0) Please, watch carefully what results your CUDA-enabled host returns and stop using CUDA SETI MB till new version will be available if you see excessive invalid results rate.
1) This package attached to this post.
Targed hosts: Windows x86, SSE3 support for AP, CUDA support for MB.

2) It consist of modified SETI MB CUDA and current SSE3 opt SETI AP binaries with corresponding app_info.xml file
3) Modification that I have done increases CUDA worker thread priority in SETI MB CUDA that allows more fully GPU usage while keeping all CPU cores busy too. That is, using of this build can increase total performance of your host for BOINC tasks.
4) MB binaries based on CUDA MB sources from Berkeley's repository (with small modification), opt AP is just repacking of current Lunatics opt AP release (SSE3 build).
5) It's not "official" Lunatics release so you could blame only me (or yourself, or BOINC bugs and so on and so forth) for any issues you encounter.
6) For best CPU and GPU usage I recommend to set number of processors available for BOINC to real_number_of_cores+1. This will mitigate current BOINC bug with CPU+CUDA scheduling and will allow fully load CPU and GPU.
Here is example of minimal cc_config.xml file you need (it also attached to this pos - example for quad)
<cc_config>
<options>
<ncpus>enter_number_of_cores+1_value_here</ncpus>
</options>
</cc_config>

 You should put it in BOINCdata folder (folder that contains projects subfolder).

7) Installation instructions are the same as for any opt app: stop BOINC, decompress all files in archive into SETI project directory, restart BOINC.

Notes for current build:
Ok, new bug fxes available - so new modded app available too
VLAR bug still with us although most advanced GPUs (fastest ones) could be free from it already - worth to test. Midrange like my own 9600GSO still suffer from it, especially under Vista.
Again, it's wort to try under XP - there are chances it will work under XP better.

Current bug fixes fight mostly with different overflows. Actually, they should eliminate overflows at all. So, please, report any overflow you will get if it not from VLAR and not from task was ran after driver crash w/o OS reboot.

ADDON: You need 180.48 or upper videodriver version to run this build successfully.

ADDON5 (for V7):
Here is new package version.
It contains the same CUDA part as before (equivalent to 6.08 stock with some mods) and latast opt AP SSE3 version (r103) )
This package will provide suboptimal performance in CUDA part versus V5b (because VLAR CUDA processing very inefficient now and V5b just abort VLAR tasks) but provides performance increase in CPU part because of opt AP rev103 increased performance.

ADDON6 (forV7 VLAR autokill mod):
And one more version.
Now changes in CUDA part of package. This build based on 6.08 sourses but with VLAR auto-kill mod enabled. It sould provide best overall performance at cost of reissuing VLAR task to another host. NOT for using on SETI beta project.

1) Yes, it's rebuild CUDA MB with increased worker thread priorty. Yes, it' not real time. It's still lower than priority of all normal processes and their threads, it's just higher than priority (1) of all other BOINC project worker threads.

2) No, combining AK8 with CUDA MB still not possible. It will require merging CUDA part into AK8 opt app and doing some synchronization between different instnces of new AK8.
This will be done eventually I hope. My current package is just temporary solution of course.

I think on the seti forums they were saying to upgrade to 6.4.5 for the CUDA app.

 Thanks for your work on this Raistmer.   

When installed on my system with a 4 day cache the system immediately goes into EDF mode and estimated time to completion is all messed up.  This happens with the stock version also so is not your fault.  Question is how does one deal with the messed up times and cache?  Just ignore it?  Will it eventually go out of EDF mode and completion times become more realistic if I continue to run it?

Yes, just ignore. My own host had same problem but now it seems it came to some realistic time predictions and all 5 tasks work in usual mode.

My GPU is working ok. 6 hours

http://setiathome.berkeley.edu/results.php?hostid=4708858&offset=40 and all time working hard
GPU 62 C temp.

overclocked from 700/170 to 720/1800 and ok (9600GT)

temp 63 celsius stable at any freq