Two 5870s, one gives higher stale rate than the other?

[Mousepotato]

I don't get it.  I have two 5870s (on Diamond reference and the other a non-reference Sapphire) clocked identically and using the exact same Phoenix parameters.  They run the exact same MH/s rate (+/- .2 or so).  However, the Sapphire always has a lot more stales than the Diamond does.  Is this just pure luck, or could it be something hardware-based? 

[tidus_13]

luck

[Dargo]

If you are seeing something like 0.5% on one card and 1.5% on the other, that isn't really a concern. But if one card is giving you a very high stale rate, then there probably is an issue. Is one card running really hot? Have you tried backing off on the aggressive settings and clock for the card that is producing more stales? I think there are more subtle signs of instability before your card locks up, so if you are running near your lock up point, it might be good to back off a bit more. Also, I've found that I get the best results (in terms of shares/hour) when one card is tuned less aggressively than the other (lower Mh/s and lower aggression in phoenix). I'm not sure what is going on, but when they are on the same aggressive settings, it is as if they are fighting each other to get work.

[DrHaribo]

If one card consistently has a lot more stales than the other, then I would suspect it's actually the GPU miscalculating, and not the data being stale. Maybe the GPU is too hot, overclocked too much?

I can't remember if Phoenix detects this sort of thing? If Phoenix doesn't double-check GPU shares with the CPU, then try a miner that does and see if that is your problem. Works with DiabloMiner, my miner (BitMinter)  and surely some others.

[wallet.dat]

I think I have this same problem too.  Though the "stales" % is about the same for my two cards according to Deepbit, one card has a lot more "Rejected" shares than the other.  Are Rejected shares the same as stales?

[DrHaribo]

Some reasons a share (proof-of-work) may be rejected by a pool:

- it's built on stale data
- you already submitted the same share before
- it's invalid (GPU calculated wrong)
- it's built on data from a different pool

Only the first is really stale. But I think most miners and pool backends treat them all the same.

[jwzguy]

In all of my rigs, there's a card that runs 8-10C hotter than coldest card, just due to the fact that not all of them can get optimal airflow. I push them all to the highest stable 24hr OC at stock voltage that's possible.

The hottest card consistently gets 50-100% higher stales than the coldest card in each rig. (Still under 1% stale.) I'm guessing these cards are making more errors in calculation. Not enough to worry about it but it's an interesting trend.

If you're worried you can increase cooling, increase voltage, or decrease OC and see how it pans out over a few days.