well your posting isn't really clear when your URL gets replaced by [Suspicious link removed].

And what on earth does the second sentence mean?  My uncle has no boobs

Christian

Difficulty is per worker.  You can set each one to 16 and see what the pool switches you to.  512 will be too much for individual workers.  Keep in mind the higher you set it, the less bandwidth you'll use, but the more variance you get.  So when I say "too much", I mean your variance will be high.  It'll still work though, just like 2048 will.

I use 256 for ~3000kh/s (multiple 280s), and 128 for < 1000kh/s (one 280).

M

That's because MRC is at N factor of 10 which is a max batch size of 2048.  Default batch size is 1024 FYI.

So when Cudaminer goes to launch the kernel it looks at N factor and your -b batch setting and uses the minimum of the two.

interesting. -H1 does 4 times more transfers over the PCI express bus than -H 2. Surprising that it's faster. I intend to get rid of all the transfers and do the hash evaluation on the GPU soon (like I do when hashing Keccak). This will allow us to run nVidia cards with 1x connections through risers without performance impact.

Christian