New sgminer support x11, x13, x14 and x15, fresh (with source)

[aznboy84]

Just realized that i might need a thread for my sgminer build for further support, so here goes ...

Optimized x11, x13, x14 and new x15 kernel

There also a "bitblockold" and "x14old" kernel working with older GPUs (78xx, 68xx ...)

Expecting X15 hash rate on 14.6 driver :

270X : 1.85MH/s (1100/1400 engine/mem)
7950 : 2.4MH/s (1050/1250  engine/mem) (mine was undervoltage / can't push engine clock)
280X : 2.9MH/s (1100/1400 engine/mem)

Github source :
https://github.com/aznboy84/sgminer/tree/v5_0-x15
https://github.com/aznboy84/freshgpu (freshcoin version )

Win32 binary :
X15_AMD_06_30.rar
X15_AMD_06_29.rar
FRESHCOIN SGMINER

there's 3 batch file for 7950, 270x and 280x and 3 pre-built binary kernel if you don't want to upgrade to 14.6, but hash rate would be 5% slower compared to 14.6 driver, just edit your pool, worker, password and run

There might be 2 "gpu-platform" on driver 14.6 (no idea why) so if your miner exit immediately after run just try  --gpu-platform (0 or 1) then it should work

How to make your own windows build

1 - Download dist files and extract to <your source path>/winbuild/
https://mega.co.nz/#!kIV1gDID!WsAh2XQo_YBIy3Ho5xLRnmMjj6V0Fh_JVS1OZOl5Zgw
(Those dist files does have x64 pre-compiled libraries so you can build x64 version too)
2 - Download ADL_SDK and extract adl_defines.h, adl_sdk.h, adl_structures.h from the archive include folder to <your source path>/ADL_SDK
http://developer.amd.com/tools-and-sdks/graphics-development/graphics-development-sdks/display-library-adl-sdk
3 - Get Visual Studio 2010 Express
http://www.visualstudio.com/en-us/downloads#d-2010-express
4 - Open <your source path>/winbuild/sgminer.sln, change configuration from Debug to Release, then press F7 to build, the compiled sgminer.exe is located at <your source path>/winbuild/output/x86/Release

some help on linux (BAMT, PIMP) build
(don't have any exprience on BAMT or PIMP, after complied the rest is up to you )

cd /<whatever folder you want>
git clone -b v5_0-x15 https://github.com/aznboy84/sgminer.git
cg sgminer
sh autogen.sh
CFLAGS="-O2 -Wall" ./configure
make

Fresh build instructions

cd /opt
git clone https://github.com/aznboy84/freshgpu.git
cd freshgpu
cp -av /opt/ADL-SDK/*.h ./ADL_SDK/
git submodule update --init --recursive
sh autogen.sh
CFLAGS="-O2 -Wall -march=native" ./configure
make

UPDATE 1 : Removing SPH_LUFFA_PARALLEL (Win32 binary rebuild too) guess someone getting few HW because of that

UPDATE 2 : Added X14 support into github source, but don't think you gonna need it because hash rate was about the same as the other miner (same code lol)
There also a x14old kernel if you wanna try

[havikk]

Great work!

Did you have any luck in resolving the issue with using both bitblock & bitblockold kernels at the same time?

Thanks!

[mammix2]

good work aznboy84, keep updating it and this could be 'The Place to go' for X15 mining information

[platinum4]

What's up dude, chiming in here since I'm usually over in the sgminer_5 multi-algo thread.

Your optimizations are pretty killer, pulling 3.6MH on a 290X.

It should be noted though I think that SPH_LUFFA_PARALLEL 1 is a killer for some cards that tend to go SICK, so people should maybe throttle back their overclocks.

I tried this version as is, with OC 1040/1300 and it threw a [questionable] card into SICK within 15 minutes, temps dropped on it to 40C and it basically just choked on a thread.  Scaled that card back to 1010/1250 [290X] and the uptime seems to be holding.

As you don't have any 290/X to test personally on, I'll definitely help wherever I can.  And I really like that you are releasing Windows binaries at the same time if not before the Linux compiles.  It was always a bitch and a half to wait for one of the few Windows compiling people to wake up and have the time to do that.

Very much fucking appreciated.  And I've backed you elsewhere in multiple threads already, as you may have seen dude.

Thanks sincerely for your contributions to the mining space so far.

Of course, if you have any more pull requests for sgminer v5_0 tree, please do not hesitate, ystarnaud has a shitload of commits going on over there as well; hopefully we can stay away from divergent forks, but whatever works right?

[divo2014]

very good job!!!! Badly that in conf  it is impossible to specify bitblock,bitblockold,chtoby at once simultaneously worked I am sorry for bad English

[aznboy84]

@divo2014, @havikk : i find out it's kidda hard to get multiple kernel into a pool configuration on this version, just threw away my 2 hours on that and still at nowhere
@platinum4 : SPH_LUFFA_PARALLEL 1 doesn't increase hash rate much so i guess just gonna remove it (copied from my x11 + x13 optimized )

[divo2014]

x11 + x13 optimized -very good job!!!! +5%

[ibfragalot2]

We need to be able to use bitblockold and bitblock at the same time... many people have old + new cards in the same rig.

Trying to bypass it by ignoring the old card(s) always results in a freeze too, doh.

