It was suggested that I=8 be used for 5xxx series ATI cards, while I=9 be used for the 69xx series cards.
I did some playing around after a few stalls here and there and actually found that for my quad 6950 rig, I actually get a higher share/min rate ("U" value) when I use I=8 rather than I=9, as suggested.....and no GPU stalling or driver crashing with a much higher overclock.
Is I=9 still suggested for 69xx cards ?
There is no hard fast rule because there are a lot of variables. Essentially Intensity determines how big of a chunk of hashes the GPU processes at one time. More hashes at once = less overhead however when a block changes it means more wasted hashes because the GPU can't be stopped once it starts a "run".
ckolivas can correct me if I am wrong but the I rating is more based on the hashing power of the card. The more hashes you can complete in 1 second the larger the optimal "batch" should be = higher I value. However other factors can alter that too. If your pool is issuing higher number of LP a lower I value reduces the amount of potential stales. So really it is a balancing act between stales & raw hashing power.
I am currently still using 2.0.6 and love it......insane performance. I pretty much use I=8 for everything, except for working-workstations, where I use Dynamic Intensity and it all works great.
If you have more than one GPU and all your monitors hooked up to a single GPU you can set the "monitor" GPU to dynamic and all other cards to a high I value. For example my 3x5970 workstation has 1 GPU set to D and the other 5 set to I=9. If you have multiple monitors hook them all up to the same GPU to minimze the number of GPU which need to be set to dynamic.
