2048 shaders.. The shader count indicate that the gtx 980 is 23% faster than the gtx 970.
The gtx 980ti has 2766 shaders.  The shader count indicate that the gtx 980ti is 66.22% faster than the gtx 970.

[sp_]

Another porblem for devs was cuda 7.5. It's release should have been an opportunity for further optimisations but it
was a flop. Most of the work on cuda 7.5 was to recover lost hash. There's no profit in that.

Cuda 7.5 is bether, but you need to recode and retune Your kernals.

Alexis have stolen some of my ideas in the decred, and made a faster opensource kernal:

http://cryptomining-blog.com/7718-new-ccminer-1-7-1-fork-by-alexs78-with-improved-vcash-and-decred-performance/

As you can see, the kernal is faster than the cuda 6.5 build for both Vcash and Decred.

Sp-mod release #8 does 3,3GHASH in the decred algo, and the Alexis kernal is doing 2,394 GHASH (Windows compile). (Source cryptomining blog)

here you can see the numbers.

This release with GTX 980 Ti intensity 31 Decred, 30 Vcash
 Vcash – 4453 MHS
 Decred – 2394 MHS

1.7.4 alexis78 release Compute 3.5 CUDA 6.5
 Vcash – 4418 MHS

Latest 1.7.6-git tpruvot with Pallas tweaks
 Decred – 2218 MHS

[bensam1231]

Thank you. Additionally, I try to fix my bugs BEFORE I sell something.

Huh... guess you haven't seen steam greenlight or kickstarter.

It's not about fixing them before, it's about fixing them in general. Just means you didn't get your code out the door as soon as you could.

This is completely putting aside that Wolf0 generally doesn't release anything and what he does he sells for a gajillion dollars to the highest bidders... but we aren't talking about that, are we?

Another porblem for devs was cuda 7.5. It's release should have been an opportunity for further optimisations but it
was a flop. Most of the work on cuda 7.5 was to recover lost hash. There's no profit in that.

Cuda 7.5 is bether, but you need to recode and retune Your kernals.

Alexis have stolens some of my ideas in the decred, and made a faster opensource kernal:

http://cryptomining-blog.com/7718-new-ccminer-1-7-1-fork-by-alexs78-with-improved-vcash-and-decred-performance/

As you can see, the kernal is faster than the cuda 6.5 build for both Vcash and Decred.

Sp-mod release #8 does 3,3GHASH in the decred algo, and the Alexis kernal is doing 2,394 GHASH (Windows compile). (Source cryptomining blog)

here you can see the numbers.

This release with GTX 980 Ti intensity 31 Decred, 30 Vcash
 Vcash – 4453 MHS
 Decred – 2394 MHS

1.7.4 alexis78 release Compute 3.5 CUDA 6.5
 Vcash – 4418 MHS

Latest 1.7.6-git tpruvot with Pallas tweaks
 Decred – 2218 MHS

I'd highly suggest focusing on merge mining SP, especially with how little utilization the GPU has for Ethereum... Claymore had a great idea and it seems to work. Nvidia hardware works great for Eth and has a lot of overhead to spare.

[sp_]

Here is decred sp-mod decred #8 With max overclocking. I use -i 31,9 on the 980ti and -i 29 on the 750ti.



Here is the sourcecode of the sp-mod #4:

https://github.com/sp-hash/ccminer/commit/3a726d90efd528ed386407bbd5b223583d80b378

[Velgelm]

I decided to test a version of the miner from Alexis78 on Decred
Again, the core of 960 and 970 graphics cards overclocked to 200MHZ.
Adjust the intensity taken from the BAT versions from SP
I compared the version Decred8Vcash6 vs Alexis78 latest from the git repository
Frankly, the results surprised me. Alexis78 great job! Your miner free, and the rate paid from SP!
SP - again there is work to do!
Alexis78 results :

SP_Decred8Vcash6

The competition continues

[sp_]

The competition continues

hehe. Why do you mine decred on the 960 and the 970. Pretty stupid when you can double Your profit With etherum or quark.
What intensities are you using?


The cryptomining blog is only getting  2394 MHS on the 980ti. (only 30% faster than the gtx 970 With 66% more cuda cores.) Why is Your build so much faster? Can you download their binaries please and test please.

[Velgelm]

Eth gives me (0.4 eth) or 3.9$ day(diff grow) 970 doing 18Mh 960 do 10Mh
Dcr gives me ~3-3.2 dcr or ~ 4.0-4.7$ (diff down) 970 do 1850Mh 960 do 1120Mh

[Velgelm]

The competition continues

hehe. Why do you mine decred on the 960 and the 970. Pretty stupid when you can double Your profit With etherum or quark.
What intensities are you using?


The cryptomining blog is only getting  2394 MHS on the 980ti. (only 30% faster than the gtx 970 With 66% more cuda cores.) Why is Your build so much faster? Can you download their binaries please and test please.

i compiled lastest git version on my PC and run with bat from your miner configs

[Velgelm]

you can download my compiled version here https://www.dropbox.com/s/x847dk636ap4x6f/ccminer.exe?dl=0

[sp_]

i compiled lastest git version on my PC and run with bat from your miner configs

I don't include the miner config in my releases. What does the bat look like.

[Velgelm]

960 ccminer -d 1 -i 26.9 -a decred --cpu-priority 1
970 ccminer -d 0 -i 28.9 -a decred --cpu-priority 2
this intensy gives better results

[sp_]

why don't you compile this:

https://github.com/sp-hash/ccminer/commit/3a726d90efd528ed386407bbd5b223583d80b378

copy the decred_ok.cu to decred.cu in the algo256 folder and rebuild.

And use -i 29.6 as intensity. It is "faster"

[pallas]

why don't you compile this:

https://github.com/sp-hash/ccminer/commit/3a726d90efd528ed386407bbd5b223583d80b378

copy the decred_ok.cu to decred.cu in the algo256 folder and rebuild.

And use -i 29.6 as intensity. It is "faster"

"faster" at the command line, slower at the pool ;-)

[sp_]

"faster" at the command line, slower at the pool ;-)

But in version #8 I have fixed the bug, and the hashrate is the same on the pool.

[sp_]

960 ccminer -d 1 -i 26.9 -a decred --cpu-priority 1
970 ccminer -d 0 -i 28.9 -a decred --cpu-priority 2
this intensy gives better results

If you remove the intensity parameter and run both the exe files without the -i my miner is much faster. Basicly you reduce the speed of my kernal by putting on an intensity brake. you should also remove the --cpu-priority parameter. My VCASH mod is only using 25 registers(little GPU Resources), and need a high intensity and overclock. The right 970 card can run stable at a 1600MHZ core for weeks.

[sp_]

decred #8 is using 29 registers

here is the speed with a proper launch configuration.

[tbearhere]

Same speed as quark... I just noticed that With sp private #5. 

[Velgelm]

what is the proper launch configuration. ?

[sp_]

Same speed as quark... I just noticed that With sp private #5.  

For sp-private #5 quark use -i 25 on the 980ti. The default is -i 24