ATI 5770 - Display driver amdkmdap stopped responding and has successfully recovered

[Dancer]

Ok I've been having the "Display driver amdkmdap stopped responding and has successfully recovered" error since day one of my brand new system running Windows 7 (64bit, fully up-to-date). So for about a month now with a Radeon 5770, AMD PIIx4 955, Gigabyte - GA-MA770T-UD3, 4GB memory.

I've had to disable aero as that usually triggered it once or twice a day. Playing any game (crysis, bioshock 2, assassins creed) triggers it randomly after anything from 5 seconds to two hours. When it happens audio keeps going, but the screen freezes (sometimes turns black) and I have to end task and when back to my desktop the error is displayed.

I've done lots of research, but despite plenty of people having this problem I've not found a solution that works. I've made sure to remove old drivers (in safemode) and am using the latest (Catalyst 10.3).

I'm thinking two things that may have an impact and would like to hear your opinions.
1) My 500w True Blue II PSU only has a 4 pin 12v plug and the gigabyte motherboard takes an 8 pin. Everything appears to be working fine, but could this be part of the issue?
2) To keep the case nice and cool I've reduced the voltage one step (I'm not overclocking). The default voltage in the bios was a bit high (1.4v) so I reduced it to 1.375v. Again could this be the issue?

 

[Dancer]

I had a sort of similar problem lately but fixed it by adjusting the RAM voltages to what it says on the sticker of the RAM sticks.

My mobo was undervolting massively and caused plenty of issues

Let us know how that works out.

Thanks for the tip, I'll get myself a new corsair 650W PSU to eliminate the 4 pin 12v as the cause and check the recommended voltage of the RAM at the same time.

@billysielu: Exactly, although what is sufficient CPU... There are plenty of people undervolting far more than I do and people overclocking with less than 1.4v for this cpu. I've also seen plenty of people reporting that using a 4 pin for the 8 pin 12v should be ok in most cases, but I'll admit it gets much more uncertain with quad cores which is why I asked. Just trying to avoid spending even more money on this build only to find that the PSU was not the problem to begin with...

Edit: @Essexraptor: Thanks for taking the time. I'll get the TX650W corsair since that is on offer and seems like a good psu witha 2/4 pin 12v. I'll let you all know how that works out.

 

[Dancer]

Update: New PSU, no undervolting, but same "Display driver amdkmdap stopped responding and has successfully recovered" and £70 poorer. Not too happy.

 

[Dancer]

i had a similar issue but with a 5870 and the problem was due to a Windows Microsoft Update can't remember which on but if you google it. Ever since uninstalling it i havn't had this problem

Ok, found this (is it you?):
"...I uninstalled this patch from Windows 7 and things seem stable for me. I had a 4870 and thought the problem was the card, I then bought a 5870 and had the same problem until I uninstalled this...

http://support.microsoft.com/KB/976264"

So I've uninstalled this and we'll see where that gets me.

 

[Dancer]

Result: Made absolutely no difference.

 

[Dancer]

Despite trying several more things the issue remains. Having done numerous searches on the issue I see that it is a common issue in Windows 7 (and vista) with ATI cards and although some people say the errors stopped after various tweaks none of those have worked for me. It seems that this driver error can be an indicator of a great range of problems from faulty gpu, bad driver install, wrong ram voltage, or combination(?) of some windows updates + hardware., etc.

What have I tried since my last post:

1. Re-installed 1.3 official catalyst driver. Yes skippy2421 I tried driver sweeper and re-installed in safe mode from 1.2 and 1.2 also had this problem, although perhaps less frequent.

2. Keeping an eye on CPU and system temps using CoreTemp (will try to download gpu-z or everest to check gpu temp, but fan is running and card does not seem especially hot)

3. Checked voltage for RAM in bios and was 1.5v as recommended for my Geil Value Ram.

4. Set windows 7 to High Performance. I will try to disable cool'n quiet and instead set the multiplier to be underclocked just to eliminate this as a cause. The multiplier goes to x16 (3.2 Ghz) so if I set it to x12 as that should be well under spec.

