Here's what I've observed with a 5970 and now a 5870 as well.  Hope it helps someone.  First and foremost, a few posts back someone noted to use the "Known Best Settings".  I added the -f5 switch and I went from about 560Mhash/sec to consistently over 600Mhash/sec on the 5970 alone. 

Next, the first post in this thread says . . .
"5970 is actually two 5870s with internal crossfire. OpenCL doesn't work with crossfire. It must be switched off. You can't switch it off on Windows. For now, 5970 can be used fully on Linux only."
I have used the 5970 for over a month on Windows 7 and get results from both cores, about 300Mhash/sec each.  So I don't know what that comment is all about.

Next, I finally was happy with my results with the 5970 and I can never leave well enough alone.  So I bought a 5870 tonight and threw that into my motherboard as well.  At first I was getting "Disabled Adapter" in Catalyst Control Center, although I could see the third GPU (now as GPU #1) in MSI Afterburner.  It did nothing and just sat there very cool.  poblcm only recognized the two 5970 cores and the CPU.  After looking at posts on Google as far back as 2008, someone said, "Plug in another monitor".  Not having another monitor, I plugged in my monitor to the 5870 instead.  Now, the 5870 GPU was the only core recognized by poblcm along with the CPU.  So I ran poblcm with a device=1 and it worked!  Then I unplugged the monitor from the 5870 and replugged it into the 5970.  Now, even though the 5870 was the disabled adapter again, poblcm was still running on the 5870.  Then I ran poblcm again on device=1 and device=0 and both of the 5970 cores were running alongside the 5870, for a total of about 875Mhash/sec.  Pretty good.  So that's just FYI for anyone running multiple 5870's, 5970's or combinations thereof.  I wish there was a way to get Catalyst Control Center, or ATI Stream, or whatever ATI driver it is to enumerate all three cores so I would not have to unplug and replug the monitor into different cards every time I want to run poblcm on the 5870 and 5970 at the same time.  I would suspect this would be a similar problem, and a similar solution, for running poblcm on  two 5970's on Windows.

Miner is now fixed. It found blocks on testnet. Added verification of kernel result in host.

I will highly appreciate any reports about hash speed with this version.

I'm getting an extra 10-20khash/s on my 5970. Kudos sir.

Anyone found a block with this new kernel on this miner? I know difficulty recently went up but my last block found was on my 5870 with the previous kernel on 01/24/22 11:46... more time has elapsed than I would have expected.

Where are you getting these .debs from?

Hello m0mchil.
I've been using your client a lot lately with a GTS 250 card...
Today I got a new shiny ATI card and I can't use it with your miner, even if they get recognized by software like GPU Caps Viewer! Let me explain...
I got both VGAs on my PC, ofc there's no "Crossfire" or "SLI" since they are from different vendors. The problem arises since the nvidia uses the CUDA OpenCL, the ATI uses (of course) the Stream impementation. This way, my 2 physical cards reside on 2 different cl_platform....
Now, I never touched python in my entire life... but those 2 lines looks... suspicious


I really suspect this is what's giving me problems: I can't see the ATI as listed in the devices I can choose! I could physically swap the cards, or maybe unistall and reinstal all the video drivers so that the first OpenCL platform is the ATI one, but this doesn't really look an elegant solution IMHO
Suggestions for solving my problem are welcome! EG: patches, or how to compile 2 different executables for windows.

I'm under windows! Everything works pretty well right now if I was to be honest, since I'm using 2 different miners, one for CUDA and the poclbm for OpenCL. But having to switch to another miner was soooo sad I loved the m0mchil one

I can confirm this trick works on my 5970 + 5770 setup. First start 2 poblcm instances for the 5970, then unplug monitor, plug into 5770, start instance of poblcm. 5970 + 5770 goodness!

~800000 k/hash/s

Fixed issue with lost results due to single output per kernel run (thanks OneFixt, ArtForz).