Consolidated Litecoin Mining Guide for 5xxx, 6xxx, and 7xxx GPUs

[chungenhung]

Using Win7 x64. AMD 5850.
Cgminer 2.8.3
cgminer --scrypt -o http://yourpool.com -u username -p password --thread-concurrency 5600 -I 18 -g 1 -w 256

That immediately crashed the cgminer. Window popped up saying "cgminer has stopped working".
What am I doing wrong here?

[chungenhung]

Using Win7 x64. AMD 5850.
Cgminer 2.8.3
cgminer --scrypt -o http://yourpool.com -u username -p password --thread-concurrency 5600 -I 18 -g 1 -w 256

That immediately crashed the cgminer. Window popped up saying "cgminer has stopped working".
What am I doing wrong here?

Now I tried to use Reaper v14 beta 4, x64.
It gives me "Error 6 getting work", which I looked up and it seems that it cannot connect to the pool.

litecoin.conf file

Code:

host http://mine.pool-x.eu
port 80
user xxx.x
pass x


protocol litecoin

worksize 256
vectors 1
aggression 20
threads_per_gpu 1
sharethreads 18
lookup_gap 2
gpu_thread_concurrency 7168

[Cablez]

Now I tried to use Reaper v14 beta 4, x64.
It gives me "Error 6 getting work", which I looked up and it seems that it cannot connect to the pool.

litecoin.conf file

Code:

host http://mine.pool-x.eu
port 80
user xxx.x
pass x


protocol litecoin

worksize 256
vectors 1
aggression 20
threads_per_gpu 1
sharethreads 18
lookup_gap 2
gpu_thread_concurrency 7168

Definitely remove the 'http://' from the host line as its not needed. If you get it to work let me know as I can't get pool-x to connect in reaper.

[chungenhung]

Now I tried to use Reaper v14 beta 4, x64.
It gives me "Error 6 getting work", which I looked up and it seems that it cannot connect to the pool.

litecoin.conf file

Code:

host http://mine.pool-x.eu
port 80
user xxx.x
pass x


protocol litecoin

worksize 256
vectors 1
aggression 20
threads_per_gpu 1
sharethreads 18
lookup_gap 2
gpu_thread_concurrency 7168

Definitely remove the 'http://' from the host line as its not needed. If you get it to work let me know as I can't get pool-x to connect in reaper.

Worked after I removed the "http://"

[chungenhung]

7970: 21000 (sharethreads 2048 and threads_per_gpu 4, the rest the same as the above and using catalyst 12.6 and AMD APP SDK 2.6) (~700 kh/s)

Is that sharethreads 2048 correct??? That is a LOT of threads!
When I put that in, it would just crash my PC, or very unresponsive.

[tacotime]

7970: 21000 (sharethreads 2048 and threads_per_gpu 4, the rest the same as the above and using catalyst 12.6 and AMD APP SDK 2.6) (~700 kh/s)

Is that sharethreads 2048 correct??? That is a LOT of threads!
When I put that in, it would just crash my PC, or very unresponsive.

i changed that because it doesn't really work.  that's for 79xx cards only though.  we're still looking for the right settings for 7970s.

[aqrulesms]

7970: 21000 (sharethreads 2048 and threads_per_gpu 4, the rest the same as the above and using catalyst 12.6 and AMD APP SDK 2.6) (~700 kh/s)

Is that sharethreads 2048 correct??? That is a LOT of threads!
When I put that in, it would just crash my PC, or very unresponsive.

i changed that because it doesn't really work.  that's for 79xx cards only though.  we're still looking for the right settings for 7970s.

What are the best OC settings for the 5850? Currently I am mining at 300 kH/s when it should probably be more, I believe.

[AForceNinja]

i get to this point and reaper.exe crashes. Any ideas?

Code:

Using all devices
         0      tahiti
Compiling kernel...this could take up to 2 minutes.
Binary size: 1417308 bytes
Program built from source.
LTC buuffer size: 640MB
          1          Intel .... i7 cpu.......
Compiling kernel...this could take up to 2 minutes.

UPDATE...I got it to work but now it seems to be using 100% of the CPU along with it and is making the computer impossible to use

im using a 7950 with the newest drivers and the settings layed out in the first post.

Mining at about 500khash/s at 1275mhz core 1600mem

You are mining on your CPU which is why you have 100% usage. To remove that from mining just add 'device 0' to your reaper.conf file right before 'mine litecoin'.

thanks, that solved the problem!!

[jasinlee]

Code:

{
"pools" : [
        {       "url" : "http://ltc.kattare.com:9332",
                "user" : "jasinlee.1",
                "pass" : "1"
        },
        {
                "url" : "http://litecoinpool.org:9332",
                "user" : "jasinlee.1",
                "pass" : "1"
        }
],
"intensity" : "10",
"gpu-engine" : "725-750,725-750,725-750,725-750,725-750,725-725,725-750",
"gpu-fan" : "75",
"gpu-memclock" : "180,180,180,180,180,180,180",
"gpu-vddc" : "1.05",
"temp-cutoff" : "95",
"temp-overheat" : "85",
"temp-target" : "75",
"auto-fan" : true,
"auto-gpu" : true,
"kernel-path" : "/home/miner/cgminer",
"vectors" : "2",
"worksize" : "64"
}

Ok, on ubuntu, this seems to just hang in terminal and never process. What do I have wrong on here? I have 4 5970s running.