5. Checked all power cables well connected (I got a whole new PSU and made no difference).

6. Installed AVG virus scanner and did a full scan (all clean). I run zone alarm as firewall.

Any other suggestions for what I can try as I'm starting to think that windows 7/vista with ATI gpus must be fundamentally flawed (my wife is using a 8800GT on a nearly identical system and has had no problems at all!).

 

[Dancer]

I've spent some time over on the AMD forums and there are several long threads with people trying to sort out the "amdkmdap" display driver problem with ATI cards from 4xxx series to the latest 58xx cards. A problem which seems to have been going on since Windows Vista. I've learned more about my gpu than ever before. I've not yet solved it, but some early results are interesting after my last changes even if bioshock2 still crashes...

I've installed gpu-z to and set it to logging and I can say that temperatures never go above 67C. 69C with PCIe slightly overvolted from 1.8v to 1.85v which by the way made no difference. The GPU idles around 30+ C same as the AMD CPU (which never goes above 59C by the way and usually stays in lower 50s celsisus when running games).

I've tried removing one stick of RAM and moving it to a different slot. I've tried moving both sticks of RAM. I've turned off Cool'n Quiet. I've tried setting the GPU fan to manually run at 70% (despite not having heat issues) and I've tried undercloking from GPU Core clock of 850 Mhz to 600 MHz. I did try to play around with the ATI tray to lower the GPU memory clock, but I've not figured out to make those settings take as I can change the sliders and hit apply but it seems to be ignore and if I close the tray and go back it has reset...

Anyway, none of the above has made any difference. Disabling Aero and using just windows 7 basic has not solved the problem either, although subjectively it seems like it crashes sooner when aero is enabled...

I did find a post in AMD forum with someone saying that plugging your PC into a surge protector could "starve" your PSU of enough power and therefore cause the dreaded amdkmdap display driver error. This seemed unlikely to me, especially as I'm using a fairly expensive belkin surge protector designed for use with computers. Anyway, having nothing to lose at this stage I tried plugging my PC into the wall socket. This was fairly late last night so I've only done some brief testing. Bioshock 2 still crashes, but I was able to play longer than I usually can (over half-an-hour!) and when it did crash there was no report about amdkmdap in the event log, but just an entry that Bioshock2 stopped responding. I also tried running Crysis running through the GPU demanding frozen landscape where it has crashed _sometimes_ in the past and it didn't crash.

I've not had time yet to do further tests and this is too early to say anything. This could all just be a fluke or it might be that my system (as opposed to bioshock 2) is now more stable after plugging into the wall socket. Will test further tonight and update.

Sorry for the long post, but wanted to make it a fairly complete log of what I did. To recap I'm still running windows 7 in basic with no aero. I'm running cpu/gpu at stock values and all voltages at bios defaults with cool and quiet disabled and win7 set to high performance. Voltages are: RAM 1.5v, CPU 1.4V and PCIe 1.8v.

 

[Dancer]

Let's try overclocking your card. Obvoiusly let's not go too far and see if these errors become more frequent.

I've tried underclocking the gpu (not the gpu memory though as this is more tricky) and it made no difference. I haven't tried overclocking, but since the crashes can happen anything from 5 seconds to 10 minutes of playing I don't see how I would be able to tell if it crashes more often when overclocked. Also, from AMD forums (where many people did RMA or upgrade their cards and the replacements had the same problem) and checking my GPU temps I don't think that is the problem, but thanks for the suggestion.

Next long shot try running with no security stuff at all and minimal programs

I've not got much installed and the problem started before I installed AVG. I've tried shutting down zonealarm, but that did not make any difference.

Out of curiosity what drivers have you tried?

I've tried 10.2, 10.3 beta and 10.3 release. From AMD forum it seems this problem occurs with all drivers, although one person reported success downgrading to 8.1. However, 8.1 is a vista driver and he had to do some fiddling to get it to work on windows 7. Crazy to downgrade this far back, but hey if it works that's what I'll do.

