GPU folding on WHS 2011

[Toconator]

Thanks to a generous benefactor I have added a 650Ti to the "garden". Slapped it in the workshop PC last nite and the trickle-down-effect was to put a 550Ti in the WHS box to bump a GT240 out. Problem is the 550Ti is barely getting 800 ppd more than the GT240 was getting. Runing the same v6.41 (systray) as before. WHS box is a hex so it has 2 more cores than the quad the 550Ti came out of. I find this very odd as it has had no problem getting 13-15K ppd on a regular basis for months. Some kinda weird Server voodoo going on here ...

 

[dragonstongue]

I admit I do not know much about folding and what should be had or not in regards to ppd, however, there is a massive difference in performance real world and theoretical between the 550ti and the gt240. Do a clean driver install might help if you have not, and maybe play around with the drivers a bit, as I know for myself doing bitcoin mining driver sets can have a massive difference in hashes or in this case ppd.

quick goggle says 9k-12k depending on clocks for the 550Ti, 3200-5800 on the GT240, so yeh, if you are only getting 800ppd difference, something is up with the WU you are working on or the drivers not letting it run the way it should, given the 500 series is more tailored to "work" tasks and the 600 series by and large is optimized for "gaming" tasks, the 200s well they are older of course but "balanced" design.

 

[bman212121]

If I'm not mistaken any 5xx cards won't work correctly on v6 of the software. You will want to upgrade to the v7 software to use 5xx and above. Were you running v6 or v7 on the old box?

 

[Toconator]

If I'm not mistaken any 5xx cards won't work correctly on v6 of the software. You will want to upgrade to the v7 software to use 5xx and above. Were you running v6 or v7 on the old box?

v6.41 on all GPUs in the garden, usually run in systray for easy referral and/or pausing for gaming. As I run CPU folding as a service and v7.xx doesn't seem to let you split folding types between service/console/systray and errors out on GPUs if you elect service mode for CPU, I am forced to utilize v6.xx for GPUs in conjunction with v7.xx for CPUs where I've upgraded the F@H program on fresh installs. Never had any problem with v6.xx on any GPU before and am currently getting 24K ppd or more on the 560Ti and 17K ppd on the 650Ti which was swapped in to the existing directory from the vacated 550Ti so I don't think v6.xx is the holdup here. I'll try a fresh driver install as suggested in the other post and see what happens. There's no adjustment for hardware acceleration in advanced properties in WHS, and besides the performance of the GT240 was not effected and it is still getting the same ppd on the older Q6600 I swapped it into. Something weird going on here ...

edit: running same driver (306.97), will reinstall anyways ...

 

[musky]

6.41 should be fine with that card. You only need v7 for 6xx cards. You might need the '-forcegpu nvidia_fermi' flag to make it grab Fermi units.

 

[Toconator]

6.41 should be fine with that card. You only need v7 for 6xx cards. You might need the '-forcegpu nvidia_fermi' flag to make it grab Fermi units.

It's not that it won't run a WU musky, it grabs the same P80xx as the 560Ti and 650Ti (650Ti DID require the fermi flag under v6.41) but only gets 4700 ppd instead of the usual 13-15K ppd. The GT240 gets the same ppd after transplant into a Q6600 rig as it did in the Server box so it appears not to be the OS (WHS vs WinXP). Puzzling ...

 

[rhavern]

Fermis have different power saving abilities vs the older GPUs. This was the final nail in WinXPs coffin for me, as the XP drivers do not have the "Prefer maximum performance" in the driver, only Vista and newer. As a result, in XP I had downclocking.

Run something like MSI afterburner and see what clock your GPU is at. My symptoms: started at maximum clocks, after a few minutes, downclocked. If this is the case, I also had some luck with disabling the screen saver. You may need to let the system go to idle before you see any evidence of downclocking.

Let us know what you see.

 

[musky]

650Ti DID require the fermi flag under v6.41)

That is somewhat odd given the a 650 TI is not a fermi card. I thought that v7 was the only way to get Keplar support. Apparently not.

I like rhavern's thought about the card downclocking. You may need GPUz to see it - dropping to 2D clocks. That makes sense anyway.

 

[Toconator]

Fermis have different power saving abilities vs the older GPUs. This was the final nail in WinXPs coffin for me, as the XP drivers do not have the "Prefer maximum performance" in the driver, only Vista and newer. As a result, in XP I had downclocking.

Run something like MSI afterburner and see what clock your GPU is at. My symptoms: started at maximum clocks, after a few minutes, downclocked. If this is the case, I also had some luck with disabling the screen saver. You may need to let the system go to idle before you see any evidence of downclocking.

Let us know what you see.

Yeah, I know. Win XP is where the GT240 went and it is still getting the same ppd as it was on WHS (you can run GPU Folding in service mode in XP to get around power saving issues if necessary). The problem is with the 550Ti on WHS which is now getting way less ppd than it was/should. The power mode in global settings is present in the WHS2011 driver and I just noticed it had reverted back to adaptive after I did the driver re-install suggested in an earlier post. It was set to prefer max performance before the reinstall, but I'll do a reboot after switching just to eliminate any issues with settings not sticking while in use. I already have Afterburner installed and checked clocks: 900/1800/2052 (4100) so it seems to be running as it should. The mystery deepens ...

 

[Toconator]

That is somewhat odd given the a 650 TI is not a fermi card. I thought that v7 was the only way to get Keplar support. Apparently not.

Yeah it actually grabbed a core 11 and then stalled and wouldn't run. Soon as I deleted the core 11 and added the fermi flag it grabbed a core 15 and off it went. Gets about 2-3K ppd more than the 550Ti did working out of the same directory on the same WU's. Smaller card, less power and slightly higher fps when my kid fires up League of Legends @ 1080p. Too bad the budget was depleted during the Boxing Day sale after Xmas when I coulda grabbed a couple for ~$80 each as I have lots of empty PCI-E slots available. Great little card putting out decent ppd for the [H]orde without making my utility meter spin alot faster

 

[rhavern]

Ah ha, it is that pesky driver setting again. Good to see a result.

 

[Toconator]

Ah ha, it is that pesky driver setting again. Good to see a result.

No I don't think it was the driver setting as it had been set to prefer max performance before the driver re-install and only reverted back to default of adaptive after the driver re-install, and while double checking after reading your post I noticed the change. Switching it and rebooting had no effect, clocks still read 900/1800/2052 in Afterburner and GPU-z but only getting 4700 ppd on WU 8071, usually get approx 13K with this card. Somethings amiss ...

edit: after finally crawling thru that WU it is now getting 13K ppd on a P7626 (whew!) Must've been something funky with the WU or something in the directory I run it from which was cloned from the other PC I transplanted the 550Ti out of ? Never seen anything like this before. Will have to keep a close eye on it. Will repost if anything useful needs reporting.

edit: grabbed another P80xx WU and back down to 7400 ppd. Now I notice in the HW monitor that GPU usage is only 51%. What could be causing this? Set to prefer max performance in driver and the power profile of the Server is set to High Performance so nothing should be throttling it. Videocard clocks are max as they should be for 3D mode. Did not/do not have this issue on PC it was transplanted from, nor with the GT240 which was folding here before the transplant (but it was folding small core 11's). This is effing weird ...