Deferred Procedure Calls killing my 6800GS

I'm having difficulties with my new 6800GS graphics card and Deferred Procedure Calls (DPCs). I won't take you through the painful process of finding out what the problem was and which component was causing it.

System specs
A64 3000+ (not overclocked)
K8v basic motherboard
1gb Corsair PC3200 RAM
2 SATA Seagate Baracuda HDs (1x140gb, 1x160gb; no RAID)
1 PATA 13gb dirve
1 DVD re-writer, 1 DVD drive
XP SP2 (originally fully updated, now bare)
6800GS card (originally running 91.36)


Background
I installed the card a few days ago. On first install I encountered some very odd slow-down which I will describe below. When I started my system everything would be fine, then over time the system would slow down. Opening the Task Manager when the slowdown started showed that I was getting 95-100% 'system idle process' (s.i.p.). Now please do note that I'm more than well aware that s.i.p. is a placeholder for the CPU and 100% should usually be healthy, but usually the graphical cpu display would show 0% cpu load if s.i.p. was at 100%. It wasn't - the cpu was choked full at 100%. On further investigation using a program called Process Explorer I discovered that actually, my s.i.p. was at <1% and I had 99% DPCs. Windows Task Manager lumps these in with s.i.p. for some reason so my unhealthy system looked healthy.

The problem in detail
So, what do I mean by slowdown? Well if I describe what it's like at 95% DPCs it will give you an idea - if I move a window around the screen it will leave a trail for several seconds until the screen updates; the mouse pointer jumps centimters at a time rather than being nice and smooth; switching tabs in a control panel can take up to three minutes to register and alter the display; programs like Notepad take five or six minutes to load up. That probably gives you the idea - in effect having 95% DPCs means that there's only 5% of CPU time for everything else, including those processes that always take up time. New programs get <1% CPU time. I can't close programs, I can't switch programs and I can't shut down, I usually have to turn off the computer at the mains.

At 75% DPCs I can use the internet and Word, but if I try to use Windows Media Player or load up a game I'll get 1FPS or less and DPCs will leap to 95%.

On first startup one of two things would happen. The first is that DPCs would start at 80% and I couldn't do anything - after two minutes of using firefox they'd be at 95% and I'd get the above problems. The other things is that the machine would start as usual with <1% DPCs and the rest of the CPU time being used on programs/s.i.p. as usual. However, when this happened if I ran WMP DPCs would immediately leapt to 50% (although I could still play the file) and then rise there after until they hit 100%. If I just used the internet or left windows doing nothing then it would sit at 0% DPCs for hours, but at some point they'd begin to rise over time until I return to the machine to find that they're at 95% again. Tellingly, if I ran a virus checker and watched the DPCs, you could see them rise over time. They'd cycle 0%-5%-0%-5%-2%-8%-1%-10%-2%-8%-1-%-12%, etc. slowely rising. It would take five minutes and thrity-three second of running the virus scanner to go from 0% DPCs to 95%.

Steps taken
My first thought was virus/trojan/malware or driver conflict, so I swept the system with six different products and cleared out the few cookies that are always present. Nothing odd found, no change. I cleaned out the drivers rebooted and installed various different drivers. No change.

A hardware conflict then? Nothing untoward in the Device Manager. Perhaps it's a trojan that hasn't been picked up by any of the virus/spyware products? Or it could be hardware-related. I have recently installed a new wireless mouse/keyboard and a graphics card. Replacing the former had no effect. Replacing my 3D card with my old Ti4400 stopped the problem totally.

Oddly, when I re-installed the new card to see what would happen I didn't get the problem back. At least I didn't seem to. I could run WMP without the ominous rise in DPCs. I then formatted and re-installing windows (partly because of these issue, partly because I wanted a nice clean system for my new card). I used an old copy of XP SP2 with no integrated hotfixes or unattended programs/settings.