 

[Dancer]

I've come up with a temporary solution that seems to work fairly well. I say temporary because underclocking my card and turning off all power saving features / noise reduction features is not really acceptable. I really hope the catalyst drivers sort out this problem, because I'm quite disappointed by the card's apparent instability.

I have yet to zero in on which of these things has the biggest impact and I need to do more testing although I've tested for a couple of hours yesterday and no crashes.

1. In my mobo bios I've set all voltages manually to prevent the auto control over or under volting. I also disable Cool'n Quiet and run windows 7 in High performance so the CPU is always stepped to x16 @ 3.2 Ghz.

2. I'm still keeping my computer plugged into the wall socket bypassing the surge protector, but I don't think this is making a difference).

3. I'm keeping Windows 7 aero turned off as I did try turning it back on yesterday and shortly after starting Bioshock 2 it froze again. (requires further investigation).

4. Using the instructions here I've used the CCC profile handling together with the overdrive to underclock my GPU to 600 Mhz GPU and 1000 Mhz memory (from 850/1200). Potentially more importantly I've set all three steps to be the same, so that the GPU will ALWAYS run at 600/1000 and I also set the fan to always run at 75% and the CoreVoltageTarget to always be the maximum (I forget the value). This effectively disables the stepping of the GPU.

5. I also researched the driver error at microsoft developer central and found this article Timeout Detection and Recovery of GPUs through WDDM. The display driver error is basically a timeout detection which triggers after 2 seconds. Creating and setting the TdrDelay value in the registry to be 30 seconds rather than the usual 2 I'm trying to eliminate the timeout detection as the cause of the problem. Interestingly after changing this value it has only frozen once when turning on aero and then running bioshock 2 and as expected the game remained frozen for 30 seconds before the screen turned grey(ish) as before.

Conclusion: Bioshock 2 used to crash every few minutes with the above it now only crashes when I also turn on aero. Plan going forward is to find out which of the above are not needed, test lots more and figure out why it seems to crash much faster when aero is enabled (could be a coincidence, further testing should confirm).

Any suggestions or feedback is welcome.

 

[Dancer]

My 5770 has been stable with no more errors after I set the clock values and voltage as below using the CCC profile.

Clock: 40000 / 85000 / 85000
Memory: 60000 / 12000 / 12000
Voltages: 1000 / 1010 / 1010

The idle values (40000 / 6000) can probably go lower as reported by fellow 5770 owner. I don't see much point in having step 2 at all... In my thinking I either want the card to be idling or going full throttle, hence why step 2 and 3 is identical.

So in conclusion increasing idle and slightly increasing voltage seems to make the card stable.

 

[Dancer]

These steps are from the AMD forums. The values below are suggested by others (I use slightly higher clock speeds), but others have reported using much higher voltages than me. Keep an eye on your gpu temps as well (for example by using gpu-z) and if it gets very hot try setting the fan to manual in the profile and set the fan speed to somewhere between 50-90% (whichever keeps it nice a cool). What is too hot is open to debate. My card is typically in the 50s and low 60s when in use, but 70s are reported as fine by some (check official documents for your gpu heat tolerance).

1. Open CCC
2. Unlock and Enable Overdrive if not already done.
3. Go to Options/Profiles/Profiles Manager. Create a new profile. Under composition make sure “ATI Overdrive” is checked. Save and Close, DO NOT ACTIVATE.
4. Go to: C:\Users\{yourusername}\AppData\Local\ATI\ACE\Profiles (you will need to have “show hidden files” turned on for this)
5. Open the xml document with the name of the profile you just created (notepad is fine)
6. Change the values of the Clock and Memory speeds to look like this (these specific values are what worked for me and my card, use judgment) EDIT ONLY THE BOLD VALUES.

<Feature name="CoreClockTarget_0">
<Property name="Want_0" value="40000" />
<Property name="Want_1" value="60000" />
<Property name="Want_2" value="85000" />
</Feature>
<Feature name="MemoryClockTarget_0">
<Property name="Want_0" value="90000" />
<Property name="Want_1" value="90000" />
<Property name="Want_2" value="120000" />
</Feature>
<Feature name="CoreVoltageTarget_0">
<Property name="Want_0" value="1000" />
<Property name="Want_1" value="1010" />
<Property name="Want_2" value="1010" />
</Feature>

7. Save and close. Go back to CCC and activate the profile you just created.

Good luck.

