Received a new GTX 470 which I had planned to run along with the GTX 260 already in my machine. Had upgraded the PSU to 900 watts. The physical install went well and as the cards are very different did not want SLI.
Installed new driver, but did not remove old (new 197.41 and old was 196.21). Attached lone monitor to GTX 470. Started BOINC mgr (6.10.45). At first things looked good as BOINC showed two GPU's being used. Old EVGA Precision showed wild numbers for GTX 470, but valid ones for GTX 260. Installed new EVGA Precision from disk. This time all numbers were valid except Core Clock for GTX 470 which was blank. Conclusion here is that someone does not want you to adjust the Core clock higher as at load the temp is already 88 degrees C.(Determined this later after I removed GTX 260). To shorten this up a bit, it seems that the GTX 470 was doing nothing even though BOINC said two GPU's were running. EVGA Precision showed GTX 470's Core clock going to zero as soon as it was started (on the graph part) Was trying to run Milkyway.
Removed GTX 260. Started processing Milkyway again. First Milkyway unit (probably version 3 - I did not check) completed, but everything else cancelled accept another version 3 unit. Have read some of the forums and know that there is a problem running "standard" Milkway.
So the question is, what can I do to successfully run both GPU's. Seems like a problem unless there is logic somewhere to not try to run work units that can not be handled. The plan was to run Milkyway and GPUGrid at the same time or Milkyway and Seti with the GPU appropriately selecting only what it could run or having CC-config help in this. Yes, I now know cc-config does not currently have this ability, but it does seem to have some special commands available only to World Grid- so maybe?

Hello,

I don´t know about the GTX470, but I have had run a GTX260 and a 8800 GT together in one host. Same as you no SLI was possible. At first only one GPU ran until I changed the cc_config.xml file in the boinc-folder.
I added: (in the options-section)

<use_all_gpus>1</use_all_gpus>

After that both GPU ran well on GPUGrid and Collatz.

Thanks for the response.
I did some research before the GTX 470 arrived and discovered the need for the GPU switch in cc_config, so it was there. I now know that the architecture of the new card is very different "programming" wise as well as hardware wise (which we all knew). I am hoping that BOINC mgr (or whatever process is necessary) will be able to differentiate between work units that will work on a GTX 470 and ones that must be redirected to the "older" style cards. Later. JB

Thanks for the response.
I did some research before the GTX 470 arrived and discovered the need for the GPU switch in cc_config, so it was there. I now know that the architecture of the new card is very different "programming" wise as well as hardware wise (which we all knew).

Fermi is basically identical from a programming perspective to the GT200, and code which was written for the GT200 will run on Fermi without modification. The only reason why the current milky way applications don't work on Fermi is because the GT200 has been "hard coded" into the host support code the app uses. Leaving aside QA and testing, precisely 1 line of code in the current 0.24 CUDA client requires changing for the application to run on Fermi cards.

Thanks for the response.
I did some research before the GTX 470 arrived and discovered the need for the GPU switch in cc_config, so it was there. I now know that the architecture of the new card is very different "programming" wise as well as hardware wise (which we all knew).

Fermi is basically identical from a programming perspective to the GT200, and code which was written for the GT200 will run on Fermi without modification. The only reason why the current milky way applications don't work on Fermi is because the GT200 has been "hard coded" into the host support code the app uses. Leaving aside QA and testing, precisely 1 line of code in the current 0.24 CUDA client requires changing for the application to run on Fermi cards.

Which begs the question, it it worth the developers time and trouble for the minuscule number for Fermi cards out there to re-make the applications that are going to be replaced with an new version real soon now?