CCminer(SP-MOD) Modded GPU kernels.

[reb0rn21]

sp_ you know 1.09 was out to alfa testers before your open love 

[sp_]

sp_ you know 1.09 was out to alfa testers before your open love 

1.09 has 5 release numbers.  #5 was released yesterday after my latest commits-.

[abanamat]

I pushed a 4% faster fugue to github + some pool tuning. Run with fixed difficulty on the pool.
The enemy dude is taking my optimized code and merge it into his 1.09 exe file. (1% fee miner)
I have only shared speedups in 8 of the 16 algos. My private can beat beat enemy 1.09 r5 with 10% dev fee.

10% dev fee?! 

[Amph]

i'm getting 13.5mh with x16r on 1070ti at 60% tdp +100 core 400 mem, what are you guys getting?

[sp_]

Pushed more speed to github (x16R, X16S)

When simd is the previous algo in the chain:

Fugue64 +26% faster
Whirlpool64 +40% faster
Hamsi64 +40% faster

https://github.com/sp-hash/suprminer/releases/tag/spmod-git4

[sp_]

i'm getting 13.5mh with x16r on 1070ti at 60% tdp +100 core 400 mem, what are you guys getting?

Depends on the sequence numbers, and how long you meassure the average rate. Testing for a few hours isn't enough. Testing x16S is more accurate, since the order is permuted and not random (every hashing algo excecuted once)

in X16S you have 16! combinations (2.092278988 E+13), in X16R 16^16 (1.8 E+18)

[agente]

i'm getting 13.5mh with x16r on 1070ti at 60% tdp +100 core 400 mem, what are you guys getting?

Im doing 8mh with normal 1070 with this settings.. (x16r not x16s)

[sp_]

1 week ago enemy v1.08 was the fastest downloadable x16r miner.

spmod-git #4 is faster, free and opensource

here is spmod-git #3 in a 9 hour comparison between the 3 mods.

Enemy 1.09 beta 5 was compiled just after I released #3.  (He use my github code and add the speedups  to his private fee miner)

[Mr.Spider703]

Final Results of Enemy 1.09 BETA 5 Vs. Enemy 1.08 Vs. sp-mod git3

FINAL Normalized Average Results:

#1    *                | Enemy 1.09 Beta5     | 106.27 RVN
#2    -7.62%  | Enemy 1.08                  | 98.75 RVN
#3    -9.41%   | sp-mod git3                 | 97.13 RVN

[sp_]

The test was made with spmod-git #3. I have released spmod-git #4 free and opensource

#4 is faster than enemy 1.08

To beat #9, I cannot share my code because he will only copy my work again.

[Bimmber]

Who cares about 1.08 ?!   1.09 is current version.

[sp_]

Who cares about 1.08 ?!   1.09 is current version.

In the spmod-git, I have only optimized 8/16 algos. My 16/16 algo private miner is around +10% faster than enemy 1.09 r5.

[Bimmber]

Who cares what your private miner does? It can be +999%, its not for public.

You are like a parrot stuck with same. If you want to compete, then stop making a clown of yourself. Release your best miner and show what you can do.

[sp_]

My private was released 2 months ago.. (big farms) it also includes a fast x17 miner (24 mhash 1080ti).
Still available pm for details.

[PallMall]

I pushed a 4% faster fugue to github + some pool tuning. Run with fixed  on the pool.
The enemy dude is taking my optimized code and merge it into his 1.09 exe file. (1% fee miner)
I have only shared speedups in 8 of the 16 algos. My private can beat beat enemy 1.09 r5 with 10% dev fee.

Hey
what difficulty you set?

[Amph]

i'm getting 13.5mh with x16r on 1070ti at 60% tdp +100 core 400 mem, what are you guys getting?

Depends on the sequence numbers, and how long you meassure the average rate. Testing for a few hours isn't enough. Testing x16S is more accurate, since the order is permuted and not random (every hashing algo excecuted once)

in X16S you have 16! combinations (2.092278988 E+13), in X16R 16^16 (1.8 E+18)

still getting 14mh for some round of some algo, i know there are 16 round of algo right? some do less hash other more

the average is 66MH with 6 gpu, so 11mh, this is a 6 hours test already, btw this is with your mod sp version 3

[sp_]

btw this is with your mod sp version 3

You should upgrade to version 4

[NorbertColon]

SP_

Excellent work and thankyou.

[sp_]

still getting 14mh for some round of some algo, i know there are 16 round of algo right? some do less hash other more

yes. My X16s (permuted sequence) implementation will fluctate with the order of the algos. You can see in my latest commit that the memory access in simd is pipelined with another hashing algo. So while the gpu is waiting for the memory to be ready, it can continue to work on other work in shared mem or in registers.. (similar to the claymores dual miner)

[JackIT]

1 week ago enemy v1.08 was the fastest downloadable x16r miner.

spmod-git #4 is faster, free and opensource

here is spmod-git #3 in a 9 hour comparison between the 3 mods.

Enemy 1.09 beta 5 was compiled just after I released #3.  (He use my github code and add the speedups  to his private fee miner)

dude, what is wrong with you? 

I realize this is the internet... but please don't take the work I do and spin it into a narrative for your own benefit.

I spend a lot of time setting up and running these tests. And I run the tests as equally and fairly as possible. I don't care who wins or loses.


You can't cherry pick a round and declare victory. The fact is, that 1.08 is faster than spmod-git3, period

The reason I do three rounds, with each miner running on an instance one time, is that my test rig has a mix of 1080Ti cards, and as much as I've done to balance the instances so that they hash the same, they're not identical and some differences are to be expected.

Analyzing the weighted and normalized performance of the three instances in this test, Instance 1 and 2 performed within 1% of each other, Instance 3 performed 3-4% better than the other two.


If I had some sort of agenda and didn't care about actual valid results.. then I would have just continued Round 1 v1.0 and called it a day. Instead, I recognized that there was an issue and stopped the test. A graph like this: wouldn't be too appealing.