Excessive stales (almost 90%!) - 6990 - poclbm - LinuxCoin 0.2a - SSH

[brwyatt]

My 6990 is getting a VERY high stale count. Right now this is what the terminal window has in it:

Code:

08/06/2011 18:23:07, 2c483cbb, invalid or stale
08/06/2011 18:23:07, long poll: new block 000008eac09a99af
08/06/2011 18:24:50, e5532eb0, accepted
08/06/2011 18:25:01, 5525c072, accepted
08/06/2011 18:25:10, 4bcd1bf5, accepted
08/06/2011 18:27:26, 9bb4e8fa, invalid or stale
08/06/2011 18:27:26, long poll: new block 000013e15300ea0e
08/06/2011 18:30:05, 77168c5f, invalid or stale
08/06/2011 18:30:25, d7838ba0, invalid or stale
08/06/2011 18:31:14, 8cfdc8bc, invalid or stale
08/06/2011 18:34:35, 096097a2, invalid or stale
08/06/2011 18:34:37, long poll: new block 000000a0b18d945c
08/06/2011 18:38:03, 699668bf, invalid or stale
08/06/2011 18:38:03, long poll: new block 0000141eb3c09154
08/06/2011 18:40:52, a6e70bab, invalid or stale
08/06/2011 18:40:53, long poll: new block 000014be352e79c7
08/06/2011 18:43:22, a1614637, invalid or stale
08/06/2011 18:45:53, 5d00ac2f, invalid or stale
08/06/2011 18:48:53, long poll: new block 000011990e2eb8f2
08/06/2011 18:50:58, 832771df, invalid or stale
08/06/2011 18:52:44, 36df5ba2, invalid or stale
08/06/2011 18:52:52, 84ea6ecd, invalid or stale
08/06/2011 18:52:58, e36f8494, invalid or stale
08/06/2011 18:54:38, 84863ec6, invalid or stale
08/06/2011 18:55:34, 80c3d78f, invalid or stale
08/06/2011 18:56:13, 8c3862ec, invalid or stale

I had both GPUs clocked at the factory overclock of 880MHz (with the switch). As I am away at the moment, I am unable to change that, but have clocked them both down to 700MHz, and got the temps down to 76C. (Heat is an issue, which is why they are underclocked. The AC is set only for 86F for a while)

My 5750 on the other side of the room is clocked at its max, running at 77C, with a normal stale rate.

I have tried using -a1 on the 6990 instances of poclbm, but that did not fix it. I initially did have issues getting both GPUs to get to 99% load (one GPU only wanted to go to 20%), but I messed with trying commands like DISPLAY=:0.1 startx, which failed, but starting the miners seemed to work after doing that. Might them running on the same xsession cause that issue?

[brwyatt]

Any one? I would really like to get my 6990 mining at least a little while I'm away for the week.

[[Tycho]]

Any one? I would really like to get my 6990 mining at least a little while I'm away for the week.

What pool are you mining at ?

[brwyatt]

Eligius right now. My 5750 is having no issues there, though. Seems to only be the 6990. Will be fine for the first few... then it just starts getting invalids/stales. I have shut the computer down for now, and can't get it back up until sometime tomorrow, but... I can try it on Deepbit again if you think it is a possibility that the pool is somehow to blame.

I partially want to blame LinuxCoin, is it possible the drivers do not fully support the 6990? It wasn't having issues in Win7 x64 overclocked to 940MHz, but right now in LinuxCoin, even putting it at 600MHz it was getting a lot of stales. I was trying to keep the temps below 85C (though I know the card can work higher than that), and was managing to keep them in the upper 70s. So, they aren't over heating...

[[Tycho]]

Eligius right now. My 5750 is having no issues there, though. Seems to only be the 6990. Will be fine for the first few... then it just starts getting invalids/stales. I have shut the computer down for now, and can't get it back up until sometime tomorrow, but... I can try it on Deepbit again if you think it is a possibility that the pool is somehow to blame.

Actually I wanted to see if they are stales or invalids (different things).

Can you try to mine on the master core of 6990 only ? To see if there are some problems with slave or internal crossfire.

[brwyatt]

I actually tried just one GPU, and it had the same issue. Is there any way to tell if it is stale or invalid in poclbm? Or were you going to check your logs on your end if I was using DeepBit?

Is there anyway to check the Crossfire settings with aticonfig on Linux? I did try setting the getwork to 1 second (instead of the default of 5)... but that had no luck, so I am actually guessing they are invalids, not stales. It was working fine in Windows before I left, so I doubt it is the card itself, and I REALLY HOPE it isn't. If it is just an issue with LinuxCoin (or the Linux drivers) I'll just give my friend my passwords for the computer and have him under clock it to 500MHz and start mining and watch it for an hour. Only downside it I can't control or monitor it while it is in Windows.

I really hate that I don't really have more information. But that is what happens when you are away for the week... Probably should have done a dry-run before leaving, I guess...