New problem
When I finished the install I added some codecs, hooked up my other drives (I'd disabled them for the install) and opened WMP. Same problem as before. Right, well uninstall the new card, put in the old one. Problem fixed. Uninstall the old card, re-install the new one. Things were fine for two or three boots and I was able to run 3D Mark 2005 to test my new rig. After three boots...

Different problem. Now windows does one of two things: It either starts up with 85% DPCs and takes ten minutes to load windows, then the DPCs drop back to 0% or it starts at 0% and the DPCs rise when I run WMP/screensaver/game. When I quit the game the DPCs slowely drop back to 0% over about an hour. As I write this they've dropped from 80% (watch the keyboard, it misses keystrokes) to 40% (seemingly normal).


So, apologies for the long post. Anyone ever seen this before? Where do I go from here? I've not tried uninstalling/re-installing the card again which may or may not have fixed the problem last time as this is clearly something deeper rooted. Help!

Many thanks in advance,
-Ben

 

Here's a snapshot from Process Explorer to clarify things:

http://img0633.paintedover.com/uploads/0633/dpcs.png

I was sitting at circa 30% DPCs with it cycling sowly back to 0%. A patch finished downloading, so I installed it. It was quite big (Company of Heroes latest beta ptach, 791mb) and you can see when I load it from the I/O graph. You can hover the pointer over the history to show the main CPU usage. The purple shading shows a brief time where the beta patch took more CPU cycles than DPCs. Outsude if that area there's a majority of DPCs (with the patch taking fewer cycles).

You can also see that DPCs increased as I applied the patch and stayed higher afterwards. Oddly, after I finished applying the patch, my DPCs dropped to nothing and I'm now back to 3% with a seemingly healthy system.

Edit: Odder still, I'm now playing a high definition file in WMP and the DPC are cycling between 3%-9%-7%-13% without seeming to rise (they've been stable for six minutes). It's 50% s.i.p., 10% DPCs, 40% WMP on average. Go figure. I'm not sure this does anything other than make me more confused, but I'm going to go and make some late lunch whilst I leave it playing the background and we'll see what happens.

http://img0633.paintedover.com/uploads/thumbs/0633/pedpcs.png

 

Pretty slow CPU...........maybe that's why.....

 

It was working fine with my Ti4400 and I don't see why it should make a diffence to the card. I appreciate that I've not got the fastest CPU, but it seems fine for most things - I'm playing Company of Heroes (which is unoptimised beta) on the highest graphical settings in a sensible resolution and it looks great and runs perfectly (assuming the DPC issue isn't happening). Surely DPCs from the CPU to the 3D card would be more about transfer bandwidth than CPU speed? Why would it work perfectly sometimes and not others.

Anyway, following on from the above, the machine played WMP HD files for an hour without any DPCs (well, a few, but nothing abnormal). I then tried it playing a game (CoH beta) and it performed fantastically. I assume therefore that this is now an intermittent issue.

 

Bump & update: The problem is still as before.

 

Problems solved - it's a hardware fault connected with the GPU temperature. At 47°C I get 0% DPCs, at 49°C I get 100% DPCs and above 50% I'm back to 0% DPCs.

RATTV3 pins the problem to a driver, videoprt.sys, creating all the DPCs. It's not a dirver related fault though, I've found hundreds of threads of people with the same problem and it always goes away with temperature related solutions (stop the fan, run a 3D intensive program, etc.). Some reports suggest that it's a fan voltage issue and that stopping the fan turning is a solution regardless of GPU temperature. The only long term solution there is to provide a different power supply for the GPU fan and I'm not prepared to mess around like that for a fault on the card...

48-49 degrees:
http://img.waffleimages.com/img/bf22c6614a953bf89c7b699dc592eaa729211300/48-49 degrees.PNG#via=salr

50 degrees:
http://img.waffleimages.com/img/88802f566d73812c876fd1cdc664c71e862f0b54/49 degrees.PNG#via=salr

51-52 degrees:
http://img.waffleimages.com/img/fb70bccaf1eb91dc48c340c3bf5ea778b02bc96b/51-52 degrees.PNG#via=salr