3 x 7970s - CGminer tweak advice

[whitetoo]

Hi guys,

Wondered if anyone would be able to provide any advice to tweak a test setup?
I am averaging 602 m/hashes per card - so 1.8G/hashes for the rig.

This is currently my config for cgminer:

"intensity" : "9,9,9",
"vectors" : "1,1,1",
"worksize" : "256,256,256",
"kernel" : "poclbm,poclbm,poclbm",
"lookup-gap" : "0,0,0",
"thread-concurrency" : "0,0,0",
"shaders" : "0,0,0",
"gpu-engine" : "0-0,0-0,0-0",
"gpu-fan" : "0-85,0-85,0-85",
"gpu-memclock" : "0,0,0",
"gpu-memdiff" : "0,0,0",
"gpu-powertune" : "0,0,0",
"gpu-vddc" : "0.000,0.000,0.000",
"temp-cutoff" : "95,95,95",
"temp-overheat" : "85,85,85",
"temp-target" : "75,75,75",
"api-port" : "4028",
"expiry" : "120",
"gpu-dyninterval" : "7",
"gpu-platform" : "1",
"gpu-threads" : "4",
"hotplug" : "5",
"log" : "5",
"no-pool-disable" : true,
"queue" : "1",
"scan-time" : "60",
"temp-hysteresis" : "3",
"shares" : "0",
"kernel-path" : "/usr/local/bin"
}

Thank you rock gurus in advance!

[BigBitz]

From what I have seen/read you're getting a fairly good rate. I Don't think you have much room for improvement unless you consider overclocking your cards.

[xwebnetwork]

Yeah, that's not bad. No errors? What's your work utility (WU) ratio? And are you using GUIMiner?

I have 4x7970's (XFX Black) and can't get them to go higher than 560. I have the mem set to 1775, engine to 1105, 22400 thread count, intensity 13, autofan, target temp 75. If I bump up the intensoty, the KHs drops and error rate goes up. Right now its 0. Did you have any issues like this getting to 600?

[MashRinx]

I have just under quite a few 7970s and all have been stable at 1125 for months.  If you do chose to give this a shot, cooling may be an issue depending on how your machines is set up, cooling, etc.  Good luck

[Jokah]

Here's mine for reference:

Core speed - 1125mhz
Mem speed - 685mhz
Voltage - 1112v (stock)
Temps - ~50° (water cooled)
Hash speed (avg) - 650
Flags - None
Miner - Default GUIminer (poclbm)


I find the biggest factor to optimum hashing power is what driver you use. I am pretty sure i am on 12.8 as anything newer doesn't seem to be as good.

I also used to have 4x 5870's mining in a dedicated rig I set up. I had a low end AMD sempron (by recommendation) but found this was bottle necking my hashing power. I upgraded the CPU and all was well. I thought this would be worth noting as I don't know what the rest of your rig is.

[alexeft]

A way to fend off any cooling issues is to use a high quality thermal paste. Here's my story:

I got an XFX black DD at 1000 Mhz. I was getting high temps that fell substantially after using Noctua's nt-h1 thermal compound.
Before using the compound, the card would reach 97 degrees C with the fans at full blast.
Now it's 80 degrees C with fans at 2700 rpms.

Of course, the card doesn't have a warranty anymore, but, WOW, what a difference.  

[whitetoo]

wow thanks for all your replies guys.

Amazingly the GPUs are running between 50-60C at 65% fan speed set with MSI Afterburner.
Core clock 1010mhz
Mem clock 5050mhz
No specific cooling other than fans setup in a cold garage.

What is the advantage of downgrading the mem speed? Energy saving?
How can I tell if I've experienced errors with CGMiner (just switched from GUIMiner) sorry for the newb question!

I'm also running catalyst 13.1 so will downgrade and report back with hash changes!

Thanks for the tip about the thermal paste.
I've applied to CPUs before but never GPUs so a walkthrough on this would be intriguing if you have time Alex.

[sagiko]

I wish I am in a colder place like you guys. I merely managed to keep my 7970 below 85C in a tropical country  where we have temperature around 32C throughout the year.

[xwebnetwork]

I downgraded to 12.10 and got an extra 30 Khs per card. Gonna try 12.8 and see what we can do there!

Follow this guide for sure:

https://bitcointalk.org/index.php?PHPSESSID=gs19k9d0k3lhoc1an3m8oovv45&topic=28402.msg1635964#msg1635964

[mdude77]

You don't state if you're going for LTC or BTC.

If LTC, I suggest bumping the engine to 1125 and changing the memory to 900.  Been doing that with my set of 7970s for some time now and I average 650 a piece.

Also, there's no need to specify things like vectors and cores in the config file.  cgminer is smart enough to know what's best.

Here's my config file:

Code:

"intensity" : "9",
"gpu-engine" : "0-1125",
"gpu-fan" : "0-85",
"temp-cutoff" : "90",
"temp-overheat" : "85",
"temp-target" : "75",
"auto-gpu" : true,
"auto-fan" : true,
"expiry" : "120",
"gpu-dyninterval" : "7",
"gpu-platform" : "0",
"gpu-threads" : "2",
"log" : "5",
"queue" : "1",
"retry-pause" : "5",
"scan-time" : "60",
"temp-hysteresis" : "3",
"shares" : "0",

Mine are in the basement with a box fan blowing directly on them, the fans usually don't go above 60%.

M

[8kmode]

try using the diabo kernel?
Using diablo over poclbm gives me an extra 30 - 40 mh/s on a single 7970

[whitetoo]

Thanks for you continued advice guys!
I've ben tweaking config setting so far but not tried diablo over kernel yet.

