CCminer(SP-MOD) Modded GPU kernels.

[s7icky]

[2015-10-14 11:17:35] quark.usa.nicehash.com:3345 quark block 2307097
[2015-10-14 11:17:35] GPU #0: GeForce GTX 980 Ti, 27665
[2015-10-14 11:17:35] GPU #1: GeForce GTX 980 Ti, 28538
[2015-10-14 11:17:35] GPU #2: GeForce GTX 750 Ti, 7109

Release 70 on windows 10

all 3 are 100+ on core.

[frazier34567]

Lyra2rev2 -i 20.2
6 750 ti 1.5.70 Hash 28720 (4786.66 per card)
6 750 ti 1.5.70 Commit 1115 Hash 28646   (4774.33  per card)

Windows 8.1 Pro  Driver 358.50

[Grim]

Lyra2rev2 -i 20.2
6 750 ti 1.5.70 Hash 28720 (4786.66 per card)
6 750 ti 1.5.70 Commit 1115 Hash 28646   (4774.33  per card)

Windows 8.1 Pro  Driver 358.50

gpu engine clock?

[nhutphan]

Ubuntu 14.04 - Release 70 - Run very simple .bat with default -i for Lyra2rev2

6 x 750ti +135 core/ +200 mem: 28880 kh/s
960: 5840 kh/s
970: 9760 kh/s

[ZenFr]

Ubuntu 14.04.
Cuda 7.5
Release 70 (one week ago vs. version 68).

Lyra2REv2 :
--------------
Faster for GTX 750 (3.4 MH/s vs 2.9) : very good job for the compute 5.0 :-)
Slower for GTX 960 (5.8 MH/s vs.6.2) :-(

I hope the best for the anniversary release

Edit : my release 70 was compiled one week ago (october 06) : the today one is a little slower on Quark (GTX 750 and 960) and just a very little bit faster on Lyra2REv2 with the GTX 750 (+ 50 kHash/s)

[sp_]

Use Cuda 6.5

My numbers are:

Lyra2v2 on the 750ti Asus strix +100 Mhz core and mem 5Mhash (windows,x86 cuda 6.5)

The gtx 750 is doing 4.3MHASH on 1500MHz core and modified bios. (Increased TDP)

[nhutphan]

Ubuntu 14.04 - Release 70 - Run very simple .bat with default -i for Lyra2rev2

6 x 750ti +135 core/ +200 mem: 28880 kh/s
960: 5840 kh/s
970: 9760 kh/s

I also compiled with CUDA 7.5

[ZenFr]

Use Cuda 6.5

My numbers are:

Lyra2v2 on the 750ti Asus strix +100 Mhz core and mem 5Mhash (windows,x86 cuda 6.5)

The gtx 750 is doing 4.3MHASH on 1500MHz core and modified bios. (Increased TDP)

In my case, Cuda 7.5 is faster than Cuda 6.5.
I don't know if i can change the BIOS of the 750 with Ubuntu, neither the GPU clock

[sp_]

I get 10.2MHASH on the gtx970 windforce G1 with +100Mhz on the core

[nhutphan]

Use Cuda 6.5

My numbers are:

Lyra2v2 on the 750ti Asus strix +100 Mhz core and mem 5Mhash (windows,x86 cuda 6.5)

The gtx 750 is doing 4.3MHASH on 1500MHz core and modified bios. (Increased TDP)

In my case, Cuda 7.5 is faster than Cuda 6.5.
I don't know if i can change the BIOS of the 750 with Ubuntu, neither the GPU clock

1. Mod Bios in Windows (fast & easy). However, I don't notice any different between TDP 38/46/65.5 on my Gigabyte gtx750ti

2. Enable overclock feature in Ubuntu: sudo nvidia-xconfig -a --cool-bits=28

[nhutphan]

Ubuntu 14.04 - Release 70 - Run very simple .bat with default -i for Lyra2rev2

6 x 750ti +135 core/ +200 mem: 28880 kh/s
960: 5840 kh/s
970: 9760 kh/s

Update: +105 core

960: ~ 6180 kh/s
970: ~ 10300 kh/s

[sp_]

But still slow. My modded lyra2v2 compute 5.2 opensource kernal is only using 7 threads per block.. and very little power..



DJm34's private kernal is already doing over 15MHASH on his GTX 980

If you look at my optimized quark kernal the gtx 970 is almost doing 3 times the speed of a 750ti.