[bullus]

We need to be able to use bitblockold and bitblock at the same time... many people have old + new cards in the same rig.

Trying to bypass it by ignoring the old card(s) always results in a freeze too, doh.

The problem lies in the v5 version from the beginning it is not supported. So I had to get rid of my 6970 because of the many problems I had. Now running with only 7950.
U can run it but u need 2 seperate sgminers to run with "remove-disabled" : true, and "device" option.

[diks]

I have nvidia
and old amd R7 
X11
amd + ~ 15-18%

Thanks! 

[ocminer]

"/tmp/OCLgJOw7A.cl", line 1110: error: identifier "INPUT_BIG_LOCAL" is
          undefined
              INPUT_BIG_LOCAL;
              ^

"/tmp/OCLgJOw7A.cl", line 1116: error: identifier "INPUT_BIG_LOCAL" is
          undefined
          INPUT_BIG_LOCAL;   
          ^
whats this ?

[aznboy84]

"/tmp/OCLgJOw7A.cl", line 1110: error: identifier "INPUT_BIG_LOCAL" is
          undefined
              INPUT_BIG_LOCAL;
              ^

"/tmp/OCLgJOw7A.cl", line 1116: error: identifier "INPUT_BIG_LOCAL" is
          undefined
          INPUT_BIG_LOCAL;  
          ^
whats this ?

Lol there was a mistake when i switch to the correct branch on github
re-download /kernel/bitblock.cl or just add

#define SPH_HAMSI_SHORT 1

at line 81

[platinum4]

"/tmp/OCLgJOw7A.cl", line 1110: error: identifier "INPUT_BIG_LOCAL" is
          undefined
              INPUT_BIG_LOCAL;
              ^

"/tmp/OCLgJOw7A.cl", line 1116: error: identifier "INPUT_BIG_LOCAL" is
          undefined
          INPUT_BIG_LOCAL;  
          ^
whats this ?

Lol there was a mistake when i switch to the correct branch on github
re-download /kernel/bitblock.cl or just add

#define SPH_HAMSI_SHORT 1

at line 81

I got this error when I tried to drop the newest bitblock.cl from 6/28 into a miner built on 6/25, which ran the original bitblock.cl

It was a very specific case, and I haven't experienced it with the 6/29 edition.

@aznboy84 I'm not sure if it's the LUFFA PARALLEL or not, but i have rigs up now for ~75 minutes after the last SICK/DEAD encountered.

Comparing to other sgminers this version knocked out 3 separate cards I haven't problems with before.  So something is especially aggressive.  The hard-coded hamsi_expand_big 1 shouldn't affect it since I'm on 290X

[ocminer]

What are the optimized starting parameters for 280x under linux ?

I'm getting around 1.000 MH/s and after 5 mins all cards are sick/dead with -I 18, -g 2 and worksize 256 (1100/1400)

[Amph]

how much is the consumption with one 280x, hashing x15?

[aznboy84]

What are the optimized starting parameters for 280x under linux ?

I'm getting around 1.000 MH/s and after 5 mins all cards are sick/dead with -I 18, -g 2 and worksize 256 (1100/1400)

They was running on default SHA256 algo i guess  
BTW here goes the config

Code:

{
  "pools": [
    {
      "name": "MAIA",
      "url": "stratum+tcp://east1.us.stratum.dedicatedpool.com:6002",
      "user": "xxx.1",
      "pass": "x",
      "algorithm": "bitblock"
    }
  ],
  "failover-only": true,
  "algorithm": "bitblock",
  "devices": "all",
  "intensity": "19",
  "worksize": "64",
  "gpu-engine": "1100",
  "gpu-memclock": "1250",
  "gpu-threads": "2",
  "gpu-fan": "50",
  "gpu-powertune": "20",
  "temp-cutoff": "95,95,95",
  "temp-overheat": "85,85,85",
  "temp-target": "75,75,75",
  "gpu-memdiff": "0,0,0",
  "shares": "0",
  "api-mcast-port": "4028",
  "api-port": "4028",
  "expiry": "28",
  "failover-switch-delay": "60",
  "gpu-dyninterval": "7",
  "log": "5",
  "no-pool-disable": true,
  "no-client-reconnect": true,
  "queue": "1",
  "scan-time": "7",
  "tcp-keepalive": "30",
  "temp-hysteresis": "3"
}

[restless]

had you made any changes to x13modold?
On 5850 I'm getting higher hashrates using your miner on x15, than with lasybear miner on x13 :S
With your version now 5850@stock makes almost 1MH/s, up from 850-860, I have to check x11 also

PS: x11 is unchanged, tho for some reason -I parameter, or intensity in conf file is ignored, always starting with I=12 :S

[ibfragalot2]

The problem lies in the v5 version from the beginning it is not supported. So I had to get rid of my 6970 because of the many problems I had. Now running with only 7950.
U can run it but u need 2 seperate sgminers to run with "remove-disabled" : true, and "device" option.

Thanks for your help  

[platinum4]

What are the optimized starting parameters for 280x under linux ?

I'm getting around 1.000 MH/s and after 5 mins all cards are sick/dead with -I 18, -g 2 and worksize 256 (1100/1400)

Right?  Aggressive as all hell huh.

[bullus]

Running it stable for 14hours now, but using the 06252014 version..didn't test this version yet.
4x7950@1035/1500 @2110kh/s