[brwyatt]

Seems to be running fine at home now under Windows 7. Might have just been the increased heat or something from the A/C being off. Or something in LinuxCoin. Either way, seems to be fine now.

Thanks to those who tried to help!

[SmokeAndMirrors]

Have you lowered the memory clock? That tends to affect the stales quite noticeably as it causes your card to sometimes not have enough time to deal with certain bits before moving on to the next one, which causes stales/invalid(false information) from your gpu.

I had been running my 6970 stock clock of  1375 Mhz and had maybe 10 stales in 2 days. I lowered the memory clock to 600Mhz due to wanting to lower the heat output(not that it really needed it at 70C, It just helps me sleep at night) and while this reduced the cards temp by a sound 10C, my stales have quadrupled.

This may not be the issue your having, but it's worth a shot if you plan on fixing your heat issue.

[Jack of Diamonds]

Have you lowered the memory clock? That tends to affect the stales quite noticeably as it causes your card to sometimes not have enough time to deal with certain bits before moving on to the next one, which causes stales/invalid(false information) from your gpu.

I had been running my 6970 stock clock of  1375 Mhz and had maybe 10 stales in 2 days. I lowered the memory clock to 600Mhz due to wanting to lower the heat output(not that it really needed it at 70C, It just helps me sleep at night) and while this reduced the cards temp by a sound 10C, my stales have quadrupled.

This may not be the issue your having, but it's worth a shot if you plan on fixing your heat issue.

Unfortunately you can't lower the mem. clock on the 6990 much below 800, or it will bounce right back up.

I managed to get to 705mhz on one 6990, right after you switch to 704 it jumps back up.
At about 300mhz it would definitely run a few degrees cooler, but it doesn't work, so...

[SmokeAndMirrors]

That doesn't sound right. I encountered the same sort of issue while trying to underclock my cards at first using MSI Afterburner. It was due to a configuration error in the afterburner settings. It allowed me to set my clocks once, then anytime after that it would jump back to the previous set clock.

It would make no sense for you not to be able to underclock your mem less than 704 Mhz unless that was the absolute speed needed to run the card, but that is highly unlikely. That is a massive amount needed for a card to simply be functional. And even if you do underclock the memory too low, i'm pretty sure you would just run into technical errors rather than afterburner automatically resetting your settings, as it doesn't really have much in terms of filters for clock speeds.

[Jack of Diamonds]

Some people say you can only make 6990's memory go 100mhz below clock frequency, but that's obv. not true (it will go to the low 700's).

However it's also the only card variety I have that wont go to minimum mem. clocks.
Also AMD clock tool doesn't detect the cores at all so no way to force it from there. Afterburner works.

Then again, it is a dual gpu requiring some massive bandwidth, so maybe it's a fools errand looking for a solution that may not exist. Perhaps the card can't be run at 300mhz.

[runeks]

I'm having this issue as well. I have 54/440 rejected:

Code:

[231.33 Mhash/sec] [440 Accepted] [54 Rejected] [RPC (+LP)]

That's over 12%.

The interesting thing is that I am also using LinuxCoin over SSH.
Could this be the culprit? Something with the SSH connection causing it?
If this keeps up I will try to mine with the exact same program/commands/clocks etc. but not over SSH. If this reduces the stale count it seems that the SSH connection is somehow causing this.

[runeks]

Just an update from my side of things. Turns out it had nothing to do with SSH, it was the aggression level that was set too high. On my first setup I had always used AGGRESSION=12 when mining, cause this got me the highest and most stable hash rate. But apparently this was too much for the same card plugged into a new mobo (with a different CPU). Not sure how this can be, but it seems to be the case. Mining for a while at aggression level 8 got me 2.6% stales instead. That's acceptable.

Could it be that running LinuxCoin from a USB stick lowers the maximum aggression level? The PC it was in before ran Ubuntu from an SSD.

The CPU in the new machine is a "Intel(R) Core(TM)2 Duo CPU     E8200  @ 2.66GHz" while the one in my old machine (that can run the card at aggression level 12 fine) is a "Intel(R) Core(TM)2 Quad  CPU   Q9550  @ 2.83GHz".  I don't imagine this has much effect since the CPU isn't used that much.

[runeks]

^ I just switched to cgminer instead of Phoenix, and it seems to have reduced my stale rate (as reported by the miner application, not the pool server) from these ~2.5% to about 1.3%. Better, but still far away from your 0.3%.

With which pool are you getting 0.3% stales? I would like to try that out, to see if it makes a difference. If it doesn't, we can assume somewhat safely that it's not the pool.

EDIT:

Could I get you to ping the pool server that you're getting a 0.3% stale count with, and post it? I'm thinking ping time could be a factor as well. I'm thinking I will try:

• Using your pool
• Using a pool with a ping time that (for me) isn't greater than your ping to your pool

[runeks]

What a coincidence, I'm using Arsbitcoin as well. I'll try running the card at stock speeds at intensity 3 (using cgminer), and I'll see in the morning what the stale rate is then.
Good night!