problems running 803x GPU WU's (may be DirectX 10 related)

[chaosdsm]

Running the "public stable" 7.1.52 client with SMP & GPU. I've solved all issues I've been experiencing except one.

Whenever I get one of the 803X series GPU work units, the system becomes choppy and un useable until I pause the GPU client. When I try to open IE9, the screen is jumpy, and the mouse pointer skips across the screen, when I open favorites, the mouse pointer sticks on every single favorite as I'm going down the list. I have to move the mouse rapidly then wait 5-8 seconds for the computer to catch up so I can see where the pointer is at, then adjust from there just to open or close a program, window, etc... It only happens when I'm folding an 803X series 3843 point GPU work unit.

With all other GPU clients and all the SMP clients so far, I can continue using my computer & processor or GPU time is released as needed. I can even play games like the DIablo III Beta with other GPU WU's folding.

System specs:
Gigabyte 990FXA-UD3 motherboard
AMD 1055t CPU - currently at default 2.8GHz clock speed
PC Power & Cooling Silencer Mark II 760 Watt PSU
Gigabyte 460GTX 1GB dual-fan video card - tried different clock settings from reference design clocks 675 core / 1350 shader / 1800 memory up to 750 core / 1500 shader / 2000 memory. Maximum temps at 750/1500/2000 = 55C

GPU Drivers tried:
285.62 WHQL
295.73 WHQL
296.10 WHQL
301.24 BETA

[Zagen30]

I believe 803x work the GPU harder than other projects. GPUs still don't have any real scheduling, so you end up seeing lots of lag when two programs want to use significant GPU time since the card is unable to process those requests well. For the IE (and other web browsers) problem, disable hardware acceleration. For gaming you'll just have to pause the GPU slot (right-click the slot to be able to pause it without pausing any other slots) and remember to start it up again when you're done gaming.

[bruce]

We don't recommend beta drivers, but others on the forum have tried 301.24 and say it doesn't have the screen lag of the 2xx.xx versions. Does it change the 803X WUs? Current thinking is that it seems to have nothing to do with FAH, itself, but rather with GPU drivers. . . unless you have better information.

[bruce]

I tried beta 301.24. (Upgrades from CUDA 4010 to CUDA 4020.) FahCore_15 now crashes on this 803x. Time to roll-back to 296.10

FahCore_15 is version 2.22 and Windows reports

Problem signature:
Problem Event Name: APPCRASH
Application Name: FahCore_15.exe
Application Version: 0.0.0.0
Application Timestamp: 4f0372d7
Fault Module Name: StackHash_0a9e
Fault Module Version: 0.0.0.0
Fault Module Timestamp: 00000000
Exception Code: c0000005
Exception Offset: 8bc35900
OS Version: 6.1.7601.2.1.0.768.3
Locale ID: 1033
Additional Information 1: 0a9e
Additional Information 2: 0a9e372d3b4ad19135b953a78882e789
Additional Information 3: 0a9e
Additional Information 4: 0a9e372d3b4ad19135b953a78882e789

[Ripper36]

I am running the new drivers on four GTX570 cards - allo good so far!

[chaosdsm]

We don't recommend beta drivers, but others on the forum have tried 301.24 and say it doesn't have the screen lag of the 2xx.xx versions. Does it change the803X WUs? Current thinking is that it seems to have nothing to do with FAH, itself, but rather with GPU drivers. . . unless you have better information.

I usually preffer not to use beta drivers either, but this is quite annoying. However, folding wise, I don't see any difference with the 301.24 beta's. The only "better information" I have is that this ONLY happens with the 803X series of WU's. All the other GPU WU's work flawlessly, even when gaming & folding at the same time, which 'seems' to suggest that it may be the work units. Wish I could fold one of the 803X GPU WU's on my ATI card then we could possibly eliminate drivers, but I imagine these are Fermi only...

[Michael_McCord,_M.D.]

The only one of my 15 Fermi cards that had crashed with this setup and WU was somewhat OC'd using EVGA's Precision. I have noticed that with these newer WU, they don't tolerate OCing at all. If you have increased your GPU core beyond stock, try lowering it back to stock speeds and trying the 301.xx driver again. My card reset to stock speeds has not crashed since on driver 301.xx

[chaosdsm]

The only one of my 15 Fermi cards that had crashed with this setup and WU was somewhat OC'd using EVGA's Precision. I have noticed that with these newer WU, they don't tolerate OCing at all. If you have increased your GPU core beyond stock, try lowering it back to stock speeds and trying the 301.xx driver again. My card reset to stock speeds has not crashed since on driver 301.xx

Although I too say start at stock... my card is now overclocked at 760/1520/2025 with the beta's & currently running a 8031 WU, and has no problems (thus far 2hours into the OC) other than the stuttering mouse / screen when trying to do anything. I've been running at 750/1500/2000 with 301.24 drivers for 40 hours without issue, though I forgot to un-pause the GPU client when I went to bed last night, so it was only folding the 803X wu's for about 31 of those hours. Factory stock is 715/1430/1800, reference stock is 675/1350/1800 - original 1GB 460GTX.

