4x R9 280X Dual-X Causing headaches..

[Rueben]

Hi,

I recently started mining with 1 card and a few days ago decided to bring up my game to a seperate mining rig.
The rig that the questions are posed about is the following:

Hardware:
1x Gigabyte GAZ87X-OC (link)
1x Intel G3220 (link)
4x Sapphire R9 280X Dual-X (link)
1x Corsair RM1000 (link)
1x Kingston USB for booting

Software:
BAMT 1.2

Other:
The cards has been flashed to a lower voltage (currently 1.12)
The cards are set at 1070 Mhz core clock.
Functional cards are delivering 0.74-0.75 Mh/s according to BAMT.

The cgminer config (broadly outlined):
-g 2 -w 256 -I 13 --thread-concurrency 8192 --lookup-gap 2 --gpu-engine 1070 --gpu-memclock 1500 --gpu-powertune 20

The problem:
One of the cards, ran stock at 1.225v, and did not function correctly at 1.09 or 1.1, I got it working at 1.12 where it had a slightly lower peformance than the other cards stock.
If I let it rip with its lower, but alas, currently acceptable 0.69 Mh/s it drops to 0.30Mh/s after 0.5-1hours, which causes temp to drops accordingly. During the drop the card load stays the same, and the clocks are not adjusted.

The dropping of hash power has also occured on one of the other cards, leaving it trailing after a few hours, dropping down to hashrates as low as .44Mh/s

My first thought was that the Corsair 1000w was not delivering sufficient power, but I abandoned this theory after reproducing the problems with 3 cards in the rig.

The problem above can be adjusted by rebooting the system, which causes the card to run stable for a while, until it eventually, drops back down. The second problem causing card described above has since not done a Mh/s drop and remains stable (for now) at 0.74 Mh/s

I am stupified as to what could cause this drop, it is almost as if the card just shuts down 50% of its calculating power and remains in some handicapped mode until i reboot it. I currently have it mounted in my main computer so I can make adjustments without shutting down the rig currently running the other 3 cards.

Any ideas?

[Rueben]

Update:

One of the cards has now dropped from 0.75 Mh/s to 0.70 Mh/s in about 10 minutes, the card was running at 72-75 degrees stable before the hash-rate drop started occuring, it has since been dropping its fansto stay at 70+ degrees.

Clocks, and load remains the same, and the rig is pulling 850W from the wall socket.

Update 2:

Now down to 0.68Mh/s and slowly dropping more... Clocks and load remain the same..

Update 3
Another 10 minutes has gone, in Gpumon the card shows a value of 925.5K/652.1Kh/s and keeps dropping, the 'K' value of the culprit card is higher than that of the other cards, where it usually matches the Kh/s rate

Also tried resetting it resulting in an error in cgminer:

[14:33] GPU1: Idle for more than 60 seconds, declaring SICK!
[14:33] GPU still Showing activity suggesting a hard hang.
[14:33] Will not attempt to auto-restart it

Tried disabling miner, Gpumon/web-dashboard shows 99% load even when mining is turned off, now rebooting the machine.

Now trying the affected card at 1060Mhz to see if it consumes to much power to keep the earlier 1070Mhz clock, Hashrates so far seems almost unaffected, but the invalid rate has gone up, its only been running for 5 mins or so, but 2 invalids already..

[MadMatt]

Are you running powered risers? I have 5 Dual-x's and I'm getting around ~730 kh/s. You're gonna have to tweak the settings around to what the cards like. I'd try putting the voltages around 1.1, core clock @ 1040, and I found out my Dual-x's like thread concurrency 11200. After you get a stable baseline for the cards, that's when you can start tweaking for better performance. 10 more hashrate is going to produce more if you're getting rejects and errors

[Rueben]

Are you running powered risers? I have 5 Dual-x's and I'm getting around ~730 kh/s. You're gonna have to tweak the settings around to what the cards like. I'd try putting the voltages around 1.1, core clock @ 1040, and I found out my Dual-x's like thread concurrency 11200. After you get a stable baseline for the cards, that's when you can start tweaking for better performance. 10 more hashrate is going to produce more if you're getting rejects and errors

I'm running without risers straight off the board for now.

As for the problems, by lowering the culprit card clock to 1060 I've now been getting a consistent hashrate of 740 Kh/s, so next step is probably to drop the core clock on the other cards to see if that has any effect on their Kh/s, if not and it stays stable I'll probably run them all on 1060 as it seems to work.
If 1060 works, I'll probably try 1050 too, and eventually maybe even 1040 as you suggested.

As for the invalids I always get them when the rig has been started, and they eventually stop coming around, now almost 3 hours after my last update of the earlier post the card that had 2/6 invalids has 3/60.

Having to tweak the cards individually is kind of a hassle, one would expect the manufacturer to slim up their margin of error a bit when they ship cards at 1.225 - 1.134v, the power consumption difference between the two is enormous.

[MadMatt]

Download Trixx, you can change core clocks and voltages while they're hashing to get real time results. Once your change them in Trixx though, they all drop down in speed by like 10-15 hash and come right back up, so wait a second til' you gauge a reading.

They should be more stable at lower core clocks, that's why you start lower and work your way up, not the other way around

[Rueben]

Download Trixx, you can change core clocks and voltages while they're hashing to get real time results. Once your change them in Trixx though, they all drop down in speed by like 10-15 hash and come right back up, so wait a second til' you gauge a reading.

They should be more stable at lower core clocks, that's why you start lower and work your way up, not the other way around

I'm running them from a headless linux, and afaik theres no trixx there, I can however easily adjust the clocks via cgminer, accessing the rig from my main computer, but thanks for the heads up on the 10-15 drop though.

I felt that I kinda started from the wrong way, but I had one card running in another computer, prior to buying 3 more, that I had been optimizing for 750KH/s earlier, and thats the .conf i started out with, in retrospect I shouldn't have because it caused a lot of trial/error with the cards that had higher stock voltage.

[MadMatt]

I would also only keep 3 running on the 1000w while you're trying to set it up if that's the only PSU you're using. Seems like very little for 4 and everything else that needs power, I could be wrong though.

[Rueben]

I would also only keep 3 running on the 1000w while you're trying to set it up if that's the only PSU you're using. Seems like very little for 4 and everything else that needs power, I could be wrong though.

Yes, I anticipated that it could cause some problems so I have another one ready to be plugged in if it is too much strain for the 1000W PSU, should have gone with the 1200 but it was not in stock friday night and I was eager to get started!

[MadMatt]

You could be able to get them running if you undervolt them enough, but seems like this is causing problems. I would suggest using something like Add2Psu if you get another PSU or there's guides on how to get 2 PSU's running together w/ a paperclip

[oroqen]

4 280x cards even under volted is going to be over 1000KW/h, 4x with 1.087v and1040core is taking 1050 watts off the wall for me, with stock 1.2 volts its roughly another 50 watts per card.

[cisahasa]

bios voltage fail
sapphire bios voltage does NOT match to gpu-z readings!!!
check true voltage with gpu-z, should be around 1.025v.
when hitting 1,025v on the bios, gpu-z shows 0,920V; CRASH!

no harm done yet, but your psu is way under too small, it can take with out mobo maybe 4 cards if you lucky