With a GTX 970 SSC @ stock I'm getting ~165kh/s with the following:

sgminer.exe --kernel hmq1725 -o http://127.0.0.1:22440 -u u -p p -I 16 -w 256 -g 2

Another 40Kh/s than my CPU and 40W more on the power draw, not great but it's openCL on a nVidia GPU Little bit more consistent going solo, pool diff jumps around too much and causes more rejects than CPU mining. I'll stick with CPU now but I'll look forward to a CUDA port!

Just a FYI on the eff value, that value isn't the output from the PSU into the case, it's how efficient the PSU is at converting AC -> DC and thus it's draw from the wall. Your 850W PSU is rated at 850W and will always output that for the components into the case if needed but it would then pull ~935W from the wall. That being said I always build a ~10% safe value into my builds anyway so I'm not running the PSU at 100% all the time and, as bigjme pointed out, the closer to 100% load the lower the efficiency gets. PSU manufacturers should post their curves online somewhere to let you target the bet place to keep your PSU load wise as well.

Thanks! Clean install on a different machine synced fine so I think I might have some remnants of the old wallet causing problems.

Anyone got a list of nodes (including port)? Tried the ones listed in the OP and pools and couldn't get any of them to work.

Make sure you are running a version >=340.52 of the nVidia drivers. Latest compiled versions of most miners require it now since the authors are ditching CUDA 5.5.

I mentioned this a couple pages back but didn't see a reply, I think the issue you are running into is because you are comparing hashrates from overclocked (stock or by the user) to your 'vanilla' 750Ti's. I have 3x EVGA 750Ti's and the rates I get from them non-OC'd are exactly what you are getting right now. However, raising the TDP and overclocking them to +142 core/+528 mem gets me the hashrates (sometimes a bit better) than what is generally quoted. Stock OC'd cards are in the range of  +100 core out of the box so that explains why some people with "stock cards" are getting better rates than you are.

If anyone is running a stock 750Ti with reference clocks and getting ~2.7MHs on X11 I'd love to see it

Can't be sure but you might be comparing 'stock' 750Ti's with versions that are coming pre-overclocked like the EVGA 750Ti SC. My stock 750Ti's only get 6.9-7MH's on DMD but when I overclock them to +160/+204 I get ~7.8MH's on them. So if you want those numbers you need to overclock your cards

I've got 3x EVGA 750Ti's on my X58 (no risers) with the TDP unlocked, +12mv, +151 GPU clock and the board is rock solid. I'm not doing scrypt/nscrypt though and haven't tested long term with it, short term it works and there is no burning smell

I'm not sure it will effect the mining community too much. This is going to be big news for places doing any sort of virtualization, especially network layer. Hardware network gear makes heavy use of FPGAs/ASICs to handle the majority of the routing/switching that goes on in the platform and the CPU is really just there to observe things. VMWare (maybe others...we're a VMWare shop) came out with NSX that handles a lot of networking in software, it's really flexible but it means the CPU on each server is having to handle those route/switch operations. If you could off-load that work to a FPGA you get back CPU time and can start getting into the 40-100GB+ realm with decent latency.

You might also see them used to handle crypto needs for things like SSL/password hashing/other offload needs. CPU time is still the reason some people are using RC4 when dealing with SSL or SHA1+salt hashing passwords stored in a DB. A general purpose device that could accelerate all those operations, allow for stronger algorithms AND reduce CPU overhead is a win for all parties.