[ANN] sgminer v5 - optimized X11/X13/NeoScrypt/Lyra2RE/etc. kernel-switch miner

[Zuikkis]

new kernel for the version in github/djm34 http://ge.tt/41D9Y872/v/0?c
(it should work with metallicjames is you use in lyra2RE.cl the blake256 code of metallicjames version.

current hashrate on my 290x at 1030/1250 with -g2 -I20 -w64   1.46MH/s (average after stabilization) was 1.32MH/s before

donation welcomed: VyF9fXUSt5iZH6GuA8WvUw4NrVhcybMiLA
(even though it was a paid job... vtc has fall off  ) so please makes updating the code worthy...

please also report problems as well as your speed (I can't reach the 1500MH mem clock... so I am a little curious)

I'm getting about 1.1MHs with 280x, but I'm getting a lot of "share above target" rejects on nicehash! Maybe 50% of rejects. This is probably the same thing others have reported, pool only sees half hashrate..

This is with latest sgminer from your github.

add "--difficulty-multiplier 2" to the command line (don't know exactly what is the equivalent for the conf file)

It's "difficulty-multiplier": "2",

This works on the command line or in the global part of the config, but not in pool specific settings... Which makes it useless for Nicehash.

The difficulty should be fixed in sgminer source.

[djm34]

new kernel for the version in github/djm34 http://ge.tt/41D9Y872/v/0?c
(it should work with metallicjames is you use in lyra2RE.cl the blake256 code of metallicjames version.

current hashrate on my 290x at 1030/1250 with -g2 -I20 -w64   1.46MH/s (average after stabilization) was 1.32MH/s before

donation welcomed: VyF9fXUSt5iZH6GuA8WvUw4NrVhcybMiLA
(even though it was a paid job... vtc has fall off  ) so please makes updating the code worthy...

please also report problems as well as your speed (I can't reach the 1500MH mem clock... so I am a little curious)

I'm getting about 1.1MHs with 280x, but I'm getting a lot of "share above target" rejects on nicehash! Maybe 50% of rejects. This is probably the same thing others have reported, pool only sees half hashrate..

This is with latest sgminer from your github.

add "--difficulty-multiplier 2" to the command line (don't know exactly what is the equivalent for the conf file)

It's "difficulty-multiplier": "2",

This works on the command line or in the global part of the config, but not in pool specific settings... Which makes it useless for Nicehash.

The difficulty should be fixed in sgminer source.

badman needs to make a new release and changes in algorithm.c the "256,256" of lyra2re kernel and replace it by "128,128"
(that would solve once for all expecting no change occurs on the pool side)

edit: just made the change on my github repositary

[jch9678]

sgminer binaries with optimized Lyra2RE algorithm (by djm34 from Dec. 18th) and support for multi-algo are available: https://www.nicehash.com/index.jsp?p=software#sgminer (thanks badman74 for windows build).

Do not forget to include ";f9=1.5" to the password parameter of your NiceHash multi-algo configuration if you want to enable Lyra2RE for multi-algo.

As far as drivers are concerned - probably the best results are achieved with 14.6/14.7 drivers.

Is the keccak unroll problem fixed?  It was broken in the dailies.  I can't get the dailies to make bins with ....ku0.bin ( I know asked in the nicehash thread but it's probably more relevant here )

Badman74, thank you for the build.  Do you have this same problem?

Not to my knowledge but then again I haven't tried that lately....
edit: ku0 works fine for me

I was able to test also.  ku0 does indeed work on your builds.  However, please don't take this the wrong because you've done so much for the community, but in my experience your builds aren't as stable as others.  I have no idea how that could be, could different compiler environments cause this, I really have no idea.  Testing your build, I tried to make a bitblock bin and not only 1 but 2 talkcoin bins got produced with different worksizes.  This happened to me in the past as well.  Very weird.  Again this is just my experience and very well could just be me doing something wrong.  I'll post my config for a critique when I get a chance.  Instead of trying to build bins with a multialgo config, I'm just going to build them one by one with separate configs and maybe that will help.  Anyway happy holidays to everybody.

