HD6950 temperature/crash issue

[Bacterius]

Hello,
I am having a recurring problem with my HD 6950 (2gb sapphire) since I bought it a few days ago. The problem is, when the temperature of my GPU falls under 50*C while doing 3D stuff (games for instance), the driver will hang (then TDR will kick in and reset everything, killing the game in the process of course).

I can reproduce it easily... I can run a benchmark in the background to heat up the GPU for a few minutes (get it to 60-70C), then if I close the benchmark and start up the game, it works for a few minutes, then crashes as the temperature falls below 50*C again (unless I run a really intensive game that can sustain 50C+ temperatures on its own like Crysis or such). If I leave the benchmark running while playing, it will never crash (although it stutters a lot so it's not a fix).

I am suspecting a driver issue and it seems some other people have had similar issues with their 6xxx cards, but apparently there is no clear-cut fix on the internet. I tried all drivers between 11.1 and 11.10, same issue, always.

Does anyone know anything about this and how to fix it? The card is not overclocked, it is not flashed, and the fan is set to run at 50% constantly (keeps the temps below 70C always). The CPU (i5 2500k) is not overclocked and is not overheating. The RAM has been tested with memtest86+, no issues. I also tested the VRAM using memtestCL.. no issues at all after like 200 passes over all 2gb.

So I don't know, I think it's pretty clear that my card is not damaged, that it's not a power problem (psu issue wouldn't explain the symptoms at all), etc... so I'm left with either a software bug or some sort of motherboard issue. And I'm not too sure what to do now.

Thanks for any advice or info on this issue - please ask for more information if you need, I'll do anything to fix this.

System specs:
- i5 2500k @ stock speeds
- Radeon HD6950 (2gb) - Sapphire
- 2x2gb G-SKILL Ripjaws (1333MHz, 9-9-9-24)
- P67 LE (B3 rev) ASUS motherboard - (one PCI-E x16 lane)
- 2x500GB Seagate 7200RPM hard drives
- 600W Thermaltake litepower PSU
- mid-tower atx case, exhaust fans, dvd drive, etc...

 

[Bacterius]

Btw this also happens when I'm developing my own 3d applications... when the GPU temperature is between 30 and 40C, the card will artifact like crazy, between 40 and 50C there's alot less artifacts but still some, and above 50C it works fine. I think games crash because they detect the artifacts and tell the driver to commit suicide to prevent damage or something.

This is kind of an unexpected behavior... I thought cards artifacted when they were running hot, not cold

 

[grumperfish]

It could be power changes from ATI Powerplay causing the artifacts (where the card clocks down due to a change in GPU load).

You can try setting a manual GPU voltage of 1.1V via MSI Afterburner and save it as a profile to be enabled at boot. This will make sure it's not undervolting the card too far when it changes power states.


However, your powersupply is pretty mediocre, so the change in power usage could be causing it to fail. Its 12V rail ratings are almost lower than a Coolermaster Extreme Power 500W unit (not a good unit btw), so if the PSU you have can't keep up with load changes you would have rapid undervolting on the card causing artifacts.

If setting a manual voltage in Afterburner doesn't fix the issue, you may want to consider getting a better power supply. An Antec Earthwatts 500W or 650W unit would be fine for your system, and they're currently the same price on Amazon.

 

[Bacterius]

It could be power changes from ATI Powerplay causing the artifacts (where the card clocks down due to a change in GPU load).

You can try setting a manual GPU voltage of 1.1V via MSI Afterburner and save it as a profile to be enabled at boot. This will make sure it's not undervolting the card too far when it changes power states.


However, your powersupply is pretty mediocre, so the change in power usage could be causing it to fail. Its 12V rail ratings are almost lower than a Coolermaster Extreme Power 500W unit (not a good unit btw), so if the PSU you have can't keep up with load changes you would have rapid undervolting on the card causing artifacts.

If setting a manual voltage in Afterburner doesn't fix the issue, you may want to consider getting a better power supply. An Antec Earthwatts 500W or 650W unit would be fine for your system, and they're currently the same price on Amazon.

I'll try locking the voltage to 1.1V in Afterburner. Wouldn't I need to lock the clocks speeds aswell, since the card has three power plans, 250MHz, 500MHz and 800Mhz? Or is it fine to let the card get a constant 1.1V regardless of whether its clock speeds are in idle or 3D mode?

I didn't see the card changing clock speeds when artifacting though - it remains at 800Mhz/1.1V whenever a 3D application is open.

