Strange problem with mining rig, 3x7950

[gerjan]

Hi,

I got a strange problem which I can't figure out.

A couple of weeks ago I built a Litecoin miner.

It's an Asrock 970 Extreme4 with an AMD Sempron 145 en Xubuntu.
It was built according the manual of Cryptobadger....

I ordered 3x Club 3D HD7950 (cgax-795713) but only 2 were delivered and the third stayed in back order.
I thought Oh well, I'll mine with 2 card at first en add the third one later.

The two cards are mining at 610 kh/s and the temperature is about 78/84 degrees Celsius. (This is good right?)

Three days ago I received the third card. I added it to the rig and ran to command to activate it.
CGminer sees the card and I start mining.

Now...the strange thing is that this card's temperature is going to 102 degrees Celsius. It starts mining at 680 kh/s but drops in 1 minute to 540 kh/sec and stays that way.
The fans of this card have a much higher RPM, the other two cards are about 2900 rpm, the new one, 3300 rpm.

I built a casing of aluminum frames, so nothing is enclosed.
I tried different positions of the card, first, second en third place, nothing helps.
The rig is in the basement with an open window. Now this particular card receives the first cold air coming in through the window.

Another strange thing is, that I set the engine and memory clock in the settings en this card seems not to takeover those settings. The other two do.

The last card is identical to the first two, same part-number, etc....

Current CGminer settings:
./cgminer --no-submit-stale --scrypt -I 19,19,19 --thread-concurrency 24000,24000,24000  --gpu-engine 1100,1100,1100 --gpu-memclock 1475,1475,1475 --worksize 256 --gpu-vddc 1.150  --gpu-powertune 20,20,20 -g 1 -o stratum+tcp://usa2.wemineltc.com:3334 -u gerjan.1 -p x --failover-only -o stratum+tcp://coinotron.com:3334 -u gerjan.1 -p x

http://www.netice.nl/cgminer.JPG

http://www.netice.nl/cgminer2.JPG

http://www.netice.nl/miner1.jpg
http://www.netice.nl/miner2.jpg
http://www.netice.nl/miner3.jpg

[Morning-Star]

102 is VERY VERY hot!    I though the card would shut down before reaching that temp.

Take a look with gpu-z and check out all your voltages and current draws, swap cards around to make sure it's the card and not MB location or power supply rail giving the card a hard time.

[zmhaha]

When I installed my 2nd 290x, the temps rised but certainly not that much, 5c at most, given all other settings stays the same.

If 2 cards are fine, then you can definitely say the mobo and the 2 old cards not having any problems.

So next, try 2 cards only but with the new card being one of them.

If you still see that problem you described, then the card itself must be the cause. (you said it is backordered, so it could be a bad batch)

If problem is gone, then it should be mobo having problems with handling 3 cards but if this is the case, the problem should not be restricted only to the new card and could happen to any of the 3 cards.

Also, make sure disable the Xfire settings manually. I don't know much about 7950s but with 290x or 290, catalyst has a tendency to reenable Xfire when you swap cards..

FYI:
A 850w PSU is low for 3x 7950s together with all other stuff.
3x 7950 will draw nearly 800w already while mining. even you unplugged nearly everything else, i would say the system still needs more than 50w to run...
So the PSU is running nearly at 100% load all the time... not a good thing... lower it down to 70-80% at least if you wish to let the PSU last.. So buy another 500w but silver+ at least and let it power one card along.

I am running 5x 290x. I put 3x on Corsair AX1200i, 2x+everything else on Corsair HX1000. Total power drawing from the wall while mining is about 1950w measured by kill a watt and this is measured after 2 hours of mining,  which means it is very close to the steady state power consumptions. (other stuff plugged in: Keyboard, mouse, 1x ssd, 2x HDD, 24'' monitor, 5x 120mm fan, 1 DDC water pump.  )

[gerjan]

OK, thnx for the input.

Last evening I made some progress. I was reading about undervoltage and I started testing. I downloaded VBE7 and this gave me some details about the cards when I opened the stock roms.

The first two cards have this memory: Hynix H5GQ2H24MFR, Elpida EDW2032BBG
The third card has this memory: Hynix H5GQ2H24AFR, Samsung K4G20325FD.

So conclusion, it is another batch.

So I fiddled around with some other voltages. The first two cards I cannot run lower than 1,100 volt. CGminer crashes after minutes when I do. I get the best results at Engine 1100 en Mem 1500: 622 kh/s. They stay at 78 degrees C.