MSI Afterburner

Core Volt: 1112
Mem Volt: 1600

Core Clock: 1118
Mem Clock 1000

Fan Speed: 65%

650 mhashes/s per card at the moment at 55-60C temp.
Pulling 790watts (measured at the wall).

I've read a few threads now concerning unvolting the cards but each recommended setting i've tried crashes the display driver (13.1). I tried the 12.8 for a while but felt the 13.1 was more stable other than this issue.

Additionally if I set the mem clock to 900mhz this pulls 20watts in addition which i thought was weird.
Does anyone have any further advice for the above?

[GuiltySpark343]

Have two 7970s in a Win7 x64 rig with Catalyst 13.1. Using BFGMiner with -I 9 --gpu-engine 1075 --gpu-memclock 1200

If I try a higher core engine clock speed, I crash. If I use a lower memory clock speed, I also crash. Voltage is locked, unfortunately, by the manufacturer (XFX) and so I cannot under or over-volt my cards.

With above settings, getting avg 620 MHash/sec.

[MisterMelancholy]

I don't think this is optimized, but I'm getting about an average of 715Mh/s on for my 2 7970s @ 1180/110 with this config:

Code:

"intensity" : "8,8",
"vectors" : "1,1",
"worksize" : "64,64",
"kernel" : "poclbm,poclbm",
"api-port" : "4028",
"expiry" : "120",
"gpu-dyninterval" : "8",
"gpu-platform" : "0",
"gpu-threads" : "2",
"log" : "5",
"no-pool-disable" : true,
"no-submit-stale" : true,
"queue" : "2",
"scan-time" : "20",
"shares" : "0",
"kernel-path" : "/usr/local/bin"

[WuNinjaTmp]

Thought I should do my due dilligence and post on this...
I *had* 3 XFX 7970 GHz Edition 1050/1500 Double Disappointment cards (I've RMA'd them back to the depths of hell whence they came, and got 4 x 7950's instead...).
I couldn't get them to mine LTC above 550KH/s. Ever. I tried everything.
For BTC, however, after lurking in these forums for a while, and tweaking/overclocking, I found the following to give me 675-700MH/s per card.  Here was the .bat file I kept in my cgminer folder, and then shortcut to desktop/startup:

color 02
setx GPU_MAX_ALLOC_PERCENT 100
setx GPU_USE_SYNC_OBJECTS 1
cgminer.exe -o stratum+tcp://MINING_POOL_URL:3333 -u WORKERNAME -p WORKERPW --shaders 2048 --intensity 13 --thread-concurrency 8192 --gpu-engine 1100 --gpu-memclock 600 --worksize 256 -g 1

-g 1 is for 1 card... change this to however many cards you have.

[-ck]

-g 1 is for 1 card... change this to however many cards you have.

No it's not. At least you're trying to help...

[crazyates]

I don't think this is optimized, but I'm getting about an average of 715Mh/s on for my 2 7970s @ 1180/110 with this config:

Code:

"vectors" : "1,1",
"worksize" : "64,64",
"kernel" : "poclbm,poclbm",

IIRC, those are the CGMiner defaults for a 7970 anyways, so I doubt you would get any better than that.

[pccs]

I just spent another day trying to get more out of my 7970 XFX DD.

The best I can do is:

LTC: 525kH/s

BTC: 660MH/s


I don't know how some people get in the 700s, I can't even get close.

[Kalen]

I have some XFX7970's, using 13.1 drivers. Had to clock it up to 1160+ to hit 700mh/s. Mine has mostly been stable around 1140mhz.

Could never underclock memory below 1300 either.

[pccs]

Yeah, I lock up too when lowering mem clock.

Got to 540K on LTC with these settings.

"intensity" : "13",
"vectors" : "1",
"worksize" : "256",
"kernel" : "scrypt",
"lookup-gap" : "0",
"thread-concurrency" : "22272",
"shaders" : "2048",
"gpu-engine" : "1100",
"gpu-fan" : "0-85",
"gpu-memclock" : "1700",
"gpu-memdiff" : "0",
"gpu-powertune" : "0",
"gpu-vddc" : "0.000",
"temp-cutoff" : "75",
"temp-overheat" : "70",
"temp-target" : "65",
"api-port" : "4028",
"auto-fan" : true,
"expiry" : "120",
"gpu-dyninterval" : "7",
"gpu-platform" : "0",
"gpu-threads" : "2",
"log" : "5",
"no-pool-disable" : true,
"queue" : "1",
"scan-time" : "60",
"scrypt" : true,
"temp-hysteresis" : "3",
"shares" : "0",
"kernel-path" : "/usr/local/bin"

640Mh/s with these settings:

"intensity" : "9",
"vectors" : "1",
"worksize" : "256",
"kernel" : "poclbm",
"lookup-gap" : "0",
"thread-concurrency" : "22272",
"shaders" : "2048",
"gpu-engine" : "1100",
"gpu-fan" : "0-85",
"gpu-memclock" : "1700",
"gpu-memdiff" : "0",
"gpu-powertune" : "0",
"gpu-vddc" : "0.000",
"temp-cutoff" : "80",
"temp-overheat" : "75",
"temp-target" : "65",
"api-port" : "4028",
"auto-fan" : true,
"expiry" : "120",
"gpu-dyninterval" : "8",
"gpu-platform" : "0",
"gpu-threads" : "2",
"log" : "5",
"no-pool-disable" : true,
"queue" : "1",
"scan-time" : "60",
"temp-hysteresis" : "3",
"shares" : "0",
"kernel-path" : "/usr/local/bin"


I had it up to 660M steady, but forgot to save the settings.