GPU2 expires, is there a new client? [Yes, download it now]

[Fadamor]

The NV_FAH_CPU_AFFINITY environment variable is still usable by the new client.

[theo343]

God this got me nervous as i have 4 GPUs folding at a location im not traveling to in the weekend. Hope it will continue until monday as said.

[ahu]

A similar thing happened with the old GPU1 client. Windows (XP and Vista) had to know that there were monitors attached to the cards for it to work. I'm not sure how it managed to bypass this in pre beta11 clients.

Did I get this right: with the new client, also in XP you now have to have a monitor or dummy attached?
Because this was my main reason to move away from Vista (I have 2 x 9800 GX2 so would need four outputs enabled).

I thought the problem was the Vista OS, since any CUDA client card had to have the output enabled. So I thought it wasn't in any way related to the GPU code.

[uncle_fungus]

The best answer I can give is, I don't know how or why the detection code has changed in it's behaviour. All I know for certain is that in order for multi gpu detection to work, windows needs to think there is a monitor attached to each device.

In the past, it wasn't necessary for a monitor to be attached permanently, just long enough to trick windows into thinking there is.

[ei57]

The best answer I can give is, I don't know how or why the detection code has changed in it's behaviour. All I know for certain is that in order for multi gpu detection to work, windows needs to think there is a monitor attached to each device.

In the past, it wasn't necessary for a monitor to be attached permanently, just long enough to trick windows into thinking there is.

You don't have to have a monitor attached to both cards in XP, it's working fine with just one. Only make sure that you have deleted old client.cfg and perhaps queue.dat and unitinfo.txt for all but the first gpu. You will then have to input new info in the client.cfg and should be able to run more than one gpu.

After upgrading to beta 20, I was only able to run one gpu, until the above action was taken. After that, no problems.

[leexgx]

@ei57
i see if this works

@uncle_fungus
the Monitor problem is an Vista Video Driver only problem ,this does not affect XP as Nvidia do not need folow the same rules on XP based OS (win3k) as thay do on Vista,
unless the folding team did something that Broke GPU2, beta 8 did not have this problem from beta 11 upwards i not been able to use more then 1 GPU2 client with that error about not suported video card, hopefuly the above will fix this

this has been pointed out more then 10 times by admins and users and me (3 times in this thread alown)

[PCZ]

Someone has changed the detection code.
Behaviour in XP is now just like Vista,
It is a PITA for users without monitors on there VC's.

[7im]

Someone has changed the detection code.
Behaviour in XP is now just like Vista,
It is a PITA for users without monitors on there VC's.

Are you sure it wasn't a driver update that caused it?

[PCZ]

If you mean VC drivers then no the driver is 177.35.
Besides i have 3 different dual GPU setups and the behaviour is the same on all of them.

One was easy to fix as it is my gaming rig and the monitor has 2 inputs so i just put the vga input into the second card.
The others were harder as they are headless and i had to override the GPU detection to get the second client working.

[7im]

Thank you, that's good feedback!

[leexgx]

guess i keep on useing beta 8 (just set the date before start), or use the old Past work around to make more then 1 card work

PCZ how do you force GPU2 to work ?

fixing pcs at this time so not had time to play with FAH

[uncle_fungus]

You can bypass the GPU detection routines in the client by using the -forcegpu flag (documented on the appropriate FAQ pages).
You simply pass one of

Code: Select all

ati_r600
ati_r700
nvidia_g80

to this flag, in addition to the regular -gpu flag.

[leexgx]

cool guess that makes it very usefull for quado cards or ATI headless cards

[Drugless]

Both 8800gts working again. Previous problems may have been related to .cfg file.
I deleted client.cfg, queue.dat and unitinfo.txt in the gpu2 folder. With no clients running, started the gpu2 shortcut and the config screen popped up. Added -gpu 1 in the advanced section and hit OK. Checked the temperatures and the right gpu was heating up. Restarted the other (default) client and saw that both gpu's were engaged.
Only one of the boards has a screen connected.

I did as suggested by ei57 above and got both clients working BUT:
The minute I stop and restart the clients I'm back to same problem.
Repeat process and leave clients running. Will check later to see WU's go through.

[uncle_fungus]

To clarify, what's going on with the new clients is the expected behaviour since our detection method relies on Windows assigning a display to each adapter (on both XP and Vista). The detection code itself has not changed between client versions (I know this because I'm looking at the source right now), so the behaviour you are observing is most likely due to a quirk in the original build process of beta 8.
I've just done a clean build of beta 8 here, and tested it on a multigpu setup and I got the expected behaviour. The second GPU is not detected until Windows associates a display with the adapter.