The third card I can run at 1.081 volt and even 1.056 volt. At Engine 1100 and Mem 1500 I get 670 kh/s. Temperature is stable at 83 degrees now, in the livingroom and not the cold basement. BUT, currently it is running at 1.056 volt and started last night at 23:00 (11:00 PM), but according to wemineltc stats, it stopped at 01:00 AM. Also CGminer was not responding this morning.

I'll try 1.081 volt today and see what it does.

I find it very strange that these cards with different memory types go by the same name and model number and have different results in mining (and perhaps in gaming as-well)

I'll see if I can get another PSU this week and I'll get a WATT-meter to see how much power it drains.

EDIT: I flashed the third card to 1.081 volt. Temperature is at 86/88 degrees C. now. Fan is rotating at 2800/3200 rpm. Hashrate 670 kh/s but sometimes drops to 600, it does climb up to 670 kh/s again. Now running 1 hour....

[zmhaha]

You will be surprised how much difference it makes just because of the RAMs.
Scrypt mining is very RAM intensive. And we all know different chips can have very different qualities even though they have the same specs.
It is just like when you OC a CPU. Some 4770k will get to 5.0Ghz, while most will only reach 4.4-4.5Ghz, given the same cooling setup.
It is like a silicon lottery. lol


You said the Kh/s drops sometimes temporarily, you are talking about the 5s average instead of the all time average, correct?
If that's the case, I can't say there must be a problem since mine does the same but just not that much.. 20-30kh/s at most on the 5s average

Oh, btw, in my first post, I said it wrong. 3x 7950 will draw at about 900w instead of 800w...
So adding another 150w for the rest of the stuff. You are looking at 1000w+... Doing this too much on your 850w will trash it..
I would definitely get a 1200w gold at least after taking the high load capacitor aging into account.

[ajareselde]

Look, 3rd card is overheating, there is definetly something wrong with cooler there, reason it drops hash is coz card is throtling, and its doing that when u pass about 90 C i think or so..

Run the card through some benchmarks, if its overheating and thrthling, RMA it. If u dont have warranty, just dissasemble the card, re-apply thermal paste and make sure cooler head fits nice and tight to the gpu.

had a milion different problems throughout the time with cards, and im 99.9 % sure this will fix ur problems.

cheers

[gerjan]

@zmhaha, I think I have another PSU somewhere. I'll look it up an try to connect it.
On the cryptobadger site I read that 1 7950 uses 200 watt, so that's 600 for 3. Then I got 2,5 watt left over. How do you feel about that?
It is the 5sec value that drops indeed.

Currently I have half-working situation:

CGminer crashes once-a-day. In screen -r mode, it doesn't respond or refreshes. Rebooting the rig is all I can do.

I think the first card is responsible for this.

Overall it is working better then before:

http://www.netice.nl/cgminer3.JPG
This is 30 minutes after rebooting.

I don't have a screenshot, but when it doesn't respond, the first GPU gives me a temperature off 65 degrees. I think it stopped hashing and CGminer responded just long enough to report its temperature. The other two GPU's are still at 77 and 89 degrees.

If it is the first card, I can lower the voltage of the third card to 1.056. Because I first thought this was the card which crashes the CGminer.
I will lower this one today and give the first GPU a bit more voltage. The temperature will go down to 84 degrees when I do.

http://www.netice.nl/cgminer4.JPG

You can see that the third GPU is listening to it's memory settings now. This is since I lowered the voltage.

EDIT: Currently I draw 835 watt from the wall. I bought a meter. This is with lowered voltage on the cards. With original BIOS I use 100 watt more!!!

[zmhaha]

200w for a 7950 is problly for gaming.
for mining, about 300w.
You do not want to push you PSU all the time.
Keep it under 80% load if you want to run it 24x7.
Keep in mind that the rated wattage will starts decreasing due to capacitor aging, and this is much quicker if you stress it a lot

[gerjan]

Ok, I got a spare PSU. Dow you know what is the best way to connect a second PSU.

It has to turn on at the same time as the other I think?

I can short circuit pin 4 and 5 on the ATX-connector to turn it on.

[zmhaha]

I guess that is the easiest way.
You have to jump start the PSU when it is not connected to a mobo.
So yea, on the 24pin mobo connector, stick one end of a paper clip into the pin with green wire first and then stick another end to any black wire pin. You have to keep the clip on all the time otherwise it will gets a "power off" signal.
Also make sure you jump the stand along PSU first before you start your PC.