Odd 5970 Problem

[Inaba]

I seem to be having an odd problem with a 5970, has anyone experienced this?

Running Linux, same build as the rest of my miners, which are not having this problem (so I know it's not the software, or if it is, I don't know why/how it's different than all my other mining boxes):

Upon first booting up and starting the miner (doesn't matter if it's poclbm or phoenix) on GPU0, things are working fine and putting out ~391 Mh/s.  After awhile, the hash rate drops and goes back up to the normal rate... it continues to do this until a reboot. 

It's like it's throttling and/or overheating, but temps are more than fine, and in the case of that core is less than GPU1.  GPU0 is humming along at 74C, GPU1 is running 75 - 76C, so it's definitely not heat related.

Does anyone have any ideas why my hash rate is dipping?  Has anyone seen this before, where one GPU on a 5970 dips and the other is solid?

/proc/interrupts shows a roughly equal number of interrupts on both cards.

I'm pretty stumped and about to reload.  There is one oddity with this box and that is the HD access seems to be slower than it should be, given it's on a SATA drive in this instance, whereas my other miners are on IDE back planes... but I can't see how a wonky HD might affect the internal workings of the GPU.

Anyway, anyone got an idea?

[1713573895]

1713573895

[Jaime Frontero]

I seem to be having an odd problem with a 5970, has anyone experienced this?

Running Linux, same build as the rest of my miners, which are not having this problem (so I know it's not the software, or if it is, I don't know why/how it's different than all my other mining boxes):

Upon first booting up and starting the miner (doesn't matter if it's poclbm or phoenix) on GPU0, things are working fine and putting out ~391 Mh/s.  After awhile, the hash rate drops and goes back up to the normal rate... it continues to do this until a reboot. 

It's like it's throttling and/or overheating, but temps are more than fine, and in the case of that core is less than GPU1.  GPU0 is humming along at 74C, GPU1 is running 75 - 76C, so it's definitely not heat related.

Does anyone have any ideas why my hash rate is dipping?  Has anyone seen this before, where one GPU on a 5970 dips and the other is solid?

/proc/interrupts shows a roughly equal number of interrupts on both cards.

I'm pretty stumped and about to reload.  There is one oddity with this box and that is the HD access seems to be slower than it should be, given it's on a SATA drive in this instance, whereas my other miners are on IDE back planes... but I can't see how a wonky HD might affect the internal workings of the GPU.

Anyway, anyone got an idea?

SATA and one PCIe slot?

seems more likely to be a wonky motherboard than hard drive or card.

if it's the same build as other miners, i'd try swapping the weird 5970 into a different miner.  if there's no problem, it's almost surely the motherboard.  at the least, try swapping slots on the two GPUs in the questionable box.

[error]

Could it be you're running graphical stuff on GPU0 that the card has to take a moment to display? This will reduce your hashrate, of course.

[charliesheen]

My hash rate goes between 1.7-2.1 gh/s over 3 5970 on ubuntu. http://www.youtube.com/watch?v=j86pORrusUo

[Inaba]

Well, I suppose it could be a wonky MB, but it seems like I'd be getting errors or even crashing if the MB was hosed, but you never know.  I'll try dropping it into another box tomorrow and see what happens if I get a chance tomorrow.

As far as running graphical stuff on GPU0, no... at least not any more than I am on other mining boxes, that is to say a basic X desktop with nothing going on and the screen saver active (I do all my work in SSH remotely).  

Charliesheen: I don't see your hashrates jumping in the video.  That one GPU goes from ~395,000 to ~224,000 and back to 395 repeatedly.

[charliesheen]

Whats your cooling system like?

[Inaba]

Adequate.  As I said, the cards are at 74C and 75C respectively.  The cards don't start to throttle until +90C

[error]

Don't bother starting X, then?

[dingus]

Some mobos just don't like 5970s for some reason. I had one mobo that would take 1 5970 in one pcie slot, but the second I put in another 5970 in the 2nd pcie slot my computer wouldn't POST. No issues with the PSU or anything; it was just the motherboard.

[randomguy7]

I've got the same problem with an HD 5870 (on default clocks, max overclocking was 900). That system was running for month and I didn't change anything. Since about a day my hashrate regulary drops to about 100 MHashs, than going back up to about 333. I can see the temps (monitored with aticonfig --odgt) going down about 10° while this happens. My monitor flickers while the hashrate goes down. It's like the card is throttling through overheating but aticonfig --odgt shows max temps at about 75°C. Manually setting the fan speed to get temps about 60° doesn't stop this.
Doesn anybody know if the card has more temp sensors and how their values can be read?

[randomguy7]

Ha, I just found a solutions for my problem (still unknowing what the problem was), I hope this works for you too.

I have two boxes, both running DiabloMiner, one runs an bitcoin instance. Both DiabloMiner connect to the same bitcoin instance.
The change I made is that I now run an own bitcoin instance on each of them, each DiabloMiner connects to localhost. Now it seems to work totally stable again.
I wish there was a setting to change ports, thereby getting both bitcoin instances well connected (I've a PAT setup).

Btw, don't know if that's important, but I run an old edition of DiabloMiner (with the old kernel) as for some mystery reason it runs faster on my boxes (no matter what settings I try).

[Inaba]

So I went ahead and put a new motherboard in the box along with a new CPU.  Reinstalled the OS and used Natty this time instead of Maverick... exact same problem. 

So it's apparently the card.  I'm at a loss as to what would cause this, anyone got any ideas at this point?  As per Randomguy's info, I tried pointing that particular miner at a different pool, but the same problem happened.

Error: How do you use the miners if you don't have X running?  It was my understanding that the X subsystem needs to be in place and running for the ATI graphics drivers to be accessible.  Are you saying it's possible to run miners without the windowing system active? That would be excellent if so.

[twobitcoins]

It sounds like the GPU is throttling due to heat -- not the core GPU temperature, but the VRM temperature.  Are you overvolting?  If so, that is almost certainly the problem.  The first GPU on a 5970 has poor VRM cooling and can't take much overvolting.

[Inaba]

Hey, now yeah that could be it!  I totally forgot that there was the craptacular VRM issue on the 5970.  Yeah, I am overvolting it...

I will investigate that route, thanks!  That's totally gotta be it.