[jasinlee]

Ok, redoing it since thats not working. Deleted .conf file and using

Code:

./cgminer --scrypt -o http://ltc.kattare.com:9332 -u jasinlee.1 -p 1 --thread-concurrency 2048 -I 12 -g 1 -w 256

Any ideas on why ubuntu's terminal just shows

[2012-10-18 00:30:53] Started cgminer 2.7.7

And never goes past that for up to 30 minutes. (longest time I waited) ?

[Rjb82]

i've figured a few things out. so here goes.

Lookup gap is tied to some kind of multiplier for memory usage or compression for what's being sent to the gpu. (raising it generally LOWERS performance)
We usually leave this at 2.
i've been playing around with it at 3+ up to 6. and figured out that at each setting it uses a specific amount of memory in relation to thread concurrencies. but we'll just focus on 3 for now.

thread concurrencies
19456 @ 3 = 812.25mb
20480 @ 3 = 855mb
21504 @ 3 = 897.25
(each 1024 is 42.75mb)

the maximum memory usage PER THREAD is limited to 512mb.

at 12288 i would get an error "your scrypt settings come to 537919488 bla blah blah..." which is 513mb. going down just 64 more.
and cgminer starts with no errors. (12224) Running a 7950 @ 1k/1750 i get ~600-610 in reaper and ~380-400 in cgminer with that particular setting.

One nifty thing i found with this also, is that you can raise the intensity to 20 and not get any (atleast in my short tests) hardware errors.

Share generation roughly lines up with the (reported) actual hash rate, i'm comparing it to a pair of 5830's that are doing 275kh/s @ 15.5 each/min, and they show as ~540kh/s on coinotron.

I tested probably 20 different thread and worksize combinations. and while i didn't find any "magic" number... i NEVER got any hardware errors.


I'm no expert, but i really hope someone with a little more knowledge can tweak what i've found to get Cgminer and Reaper on level ground.

Thanks.

[tacotime]

As I stated in the op, you have to bypass the ocl error check to use a large amount of memory as with the 7950. Coblee has stated he will not remove cgminers error checking.

[LiteBit]

Has anyone found that running multiple 5970 cards is more advantageous on one versus the other (cgminer vs reaper)?   I have 3 that I'm trying to run on cgminer now and getting some SICK gpus when I use them together.  Might try them on reaper and see if the same thing happens.

my settings are: cgminer --scrypt -o server:port -u -p --lookup-gap 2 --thread-concurrency 8192 -g 1 -I 18 -w 256

[meebs]

As I stated in the op, you have to bypass the ocl error check to use a large amount of memory as with the 7950. Coblee has stated he will not remove cgminers error checking.

who is the author of bfgminer? That is what I use for ltc mining on my 7770's.. I wonder if he would be interested in doing a little ltc fork without error checking?

[Rjb82]

As I stated in the op, you have to bypass the ocl error check to use a large amount of memory as with the 7950. Coblee has stated he will not remove cgminers error checking.

Right. I've heard that they (coblee and/or Con Kolivas)  won't change it. But what i was saying is that with the right settings (combination of lookup gap/concurrency/other settings) we may not NEED a fix.

i can get ~66% of reaper's performance (600kh/s vs 400kh/s) by messing with it for 90 mins in one night. Surely someone with working knowledge of the algorithm and how it functions in relation to memory usage can figure out "optimal" conditions to get it  "closer" if not even.

I forget the setting, but i get the same error about my scrypt setting being too high... and it still runs. so it's not a HARD set limit. or atleast it looks like there ARE ways to bypass the check. (i think it was lookup-gap 1 with 4096 concurrency. see Ss below)


http://imageshack.us/photo/my-images/844/ss1rwb.jpg/

^ so this shouldnt be possible right? since its over the memory size... 512mb setting and 128mb buffer.

[Nolo]

I'm sure it's something obvious, and I'll say "duh!" as soon as I'm told, but in cgminer, what's the difference between the two hash rates reported?  For example 289.1K/285.6Kh/s 

[superfastkyle]

I believe its average over 5 seconds followed by average since you started cgminer

I'm sure it's something obvious, and I'll say "duh!" as soon as I'm told, but in cgminer, what's the difference between the two hash rates reported?  For example 289.1K/285.6Kh/s 

[philips]

As I stated in the op, you have to bypass the ocl error check to use a large amount of memory as with the 7950. Coblee has stated he will not remove cgminers error checking.

Coblee?

Anyway, from the latest cgminer changelog (2.8.4) :

Scrypt will now not fail when setting high thread concurrency values that still return some ram even if opencl returns an error on that ram allocation.

Maybe you could try again with cgminer ?

[Nolo]

Also this morning I woke up and cgminer said DEAD where the hash rate should be.  I restarted it, and everything went back to normal.  What happened?

[philips]

Also this morning I woke up and cgminer said DEAD where the hash rate should be.  I restarted it, and everything went back to normal.  What happened?

If you are overclocking the card try a lower value, if not, try a lower intensity.