[Michael_McCord,_M.D.]

I now have to also report seeing severe lag with the 8031 WU and 301.24 drivers, specifically it is WU 13,22,45. Resolved after rollback to 285.xx driver

[bruce]

Would it be reasonable to remove the "V7.1.52" from the title of this topic?

Granted Projects 803x act differently than most earlier projects.
. . . because of

• a change to the client
• a change to the drivers
• a change to the FahCore
• different characteristices of the data being processed
• something else
• some combination of those things

The previous post strongly (as well as the title) strongly suggests that it's related to the driver version. Has anybody been able to demonstrate that keeping the same WU and the same driver version but rolling back to V7.1.xx fixes ANYTHING that can be attributed the version of FAHClient? . . . or is the Client version just something that coincidentally changed about the same time so it SEEMS like it's related.

Note that the drivers are ALWAYS different between 32-bit and 64-bit windows.

[chaosdsm]

The previous post strongly (as well as the title) strongly suggests that it's related to the driver version.

I don't know who added the part about drivers to the title... I strongly dissagree that it's driver related unless it is something that has existed within the Fermi driver code since at least the 275.33 WHQL release, which is the oldest version I've tested so far.

When it comes to Folding at Home on my GTX460, other than tiny performance differences (under 0.25%) none of the old or new, beta or WHQL drivers changes the behaviour, though I haven't checked the monitor sleep / power off thingy yet under 301.24 beta. However the newest beta's do allow me to overclock the card further, but that doesn't change the issue at hand. This tends to point towards a WU issue, or possibly a previously unrecognized hardware issue common to the Fermi architecture. Yet, it does not leave out the possiblity of drivers, because it could be something that has existed in the drivers since the first Fermi supporting driver was released. But it's definitely not enough (IMO) for someone to come in & edit a thread title to read (apparently drivers)...

I just wish I had more time & hardware at my disposal so I could conduct a thorough scientific investigation across multiple GPU's (430's 450's 460's 470's 480's, 550's, 560's, 580's, etc...) & Client versions. Although, I may be willing to donate my 460GTX relatively soon for testing, since I seem to have one that doesn't have any of the other issues that most people are having with them.

[bollix47]

FWIW

I have 3 GPUs (460,480,560) and all of them needed a fan speed increase when the 803x WUs appeared on the scene. The 460 was particularly sensitive to heat and would EUE if the temperature went over 65 degrees. Since using MSIAfterburner to increase the fans' speed and adding addition cooling for the 480 no WUs have failed.

All three are using 285.62 drivers and running on v7.1.52.

The 560 is running on my daily use computer(Q6600 @ stock) and I see no lag using Vista 32 and an aero theme.

[chaosdsm]

FWIW

I have 3 GPUs (460,480,560) and all of them needed a fan speed increase when the 803x WUs appeared on the scene. The 460 was particularly sensitive to heat and would EUE if the temperature went over 65 degrees. Since using MSIAfterburner to increase the fans' speed and adding addition cooling for the 480 no WUs have failed.

All three are using 285.62 drivers and running on v7.1.52.

The 560 is running on my daily use computer(Q6600 @ stock) and I see no lag using Vista 32 and an aero theme.

Hm... the 560 has 336 shaders, same as the 460, assuming that you are of course talking about the original 560 & not any of the TI's or the OEM (came with PC) version which can have between 352 & 448 shaders. Assuming all that, then shaders are of equal quantity with different results, so that could mean hardware problem on my end, problem between Windows 7 Pro 64bit, GPU, & FAH, &/or driver issue...

If however you had an OEM version or one of the TI cards, then those all have more shaders than the 460 and could point towards 336 shaders not being enough to effectively run a 803X work unit meaning the work unit itself might be the cause of the trouble, but would need more supporting evidence.

I'll go back yet again to the 285.62 drivers tomorrow & see what happens. Who knows, maybe last time around I missed a step in the uninstall / delete of previous drivers. Maybe I'll also throw one of my 430GT's with its 96 shaders into the mix & see what happens

[bollix47]

My 560 is indeed a Ti and has 384 shaders. The 803x WUs ran hot and I increased the fan but only to 65%

460 - 336 shaders ... runs the 803x fine but as I said above I did have to increase the fan speed to 80%

480 - 480 shaders ... the 480 has always run hot since I got it but the 803x WUs were too much for it even with the fan @ 100% so I took the side off and have a room fan blowing on it ... np since doing that ... I did order and receive a Zalman cooler for it but haven't installed it yet

They all failed prior to increasing the fan speed.

[bruce]

They all failed prior to increasing the fan speed.

There seem to be a number of problems with the fan profile not increasing the fan speed enough when the GPU is used heavily.