[badman74]

new kernel for the version in github/djm34 http://ge.tt/41D9Y872/v/0?c
(it should work with metallicjames is you use in lyra2RE.cl the blake256 code of metallicjames version.

current hashrate on my 290x at 1030/1250 with -g2 -I20 -w64   1.46MH/s (average after stabilization) was 1.32MH/s before

donation welcomed: VyF9fXUSt5iZH6GuA8WvUw4NrVhcybMiLA
(even though it was a paid job... vtc has fall off  ) so please makes updating the code worthy...

please also report problems as well as your speed (I can't reach the 1500MH mem clock... so I am a little curious)

I'm getting about 1.1MHs with 280x, but I'm getting a lot of "share above target" rejects on nicehash! Maybe 50% of rejects. This is probably the same thing others have reported, pool only sees half hashrate..

This is with latest sgminer from your github.

add "--difficulty-multiplier 2" to the command line (don't know exactly what is the equivalent for the conf file)

It's "difficulty-multiplier": "2",

This works on the command line or in the global part of the config, but not in pool specific settings... Which makes it useless for Nicehash.

The difficulty should be fixed in sgminer source.

badman needs to make a new release and changes in algorithm.c the "256,256" of lyra2re kernel and replace it by "128,128"
(that would solve once for all expecting no change occurs on the pool side)

edit: just made the change on my github repositary

i did that to start when 256,256 didn't work

[badman74]

sgminer binaries with optimized Lyra2RE algorithm (by djm34 from Dec. 18th) and support for multi-algo are available: https://www.nicehash.com/index.jsp?p=software#sgminer (thanks badman74 for windows build).

Do not forget to include ";f9=1.5" to the password parameter of your NiceHash multi-algo configuration if you want to enable Lyra2RE for multi-algo.

As far as drivers are concerned - probably the best results are achieved with 14.6/14.7 drivers.

Is the keccak unroll problem fixed?  It was broken in the dailies.  I can't get the dailies to make bins with ....ku0.bin ( I know asked in the nicehash thread but it's probably more relevant here )

Badman74, thank you for the build.  Do you have this same problem?

Not to my knowledge but then again I haven't tried that lately....
edit: ku0 works fine for me

I was able to test also.  ku0 does indeed work on your builds.  However, please don't take this the wrong because you've done so much for the community, but in my experience your builds aren't as stable as others.  I have no idea how that could be, could different compiler environments cause this, I really have no idea.  Testing your build, I tried to make a bitblock bin and not only 1 but 2 talkcoin bins got produced with different worksizes.  This happened to me in the past as well.  Very weird.  Again this is just my experience and very well could just be me doing something wrong.  I'll post my config for a critique when I get a chance.  Instead of trying to build bins with a multialgo config, I'm just going to build them one by one with separate configs and maybe that will help.  Anyway happy holidays to everybody.

don't know why it would do that... i do get some extra bins built at times but that is because there are pools in my config that uses them
i don't think there is anything really different in my version that could make it build weird bins
don't know why it may happen to you but i have never had a bin built that wasnt specified in my config (other than ckolivas when something was spelled wrong...)

[go6ooo1212]

What are you guys getting for Neoscrypt efficiency? Hash per watt, power draw at the wall?

Dick.  But it appears to be most profitable on nicehash again; even with the salty bottlenecked neoscrypt.cl you released for us around Veteran's Day.  (thanks again)

Actually, that was the best I had at the time - also, you aren't entitled to jack shit from me, need I remind you? Complain to Vehre - his code got less than 100kh/s on a 280X. Would you rather have that instead?

For the rest of you, since I've gotten a lot of people asking, I made another SGMiner 5 build off the git master branch: https://litecointalk.org/index.php?topic=17533 -- see the bottom of the OP for the latest.

