The <4x PCI-e slot was a suspected issue for a long time but no one had actually verified it to my knowledge. I've only ever mined on 4x or greater slots. The transfer speed is only 500MB/s per lane with PCI-e 2.0, so it's possible that using only a single lane may cause a bottleneck.
You can verify this simply by moving around the cards. If the same cards are fine in an x16 slot but awful in an x1 slot, then you probably need the extra bandwidth.
I think that I am already actually doing that. Correct me if I am wrong. My system is all 5850 Reference cards, they are all on identical V,E & M settings on CGMiner and give near to identical performance in K/hs, HW errors and Rejects regardless of their using x16 or x1 extenders in their current slot positions. Therefore swapping them around would seem to be pointless. They are all underclocked to about 220 K/Hs. They dont seem to have a hardware problem.
It is the problem with stales that I am trying to resolve. 99% of all stales that I get occur just before and just after the LP. I have tried almost all the LTC Pools. I have found that if I use a pool with frequent LP's it just kills my profitability with "prev hash stale" stales. On a pool with fewer LP's I will still get stales at the LP but far fewer and sometimes none at the LP. It looks like the system cant cope with the transition/data from the LP. That's why I was interested when I saw the comment about the PCIe lanes maybe causing communication problems.
It is currently quite happily mining BTC at 2 Gh/s. I really dont want to start ripping the whole thing apart to "re-invent the wheel" if somebody has already solved this issue.
Edit:- After a fair bit of scratching around I found this command:- lspci -vvv Run that after sudo su and you will get mountains of info regarding the current PCIe status. Including what I think are the current lane allocations listed at LnkSta:
Run it without sudo su and you will not get all the available data.
There is more there than I care to try to understand.