Gonna give MSI Afterburner a shot

 

[grumperfish]

I'll try locking the voltage to 1.1V in Afterburner. Wouldn't I need to lock the clocks speeds aswell, since the card has three power plans, 250MHz, 500MHz and 800Mhz? Or is it fine to let the card get a constant 1.1V regardless of whether its clock speeds are in idle or 3D mode?

I didn't see the card changing clock speeds when artifacting though - it remains at 800Mhz/1.1V whenever a 3D application is open.

Gonna give MSI Afterburner a shot

I run a constant 1.1V to mine. Although when I unlocked the shaders on my 6950 I flashed a BIOS that only has 2 power states (no 200mhz power state), so I see shifting only from 500 to 880mhz.

 

[Bacterius]

I run a constant 1.1V to mine. Although when I unlocked the shaders on my 6950 I flashed a BIOS that only has 2 power states (no 200mhz power state), so I see shifting only from 500 to 880mhz.

Okay. I think the voltage lock worked, GPU-Z is reporting 1.1V even in idle mode (didn't dare enable voltage monitoring on Afterburner because of the big red warning lol)

Gonna try running a benchmark, see if its stable, then I'll try to reproduce.

 

[Bacterius]

Ok it didn't fix it. The voltage is now constant but the core clock seems to fluctuate like crazy when running Furmark (or any 3d app). I can remove that behavior by using CCC to set the "power control" slider to 20%, but if CCC is installed then I cannot lock the voltage via Afterburner because CCC overrides it. And the "unofficial overclocking" feature in Afterburner that would fix that apparently only works with MSI-brand cards, while my card is a Sapphire.

And the card still crashes/artifacts. Here is a screenshot of Afterburner:

 

[grumperfish]

Ok it didn't fix it. The voltage is now constant but the core clock seems to fluctuate like crazy when running Furmark (or any 3d app). I can remove that behavior by using CCC to set the "power control" slider to 20%, but if CCC is installed then I cannot lock the voltage via Afterburner because CCC overrides it. And the "unofficial overclocking" feature in Afterburner that would fix that apparently only works with MSI-brand cards, while my card is a Sapphire.

I have an XFX card and use MSI Afterburner exclusively, without CCC, so this isn't true FYI.

Wow, yeah, looking at the graph, something weird is going on. Do you have any dual-screen utilities active (ie: ultramon)?

Do you have a reference card or a recent revision? What's the exact model # of your 6950 and where/when did you buy it? If it's reference, it might be worth it to flash the BIOS to a version with different power states. List your card and any relevant information before considering this, and don't outright attempt this first.

Check google for how to disable ULPS as well. Technically it's for crossfire, but it might help as it applies to power states.


Additionally, install the latest beta of HWiNFO64 & click "show sensors" when it's starting. Run furmark or Kombuster (via Afterburner, click the "K" button on the left side of the window) and post a screenshot of HWiNFO (including the GPU/CPU/board temps & voltages while it's running. I want to see if the PSU is giving inadequate voltages somewhere.



Your clock speed changes in Afterburner graphs should look similar to this while running Kombuster (temperatures, fanspeed, and clocks will be different of course - running a different BIOS & custom cooling):

 

[Bacterius]

Wow, yeah, looking at the graph, something weird is going on. Do you have any dual-screen utilities active (ie: ultramon)?

No, I don't have any dual monitor utilities active (I have only one monitor and I'm pretty diligent about only installing software I need, so pretty sure there's nothing like ultramon or screenpanels on my computer).

Do you have a reference card or a recent revision? What's the exact model # of your 6950 and where/when did you buy it? If it's reference, it might be worth it to flash the BIOS to a version with different power states. List your card and any relevant information before considering this, and don't outright attempt this first.

I'm not sure, I have a "Sapphire HD6950 2GB GDDR5 PCI-e" card. It didn't have any special stuff on it, it has a single fan and a Sapphire logo on it, it's not a Dirt3 or whatnot edition, not sure what you mean by reference. I bought it back in May or so but it didn't get unboxed until a couple weeks ago. Bought in New Zealand online (Paradigm PC's)

Check google for how to disable ULPS as well. Technically it's for crossfire, but it might help as it applies to power states.

Ok I'll have a look after getting the screenshot.

Additionally, install the latest beta of HWiNFO64 & click "show sensors" when it's starting. Run furmark or Kombuster (via Afterburner, click the "K" button on the left side of the window) and post a screenshot of HWiNFO (including the GPU/CPU/board temps & voltages while it's running. I want to see if the PSU is giving inadequate voltages somewhere.