 

[Dancer]

A quick couple of questions, if you do not use CCC, just the drivers themselves, does the card still downclock itself in 2d (and if so to what value) ?
Would the above method of changing the xml be able to overide CCC's overclocking limit ?

I've never tried just the drivers without CCC installed, but I suspect it still downclocks to save energy and reduce noise and heat. I think the rates would be determined in the driver and the profile just overrides these values. The best way to check is to run something like gpu-z and check what the clock and memory rates are set to. You can turn on logging to see if they change while gaming.

I don't know if it is possible to go past the CCC overclocking limit, and it's not something I would try as you might risk damaging the card.

 

[Dancer]

Nice write up Dancer, i assume you've had no more problems since this fix. I may explore this avenue myself as i am currently the victim of the overclocking and dual monitor screen flicker issue. As the issue is caused by the 2d clocks dropping too low for 2 monitors this may be a viable option to get around this if the new beta drivers don't work for me.

Your assumption is correct, so far so good.

 

[Dancer]

I got this from the AMD forums..

"Manually forcing the fan to run at 60% pretty much took care of my problem as well. However, as I mentioned above, I don't think it's because the driver isn't auto-adjusting the fan speed. I believe it is, but it's adjusting it to keep the GPU chip cool, which is where the card's temperature sensor is. I think the problem is that other chips on the board are getting hotter than the GPU (or hot enough to become marginal), but the fan doesn't kick in yet per auto-adjustment because the GPU chip's temperature isn't hot enough yet. By forcing the fan to 50% or 60% all the time, rather than peg it to the GPU temperature, ALL the parts on the board are getting cooled all the time, including aux or memory chips that are more temperature sensitive than the GPU itself."

I have updated my bios the latest version. That helped a great deal with only two crashes since then. Lately I've also manually adjusted my fan to 50%, crashes are very few. However this remains to be seen...still testing.

By the way I have a single 19" monitor at 1280x1024

I need to do more testing before I can declare my system 100% stable, but since I changed the idle (aka 2d) clock speed and voltages in the CCC profile I have had zero crashes/freezes. I really wish there was a way to set the fan speed to match the clock speeds in the profile so that I could have a "quiet" fan speed when being idle and "gaming" fan speed otherwise. I use a fan speed of 55% which keeps the card nice and cool and is much quieter than 60%.

 

[Dancer]

I hasn't worked. It was most of the night, then I had a major crash and my fans went wild. This was without using your ccc profile method.
I will go back and try that again. I had notice the GPU was at permantently 100% activity. Can that be right or have I set that profile up wrong? I got the info from GPU-Z.

I can only tell you about my own experiences as I'm not an expert on ATI or gpus, however, it should not show 100% activity all the time. I wasn't looking too closely at gpu activity, but I'm pretty sure it was fairly low most of the time and hardly ever went to 100%. Also, what is a "major" crash as opposed to the "usual" crash? What were you doing when it crashed? What does the windows event viewer indicate happened at that time?

So if you haven't used the ccc profile method what did you do?

You can try posting a screen of your gpu-z page showing the clock speeds and gpu acitivity and maybe someone will spot something, but was the gpu at 100% before you updated the bios? Which bios did you use? Was it one from your card manufacturer? You might want to try to contact the card manufacturer if you cannot find answers online.

 

[Dancer]

hi

thanks for the fix info. I am going to try this as I have had lots of display not responding errors on my 2x HD5770's

I wanted to ask, do I need to select the new profile manually each time I go into windows, or will it be loaded automatically when windows starts?
Also must ATI overdrive be enabled? ( I disabled this to be on the safe side)
cheers

Once you've activated the profile it should load automatically every time you reboot.

I have overdrive enabled, but then I'm also increasing the fan speed. I'm not sure if you can leave it off, but if you wish to try you can test that easily enough. Just create your new profile, activate it and use something like gpu-z to see your speeds, voltages and fan speed.