I'm sorry, I missed this message originally, as I wasn't checking the forums as much. The performance difference is bad, especially when multiplied by the number of rigs I have. I think my biggest issue is that Luke has already stated that he doesn't care about Scrypt, so it's going to turn out exactly like CGminer did I think. However, it appears that Luke at least works with you, where CK and crew were a bit difficult from what I'm reading into things.
Luke is fine with someone maintaining the GPU and Scrypt code which I am up for if I can get support from users.
Here are my arguments for CGminer on this sample rig with 4 x 7950s:
-s 1 --intensity 20 --gpu-engine 1020 --gpu-vddc 1100 --thread-concurrency 24000 --lookup-gap 2 --worksize 256 -g 1 --expiry 10 --queue 0 --auto-fan --failover-only
Output with CGMiner:True GPU opencl Litecoin 3231.18329466 .02937 1610% 74 641.5 4963 76 0 1.9223 20
True GPU opencl Litecoin 3231.18329466 .02937 1610% 82 628.2 5001 82 0 1.9371 20
True GPU opencl Litecoin 3231.18329466 .02937 1610% 90 628.5 4782 99 0 1.8522 20
True GPU opencl Litecoin 3231.18329466 .02937 1610% 74 641.4 4961 79 0 1.9216 20
Output with BFGMiner:True OCL opencl Litecoin 3231.18329466 .02929969 1606% 73 616 23 0 0 2.441 20
True OCL opencl Litecoin 3231.18329466 .02929969 1606% 81 590 19 0 0 2.016 20
True OCL opencl Litecoin 3231.18329466 .02929969 1606% 88 592 26 0 0 2.759 20
True OCL opencl Litecoin 3231.18329466 .02929969 1606% 74 621 25 0 0 2.653 20
I use a rig of 4x7950s myself, and never had a big performance difference between the two. I get a solid 650 Kh/s from either with a 1100 clock and 1350 memory. However, I do not specify all those arguments. The only thing I specify in MultiMiner is:
-I 20 -g 1 --thread-concurrency 24000
Multiply this by 40+ rigs and I've got a significant loss in hashing power. Programatically, I can't offer any help, as I don't have that skill set, but I'd certainly love to test a solution, as I miss having the latest Multi-miner, really grew to love it.
Could I possibly get you to set me up with TeamViewer access to a single one of your GPU rigs that I could test on? It would help dramatically.