Here you go (this is without disabling ULPS, and I had to hide a few sensors like hard drive, and some others to make everything fit on the screen, but I kept the info you wanted):


PS: I didn't really bother making all the AUX fan connections for the case fans, so they're all spinning at max speed (pretty silent so hey why not).

PS2: why does your card say "AMD Radeon HD6950 Series" while mine says "AMD Radeon HD 6000 Series" (even tho I have a 6950)? Is it up to the vendor to put the description in? Just wondering

 

[grumperfish]

Here you go (this is without disabling ULPS, and I had to hide a few sensors like hard drive, and some others to make everything fit on the screen, but I kept the info you wanted):

If HWiNFO64 is accurate (and it usually is) your 12V rating at load (11.8V) is way too low. I'm not positive that it's the source of your issues, but it cannot be helping things.

I would suggest either borrowing a known-good power supply if possible, or if not then getting a better unit ASAP. If the PSU is struggling to provide the CPU with enough power at load, it will have less to give to the GPU causing erratic performance (as you've seen so far). Without a more stable PSU, it's impossible really to figure out the root of your issues. It's highly-likely that simply replacing the PSU will stabilize performance.


Your other voltages aren't terrible, and the 12V rail is in-spec, but it's stressed well below where it should be. It should ideally be 12V even or a bit higher to supply adequate voltage to your PC components.

PS: I didn't really bother making all the AUX fan connections for the case fans, so they're all spinning at max speed (pretty silent so hey why not).

Yeah I've hidden a few things as well like the PCH lines and the SMART data for my SSD, as it always reports 30C no matter what, so, eh, wasted space

PS2: why does your card say "AMD Radeon HD6950 Series" while mine says "AMD Radeon HD 6000 Series" (even tho I have a 6950)? Is it up to the vendor to put the description in? Just wondering

We're using different apps for testing. Furmark & Kombuster are similar, but they probably report the card data differently. Run Kombuster and see if you still have the same issues you're experiencing.


The reason I asked about the card's model number as if it's reference, you may be able to flash to another Sapphire reference BIOS (from techpowerup's BIOS database) that has dual clock changes, and not get hit with the excessive clock changes. Knowing the model # would help ID whether that's a possibility to look into.


If replacing the PSU doesn't fix the performance issues (try this first, and really, you need to do it anyways if you want your system to last for awhile), the next step I'd suggest would be to uninstall the CCC and just try using afterburner by itself to make sure there are no conflicts between them.

Beyond that you can use the unofficial overclocking feature in afterburner to disable clock changes and keep the card at 800mhz for testing (I forget what the setting is; it's a number value [0,1,2, or 3] but there's guides on google). If it doesn't have the same issues while stabilized at one clock, it's definitely an issue with Powerplay so we'd want to investigate a different BIOS with dual-clock changes instead.

If none of this works, and the card's under warranty, you'd need to contact Sapphire for an RMA, but we should troubleshoot a bit before that step.

 

[Bacterius]

If HWiNFO64 is accurate (and it usually is) your 12V rating at load (11.8V) is way too low. I'm not positive that it's the source of your issues, but it cannot be helping things.

But then how come it doesn't artifact or crash on Furmark or MSI Kombustor (which works well, btw, tried clicking on the "K", nifty) and works perfectly on most games/apps, and just randomly commits suicide only in certain specific (and reproducible) situations? Sorry if I seem kind of hard-headed, it just seems counter-intuitive.

I would suggest either borrowing a known-good power supply if possible, or if not then getting a better unit ASAP. If the PSU is struggling to provide the CPU with enough power at load, it will have less to give to the GPU causing erratic performance (as you've seen so far). Without a more stable PSU, it's impossible really to figure out the root of your issues. It's highly-likely that simply replacing the PSU will stabilize performance.

Okay... I was hoping I wouldn't need to invest in another PSU but after all it makes sense. I'll try and grab a good PSU in a month or so, hopefully it won't be too expensive. Maybe I can borrow one off the hardware shop and return it after I tried it to see if it makes a difference.

Btw you mentioned "erratic performance": the GPU is actually performing as it should when it's not busy crashing (and the clock fluctuations on my screenshots don't actually make the framerate change - strange). Its framerate is consistent with other 6950's (from the furmark score database + other forums etc..) Just thought you'd like to know.

