SGminer 5.3.1 better speed for Skein algo

[mkarhade]

Hi, Rather than using Cgminer for Skein algo, please use SGminer 5.3.1 which does skein at about 40% higher rate. That will be more profitable . Thank you

UPDATE: NVM you are using sgminer but the intensity is set to 11 and work size to 64 which is too low for any newer AMD card. Work size 256 and intensity 22-23 works best. So you might want to integrate that

[aaronsace]

What card are you using?

If anyone has any suggestions on two versions of the settings (one safe for old cards and one optimal for newer cards) then I would appreciate it:

    Lyra2z = ' --worksize 32 --intensity 18'
    Equihash = ' --gpu-threads 2 --worksize 256'
    Cryptonight = ' --gpu-threads 1 --worksize 8 --rawintensity 896'
    Ethash = ' --gpu-threads 1 --worksize 192 --xintensity 1024'
    Sia = ''
    Yescrypt = ' --worksize 4 --rawintensity 256'
    BlakeVanilla = ' --intensity d'
    Lyra2RE2 = ' --gpu-threads 2 --worksize 128 --intensity d'
    Skein = ' --gpu-threads 2 --worksize 64 --intensity 11'
    Qubit = ' --gpu-threads 2 --worksize 128 --intensity d'
    NeoScrypt = ' --gpu-threads 1 --worksize 64 --intensity 13 --thread-concurrency 64'
    X11 = ' --gpu-threads 2 --worksize 128 --intensity d'
    MyriadGroestl = ' --gpu-threads 2 --worksize 64 --intensity d'
    Groestl = ' --gpu-threads 2 --worksize 128 --intensity d'
    Keccak = ''
    Scrypt = ''

[mkarhade]

Hi Aaron ! I am using RX480 and Skein algo is the one that I am pointing out Skein = ' --gpu-threads 2 --worksize 256 --intensity 23' works best for me. about 250-270mh Skein at 70W power usage with rock stable performance. With intensity 11 and worksize 64 it is too low

[aaronsace]

I'll update the settings on that algorithm.

Do the other algorithms perform as expected? I have an RX480 but tweaking setting takes forever so I haven't checked that they are the best.

I'm worried they will crash older cards as well so I don't like to mess with them unless anyone reports it.

[bandit420]

Sgminermph.txt

Still using the "old" SGminer 5.3.1 from Miningpoolhub and in my opinion still the fastest, and can also confirm its faster with the mentioned parameters (5*RX480, around 1.3 GH).

[aaronsace]

The only improvement to the actual miner that he made was adding skein; everything else is the same. Odd if NiceHash added a slower version though...? Have you recorded the speed difference?

I can add it back; is MiningPoolHub's fork definitely faster?

[jjziets]

I am not getting 5.3.1 to work on any of my systems so can't test. But my 6x RX480 8Gb gets 1,55 GH/s with reorder/cgminer.exe. (i 13) That looks to be the sam as what @bandit420 get with 5.3.1. I can push this to +300 by lifting my power limits. But running at 89C is not good for the cards so I am happy with 270 at 75c

"--gpu-threads 2 --worksize 64 --intensity 13"

[bandit420]

worksize256_i_22.txt worksize_256i_d.txt worksize_64_i_d.txt worksize64_i11.txt

So weird thing: reorder and sgminer (nicehash 5.6.1) allways failing. But sgminer is also not recognizing the kernel and falls back to the standard ckolivas kernel. Maybe a problem of cpu, because I had only 1 core free to test of 8. Worksize 256 and Itensity d with sgminer from MPH seem to be the best results (labeld "sgminermph" on the benchmarks).

[bandit420]

Just as an addition: running -15% on the core and +150 on memory clocks, and newest driver (AMD 17.5.2) for the benchmarks.

[lekiet0917]

Please give me sgminer release for skein and conf. Thanks

[imrecsoka]

when you specify gpu threads and worksize do they really help in getting a better hash rate? sorry i am living in the past , i am just catching up . i have 3 RX470 cards i really hope they are worth while I used to get 720 mh/s but now i get 650 mh/s i am not sure why (its been months that have gone by and i am also using a different windows 7 version , though i have more memory now 8gb instead of 4gb. Same board and cpu and hard drive. My hard drive is really old though. Trying to figure out if i can lower voltage and raise the mhz core , not sure how relevant the memory speed is ?