First off, the original CGminer for GPU is dead. All 3.8 and newer builds are made for ASICs only. Luckily, it has been forked, and scrypt-mining on GPU has been kept. Get kalroth's CGminer fork at
http://k-dev.net/cgminer/. Also have a look at this thread at
overclock.net.
Configuring CGminer is quite complicated (it is JSON formatted), but my basic one goes like:
{
"pools" : [
{
"quota" : "20;stratum+tcp://server1:3333",
"user" : "x.x",
"pass" : "x"
},
{
"quota" : "50;stratum+tcp://server2:3333",
"user" : "x.x",
"pass" : "x"
},
{
"url" : "stratum+tcp://server3:3333,
"user" : "x.x",
"pass" : "x"
},
{
"quota" : "10;stratum+tcp://server4:3333",
"user" : "x.x",
"pass" : "x"
},
{
"quota" : "20;stratum+tcp://server5:3333",
"user" : "x.x",
"pass" : "x"
}
],
"failover-only" : true,
"load-balance" : true,
"scrypt" : true,
"kernel" : "scrypt",
"intensity" : "15,13",
"thread-concurrency" : "4476,3198",
"gpu-threads" : "1",
"gpu-platform" : "1"
}
This config turns on the normal scrypt module, as you can see.
Also it enables failover and load balancing. As you can see, server 1,2,4,5 shares their quota amd server 3 is only a fail-over, but it will still failover to any alive pool as long as a server with quote demand is dead.
20+50+10+20 = 100 %.
Notice the "quota" parameter replacing the usual "url" parameter, and that is add a comma separator after the quota specified before the url!
Intensity: Is just how much steam the cards can handle. If you see in the CGminer console that you are getting HW errors, lower the intensity. 15 is a bit high, but I get no errors. 13 is more common, but still gives me a HW error of 1 per day or so. That is not good, so I should lower intensity to 12 and see how it goes.
Note the comma-separation on the parameters that supports it. Generally PCIe slot 0 is the first, PCIe slot 2 is the second, etc!
Thread concurrency: Is calculated by taking number of GPU shaders and multiplying by 4 (for the 6xxx and 7xxx series, other models has other optimal calculations.) and remove 4.
GPU Platform: Normally, the platform available on PCIe slot 0 becomes gpu-platform 0. That means if you have an Nvidia card in slot 0 like me, and 2 AMD cards on slot 1 and 2, it becomes platform 1. So platform 1, card 0 and 1 is my active cards in CG miner. cudaminer would then need the opposite config, pretty much.
Starting up CGminer from a batch file:
@echo on
ECHO "cgminer miner startup"
ECHO.
setx GPU_MAX_ALLOC_PERCENT 100
setx GPU_USE_SYNC_OBJECTS 1
cgminer.exe --config=cgminer.conf
setx works on Windows, as some people dispute. Look at the 'succeeded' return-message in the console.