Your other voltages aren't terrible, and the 12V rail is in-spec, but it's stressed well below where it should be. It should ideally be 12V even or a bit higher to supply adequate voltage to your PC components.

Hmm but if this reading is incorrect, would the reading within the BIOS be correct? Not sure how to put GPU load in the BIOS though.

The reason I asked about the card's model number as if it's reference, you may be able to flash to another Sapphire reference BIOS (from techpowerup's BIOS database) that has dual clock changes, and not get hit with the excessive clock changes. Knowing the model # would help ID whether that's a possibility to look into.

I was told on other forums not to flash a BIOS because it's really easy to end up with a dead card apparently. I downloaded "Radeon Bios Editor" and read the tutorial but I got frightened with all the options and closed it
I don't want to mess up my card, it cost me alot of money and I'm not really interested in overclocking, just want it to, like, work. Is there a safe way to "flash" the bios correctly without risking breaking my card?

If replacing the PSU doesn't fix the performance issues (try this first, and really, you need to do it anyways if you want your system to last for awhile), the next step I'd suggest would be to uninstall the CCC and just try using afterburner by itself to make sure there are no conflicts between them.

I already uninstalled CCC (since it doesn't work well with Afterburner, they would keep fighting over the fan speed and voltage value). I just have the ATI display driver right now with Afterburner - no CCC.

Beyond that you can use the unofficial overclocking feature in afterburner to disable clock changes and keep the card at 800mhz for testing (I forget what the setting is; it's a number value [0,1,2, or 3] but there's guides on google). If it doesn't have the same issues while stabilized at one clock, it's definitely an issue with Powerplay so we'd want to investigate a different BIOS with dual-clock changes instead.

I tried to do it but it didn't seem to have any effect. (I followed the tutorial and went into the .cfg, and entered the details, then rebooted and starting Afterburner but core clocks didn't come out any differently). I'll have to try it again, I probably made a typo or something.

If none of this works, and the card's under warranty, you'd need to contact Sapphire for an RMA, but we should troubleshoot a bit before that step.

Yes I really hope it doesn't come down to the RMA, since it would take forever considering I'm in NZ. Ah well~

 

[grumperfish]

But then how come it doesn't artifact or crash on Furmark or MSI Kombustor (which works well, btw, tried clicking on the "K", nifty) and works perfectly on most games/apps, and just randomly commits suicide only in certain specific (and reproducible) situations? Sorry if I seem kind of hard-headed, it just seems counter-intuitive.

I think it's a combination of the card changing power states in combination with the installed PSU. Switching that out may make a world of difference by allowing the card to give proper voltage when changing power states, especially as the card is otherwise running at expected and proper performance.

Hmm but if this reading is incorrect, would the reading within the BIOS be correct? Not sure how to put GPU load in the BIOS though.

There's not really a way to apply GPU load during BIOS, but HWiNFO is usually pretty spot-on. It may be off by .05V, but even still 12V would be running too low for your system indicating a PSU replacement.

I was told on other forums not to flash a BIOS because it's really easy to end up with a dead card apparently. I downloaded "Radeon Bios Editor" and read the tutorial but I got frightened with all the options and closed it
I don't want to mess up my card, it cost me alot of money and I'm not really interested in overclocking, just want it to, like, work. Is there a safe way to "flash" the bios correctly without risking breaking my card?

Flashing a new BIOS is fairly safe if done correctly, but we'd need to know the exact model# of the card to find the correct BIOS to flash. Do not do this before replacing the power supply though as it may be unnecessary. I used W1zzard's instructions with atiwinflash when I flashed by 6950, but I spent alot of time making sure I was using the right BIOS for my exact card. It's reference, so I could have used several different BIOS', but I wanted to make sure I was using the proper XFX BIOS.

I tried to do it but it didn't seem to have any effect. (I followed the tutorial and went into the .cfg, and entered the details, then rebooted and starting Afterburner but core clocks didn't come out any differently). I'll have to try it again, I probably made a typo or something.

They changed the instructions fairly recently, so make sure you're using recent instructions and restarting Afterburner. Use the option that disables PowerPlay to keep it at a single clockspeed.

Yes I really hope it doesn't come down to the RMA, since it would take forever considering I'm in NZ. Ah well~

Hopefully not, yeah, but after you replace the power supply the card may be fine. It's better to do that first rather than RMA a card that's in all likelihood working properly.

 

[Bacterius]

There's not really a way to apply GPU load during BIOS, but HWiNFO is usually pretty spot-on. It may be off by .05V, but even still 12V would be running too low for your system indicating a PSU replacement.

