Topic: CGMiner only mines 1 card scrypt, 4 cards otherwise. Why? (Read 1168 times)

OK, tried a few more things and got mixed but not results that I can make use of but thought I'd share in case it helps anyone else or someone has some ideas.

I set
with 2 I only got 3 GPUs #ing.

Now with 4 GPUs running I started to fiddle with the clocks.

With all memory clocks on 1725 and all gpu on 1030 I get 4 cards running.
With all memory clocks on 1750 and all gpu on 1045 I get 3 cards running.

Now I try to set the mem clocks individually (all gpu clocks back to 1030):
1750,1725,1725,1725 = 4 cards OK
1750,1750,1725,1725 = 3 cards OK
1750,1725,1750,1725 = 2 cards OK = now this is strange as I'd expected it to be 3 cards at worst.
1750,1725,1725,1750 = 4 cards OK.

At this point # rate is dire: 416, 276, 157, 283. Useless.

I then start to fiddle with the GPU clocks but this just leads to cards doing nothing again (even though the clock is in range for the card).
1045,1030,1030,1045 = 3 cards
1045,1030,1030,1030 and removed --shaders but still only 3 cards

Leaving --shaders out I try all cards on 1725 again (all back to 1030 gpu) and I only get 2 cards going so I put shaders back in.

I go back to 1750,1725,1725,1725 (with shaders back in) and I get a # of 417,259,373,311 which is a slight improvement but still useless.

I now try tweaking the GPU based on the lower limit with the 0.57 - 0.6 ratio and set the gpu clocks 1030,1015,1015,1015
which gives 417,276,408,281 #.

I'm using Ubuntu 12.04 LTS with CGminer 3.2.2 and one thing that makes this process slower than is should be is that when I hit Q to stop CGminer it does stop but the process does not end. I can kill it but if I then try to use CGMiner again it doesn't work so every time I try some new parameter I'm having to reboot. Perhaps this is a clue?

Arrrgh! It's a hot day & even hotter than the devils ring after a napalm curry where I have the machine (even with aircon) so I set the machine back to SHA-256 & give it a rest while I read up for more ideas... but with all the ASICs coming on-line I need to turn this machine around for scrypt.

No idea, it shouldn't be used at all

Example of something close to my config:

Cgminer.exe --scrypt -o stratum+tcp://POOL_ADDRESS:PORT -u username -p password --thread-concurrency 22400 --gpu-engine 1100 --gpu-memclock 1500 -I 19 -g 1 -w 256






Don't forget to add the --scrypt to the beginning like in my example! Let me know how my suggestions work out for you!

Why do people continue to use the 'shaders' flag/variable when the 'thread-concurrency' flag/variable supersedes it?

Sorry I butchered that last post. Phone keyboard...

Keep -w (worksize) to 256 for your cards. That will work best for 7 series. Have you checked some of the example configs on github? try taking any extra parameters out except for thread concurrency, worksize, and intensity. Also remember scrypt is very memory intensivr. Bandwidth is not an issur since the hashing os all calculational on the cards. Its good to have some thing like a .70 ratio of engine to memory speed. For example on my 7950s im getting about 650 kh/s per card with three of them. Im using an engine speed of 1100 and memory speed of 1500.

You have to add --gpu-engine (speed you want) and --gpu-memclock (speed you want) as a parameter. Give it a try, i bet it will work

Here's a couple of tips based on my observations and experience. Your mileage may vary. Limit one offer per customer. Batteries not included. Always ride safely and wear a helmet. Don't forget your towel.
 
- Dial in each card separately
- Use a *.conf file instead of the command line, it makes life a lot easier.
- Thread concurrency may differ for the same cards plugged directly into the MOBO vs plugged in with 1x-16x or 1x-1x risers, likewise for powered/unpowered risers.

Go forth, tweak, and you're welcome.

OK, I changed


To:
(changed lookup-gap to 0 and -g to 1).

This worked "OK" on 3 cards but nothing on card 2. I didn't bother to check the # rate at this point.

To:
(Took out -w, changed lookup-gap to 0 and -g to 1).

Now all 4 cards work, "YAY"  but the # rate is so bad (200 to 400K per card) where I'd expect 600-700+

I do wonder if I'm coming up against a bus/memory bandwidth issue or something like that.

Anyone else nailed this?

Try doing your settings times 4, separated by a comma. For example, -I 13,13,13,13 --shaders 2048,2048,2048,2048 etc.....

Always start low and work your way up

Indeed, great movie!

The problem is not one of stability, it runs fine but just on one of the cards while the other sit there doing jack all.

I will try the -g 1 ASAP.

Thanks for the link to the litecoin stuff. I did do a search & did not find the problem I'm seeing (but that thread is massive so I guess if it's anywhere it'll be in there. Thanks.

And maybe thread concurrency 22400

Indeed, I love it.

On a serious note to the OP try -g 1 DO IT NAO

Um, not using the 'search' function for the forum? https://bitcointalksearch.org/topic/consolidated-litecoin-mining-guide-for-5xxx-6xxx-and-7xxx-gpus-117221

Ahhh thread jack. 

What a great movie.

"I told that fucking kraut down at the league office a thousand times I don't roll on shabbos"

"They already posted it walter"

"WELL THEY CAN FUCKIN' UNPOST IT!!!!!"

Fuck it man.  Let's go bowling.

Dude, your picture dude.

Right! 

Run everything at default and get them all working first.  Then play with thread concurrency, intensity, and the other stuff.

My suggestion would be to stop trying to run 4 threads per GPU before you get everything stable