But only 2x in the lyra2v2

[bensam1231]

Digibyte popped. Based on how hard it's getting hammered by miners right now I don't think it's much of a secret.

Seems as though there is some problems with myr-groestl and skein on ccminer. Comparing it to the Tpruvot version myr-gr, the SP version produces higher hashrates even though the hashrate is much lower poolside. The SP is reporting about 50% higher hashrate, however poolside it is operating about at 10% of the Trpuvot version.

Skein is in the same boat. SP is reporting it about 20% higher then the Tpruvot version, but poolside it's about 30% slower.

It seems as though the SP is also reporting a bunch of 'above share target' messages, where as the Tpruvot version is not.

Also worth noting it seems as though Tpruvot is using the CPU assisted method with -C and it pegs your processors for all the algos, however turning -C on for SP does not fix the aforementioned problems.

The only version of SP-CCminer that is working well with DGB right now is qubit.

None of this really matters though. Nvidia hardware is about on par, if not a bit slower then AMD hardware in all the above algos with publicly available miners (Pallas's included), so it's not going to matter much as the Ethereum horde switches over to DGB and crushes it.

[rednoW]

2 sp_

Latest github compile is 20-30khs slower then release 70 on lyra2v2. gtx750, win7x64, cuda6.5 x32 build.

[bathrobehero]

For n-factor 15 I get:
750 Ti: 1.26khs (-L 5 -l t5x16)
970: 3.27khs (-L 10 -l t13x24)

N-factor 16:
750 Ti: 0.53khs (-L5 -l t5x15)
970: 1.15 (-L 14 -l t13x28)

Correction: the numbers above are from Ubuntu with stock speeds and not from Windows.
Actual Windows + OC hashrates for N factor 16:
750 Ti: 0.36khs (-L 9 -l t5x14)
970: 0.97khs (-L 19 -l t13x23)

The N factor 15 hashrates are much closer (Win7+OC = Ubuntu Stock).

So yeah, LEO is linux only from here on out.

[Genoil]

Well someone did a very impressive job at creating the ultimate eth kernel, and now this dude is blatantly reselling this piece of craftsmanship: https://bitcointalk.org/index.php?topic=1205258.0 because he needs money! Holy batman copyspeed kernel, who made this!? I just need to know how it was done!

[bensam1231]

Well someone did a very impressive job at creating the ultimate eth kernel, and now this dude is blatantly reselling this piece of craftsmanship: https://bitcointalk.org/index.php?topic=1205258.0 because he needs money! Holy batman copyspeed kernel, who made this!? I just need to know how it was done!

And such is the way of private miners. I'm sure whoever made it has some pretty pissed off clients now. It would explain why Ethereum seems to be being mined into the ground more then 'free power'.

[chrysophylax]

Well someone did a very impressive job at creating the ultimate eth kernel, and now this dude is blatantly reselling this piece of craftsmanship: https://bitcointalk.org/index.php?topic=1205258.0 because he needs money! Holy batman copyspeed kernel, who made this!? I just need to know how it was done!

i posted a couple of questions to see what the answers will be ...

so far - nada ...

i have also 'heard' ( by means of pm ) that the whole thing is a ripoff and contains a trojan that will try and hack your wallet accounts ...

genoil - the chances are that if you and the other devs are having one hell of a time to get THOSE type of hashrates - then the app ( and the guy doing it ) is a lie to try and scam people out of their btc ...

in this case - it certainly is ...

be careful guys - especially windows based miners ( unless its from a well respected dev ) ...

and tanx to those of you who messaged about it ...

#crysx

[sp_]

2 sp_
Latest github compile is 20-30khs slower then release 70 on lyra2v2. gtx750, win7x64, cuda6.5 x32 build.

yes, I tested last night. The latest is slower on the 750/750ti. I will fix it.

[sp_]

Seems as though there is some problems with myr-groestl and skein on ccminer. Comparing it to the Tpruvot version myr-gr, the SP version produces higher hashrates even though the hashrate is much lower poolside. The SP is reporting about 50% higher hashrate, however poolside it is operating about at 10% of the Trpuvot version.

Looks like a problem with the diff on the poolside. Try to run with --diff 2 or --diff 0.5.
I haven't tested the skein algo in a while, but it worked on yaamp some months ago when I optimized it.

Some pools are cheating by accepting low difficulty shares and weight them equal to shares that are 2x more difficult to find.