In Dec/2012 I bought the new HD7870BB-Tahiti Edition card for my rig.
Now this card completes one Astropulse in 40mins thru Lunatics 0.40.
(Means two APs in 80mins with count=0.5 and device=2)
Not OC'ed.

Have anybody else made some experiences with this single GPU since release ?

yes my 7970 can handle 3 WU's at a time. However for optimal throughput I keep 2 CPU cores idle to feed the GPU

---

In a rich man's house there is no place to spit but his face.
Diogenes Of Sinope

Also. Few weeks ago I switched my cpu load to 1.00CPUs+0.50GPUs per ap wu.
Based on HTmode so these are 2 logical cores.

Have anybody else made some experiences with this single GPU since release ?

I don't have that card, I have a 6670. You have 4 times more compute units than me, you have cgn and I have vliw 5, your memory interface is 256 bits mine is 128, your clock is 925/975 I'm at 810.

All in all I'd expect you to be running ap 3 to 5 times faster than me, maybe more. But you are only about 2.5 times faster. You can speed this thing up.

Check your GPU utilization, if it's not at *least* 70%, idle a core to feed the gpu. Also you may need to tune the command line parameters unroll, ffa_block, ffa_block_fetch. My experience was that unroll in particular had to be set properly (I use 8, might be a function of the number of compute units you have, which is 20 in your case).

Have anybody else made some experiences with this single GPU since release ?

...Also you may need to tune the command line parameters unroll, ffa_block, ffa_block_fetch. My experience was that unroll in particular had to be set properly (I use 8, might be a function of the number of compute units you have, which is 20 in your case).

At this time my parameters in the app_info.xml are switched to:
<cmdline>-instances_per_device 2 -unroll 12 -ffa_block 8192 -ffa_block_fetch 4096 -sbs 128</cmdline>

In the next days I'll try to set the values higher.