Oh ok, it's at 12.096 idle, and 11.7-11.8 on gpu load. Not good

Ok, going to grab a new PSU in a few weeks and see if that fixes it. Luckily I still have the card's package box so I'll be able to get the exact model off of it so I can try flashing it if the PSU change doesn't work (but hopefully it will).

Thanks a lot for your time Grumperfish I'll come back to you after the new PSU is installed see how it goes.

 

[grumperfish]

Oh ok, it's at 12.096 idle, and 11.7-11.8 on gpu load. Not good

Ok, going to grab a new PSU in a few weeks and see if that fixes it. Luckily I still have the card's package box so I'll be able to get the exact model off of it so I can try flashing it if the PSU change doesn't work (but hopefully it will).

Thanks a lot for your time Grumperfish I'll come back to you after the new PSU is installed see how it goes.

No problem, glad we could ID at least one of the issues. It would not be fun to RMA a card and then find out the issues persist due to another problem. Don't flash the BIOS until you try out a new PSU (check the stickied "Good PSU" links in that section of the forum). If you card is reference you can likely find a BIOS that will work fine on Techpowerup's site (check the GPU BIOS) section, and you could probably unlock the extra shaders if the card is a reference model.....

If the new PSU fixes the problems you're seeing of course.


Edit: Subscribed to this thread, so I'll see it when you post back with the new PSU.

 

[OnDborder]

You could give this a try..
Run regedit
First "export" your current registry.
Disable EnableUlps
Do a "find" "enableUlps". Or..
HKEY_Local_Machine>system>ControlSet001>control>class>{4D36E968-E325-11CE-BFC1-08002BE10318}>0000,0001,0002
Open the "EnableUlps" and change the 1 to 0.

PSU-"12.096 idle, and 11.7-11.8" is way too much flucuation.

 

[Bacterius]

You could give this a try..
Run regedit
First "export" your current registry.
Disable EnableUlps
Do a "find" "enableUlps". Or..
HKEY_Local_Machine>system>ControlSet001>control>cl ***>{4D36E968-E325-11CE-BFC1-08002BE10318}>0000,0001,0002
Open the "EnableUlps" and change the 1 to 0.

This didn't work - the relevant keys didn't exist and creating them did nothing. I read up on ULPS and it doesn't seem to apply to single cards though.

I'm going to be looking at a Corsair PSU because they seem to have a good reputation (even though they are more expensive but you get what you pay for I guess). Hopefully it'll be in before the end of the year and the issue will be fixed.

Meanwhile I cobbled up a ghetto fix for my issues, I tweaked the fan parameters a bit so that its temperature stabilizes around 55C even under minimal load (by setting fan speed to 20% under 55 degrees and to 60% above 60 degrees). I still need some heat to get it going but at least it doesn't crash on me anymore, so I can still use my computer. I might downclock the card a bit until I get a new PSU to make sure I don't end up damaging the card by accident.

 

[grumperfish]

^^^
Sounds like a good plan. For PSU's anything Corsair is generally OK (above the "Builder"/CX series). Seasonic is good as well, and most of XFX's PSU's are re-branded Seasonics. On top of that, Thermaltake's "ToughPower" line is good (but not litepower/TR-2/PurePower series'), and Antec's Earthwatts series is very good.

You'll want to look for a 550W+ unit from one of these brands. Avoid OCZ PSU's.

 

[Bacterius]

Okay I'm about to buy the PSU and I'm thinking of getting this:
CORSAIR Professional Series Gold AX750 (750W) but I'm reading what you said about Corsair ("above the "Builder"/CX series) and I'm a bit confused with what "above" means can you tell me if this is good?

Otherwise there's a Corsair Professional Series HX750W (750W).

Which one (if either) would be best for my system?

 

[grumperfish]

By 'above' I meant to look at the TX/HX/AX series units.

Either of those you linked would be fine. The AX has slightly better components, and the fan will completely shut off when it's not being loaded. Personally I'd go with the HX as it's a very solid unit, and it's a little cheaper than the AX.

 

[Bacterius]

By 'above' I meant to look at the TX/HX/AX series units.

Either of those you linked would be fine. The AX has slightly better components, and the fan will completely shut off when it's not being loaded. Personally I'd go with the HX as it's a very solid unit, and it's a little cheaper than the AX.

Great! I'm going to order the HX one sometime next week I think. Just in time because the crashes have been happening more and more often recently.