I taught it will be some Cristmas optimization gift  , but I reach only 322kh/s (neoscrypt) again with my 280X

[tlmoscow]

5xASUS 280x - 1560kH, 1150Wt whole rig.

[go6ooo1212]

I havent WATmeter unfortunately , cant tell the exact power-consumption...

[tlmoscow]

5xASUS 280x - 1560kH, 1150Wt whole rig.

1.36kh/s per W. Nice - undervolted, right?

littlebit

[cat77]

4 x 280x   1390Khs, 710W at wall
4 x 7950   1280KHs, 680W at wall

[xelbyb]

NeoScrypt

4x7950  710W

[semajjames]

What are you guys getting for Neoscrypt efficiency? Hash per watt, power draw at the wall?

Dick.  But it appears to be most profitable on nicehash again; even with the salty bottlenecked neoscrypt.cl you released for us around Veteran's Day.  (thanks again)

Actually, that was the best I had at the time - also, you aren't entitled to jack shit from me, need I remind you? Complain to Vehre - his code got less than 100kh/s on a 280X. Would you rather have that instead?

For the rest of you, since I've gotten a lot of people asking, I made another SGMiner 5 build off the git master branch: https://litecointalk.org/index.php?topic=17533 -- see the bottom of the OP for the latest.

Nice one Wolf0
                      yet another fine example of your crypto politics I'll give the miner a go asap

 

[xelbyb]

NeoScrypt

4x7950  710W

Hashrate?

1.060

[djm34]

what is the hashrate of the 750ti on neoscrypt ?
thanks

[katak]

Hello Guys,

Do we have any windows shminer which gives more than 320 Khash/s on a 280x in neoscrypt ?
I could see amazing amounts on this card on linux...

thanks in advance

[willrain]

Down are kernels and generated bins for X algos:
***
This kernel&bins are for Tahiti&Hawaii. I have no Kaiman and Pitcairn to test.

Thanks. Can somebody share optimized binaries for X13 for Pitcairn?

[willrain]

I'm selling them

I know ^__^

[willrain]

Then you're kind of shooting yourself in the foot, lol. Instead of buying it for cheap, you're asking for it to be published, which will of course kill any possible profit you could make with it... 

I understand, but binaries for x11 and some for x13 is already in public. But most of my cards is 7850

[damm315er]

What are you guys getting for Neoscrypt efficiency? Hash per watt, power draw at the wall?

Dick.  But it appears to be most profitable on nicehash again; even with the salty bottlenecked neoscrypt.cl you released for us around Veteran's Day.  (thanks again)

Actually, that was the best I had at the time - also, you aren't entitled to jack shit from me, need I remind you? Complain to Vehre - his code got less than 100kh/s on a 280X. Would you rather have that instead?

For the rest of you, since I've gotten a lot of people asking, I made another SGMiner 5 build off the git master branch: https://litecointalk.org/index.php?topic=17533 -- see the bottom of the OP for the latest.

Did you ever get around to identifying your neoscrypt kernels so that people could tell which is which?

[damm315er]

What are you guys getting for Neoscrypt efficiency? Hash per watt, power draw at the wall?

Dick.  But it appears to be most profitable on nicehash again; even with the salty bottlenecked neoscrypt.cl you released for us around Veteran's Day.  (thanks again)

Actually, that was the best I had at the time - also, you aren't entitled to jack shit from me, need I remind you? Complain to Vehre - his code got less than 100kh/s on a 280X. Would you rather have that instead?

For the rest of you, since I've gotten a lot of people asking, I made another SGMiner 5 build off the git master branch: https://litecointalk.org/index.php?topic=17533 -- see the bottom of the OP for the latest.

Did you ever get around to identifying your neoscrypt kernels so that people could tell which is which?

Well, now there's really just one major one, and the Hawaii one that wasn't a good idea.

Well, it wasn't until I downloaded the new SG5-wolf from yesterday and compared the neoscrypt files that I found out that it had the same 2 month old kernel..  For some reason I was expecting a different neoscrypt kernel in it.