Well I tried .33 setting per gpu, but without any success.
See no point in trying anything higher.
Anything I'm missing ?? Or just an Nvidia thing.
Max I seem to be able get without errors on each 690 is 4 WU's
Pathetic compared to what the 79xx's are pumping out.
It was kinda expected though, as I've read in many places that 2 (maybe 3) concurrent on them is the limit.
Was hoping to at least get 3 running per gpu (6 per card) BUT alas not.
11 of the 12 WU's were running fine, BUT 1 kept erroring out, one every 30 seconds or so.
So had to drop back to .5 setting again.
Might keep one of the 690's for gaming rig, try to sell other and pick up a couple 7970's for the SR-2.

Tried this today on a 4 core/thread i5 3570 with a GTX 570
<max_concurrent>5</max_concurrent>
<gpu_usage>.5</gpu_usage>
<cpu_usage>.75</cpu_usage>
Doesn't add up I know (4.5 cpu threads) but works perfect.
Not gone over 2 WU's on the GPU as I'm fairly certain that errors will result.
The above results in 3 CPU task running and 2 GPU tasks running, even though total cpu total = 4.5 on a 4 thread cpu.
No errors on cpu or gpu tasks, even when both the gpu tasks are either at half way OR finishing stage (which is when (I think) the cpu part gets done)
Any thoughts as to why this "exploit on the system" is working ??
Reason I tried was to get the cpu tasks in cache to run without losing the 4 gpu tasks functionality - worked !!
Interesting I reckon.

Anyone have errors crunching more GPU than CPU threads? Would think even at .5 cpu that the cpu wu would take control of the whole thread until finished or will it time slice between wu's? Am crunching 8 GPU now on a 3770K (only HCC1) and with the hyperthreading threads not being a full core, would think the system would choke. Not sure though.

I am crunching 12 concurrent GPU WU's (one HD 7750 max OC'ed) on 3 CPU cores (i7-3770 non-OC'able). After less than 1 day, tasks order in the way that only 2 - 3 of them are in the CPU phase (rarely 4 of them), so they nicely utilize 3 CPU threads I have dedicated to them . Each task run app. 37 mins.

The maximum I have (successfully) tried with app_config was 24 concurrent GPU tasks on all 8 CPU threads (each task finished in circa 1.40 - 1.45 h).

If you are trying to optimize GPU tasks only, do not bother with <max_concurrent>N</max_concurrent>. This is only for setting max CPU tasks and you can delete it from app_config. You are controlling the number of GPU tasks via
<gpu_usage>X</gpu_usage>
<cpu_usage>Y</cpu_usage>

With 7770 I will recommend to go with 3 GPU tasks per CPU thread, max 24 concurrent GPU tasks per GPU. If you have 8-threads CPU, app_config should include:

<app>
<name>hcc1</name>
<gpu_versions>
<gpu_usage>.041666</gpu_usage>
<cpu_usage>.333333</cpu_usage>
</gpu_versions>
</app>

Since 7770 is much more powerful than my 7750 (1.28 TeraFLOPS SP compute power vs. 819 GigaFLOPS SP for my 7750), I would definitely try 32 concurrent GPU tasks with:

<app>
<name>hcc1</name>
<gpu_versions>
<gpu_usage>.03125</gpu_usage>
<cpu_usage>.25</cpu_usage>
</gpu_versions>
</app>

Cheers

I gave 32 tasks concurrently on my 7770 a try on my 8-threaded i7 2600 with 16GB RAM.
If anything would have been fine, they are running for about 80 minutes with of course all 8 threads dedicated.
But it was a pity, that 13 out of that 32 almost ready tasks restarted at the 50% checkpoint - start of the second image.

So I stick to 3 tasks on 2 threads and 6 threads free for other CPU-tasks.
Elapsed time of 3 HCC GPU's concurrently in that config about 8.5 minutes.