7950 crashing in some games

[pinkeen]

Hello!

First of all, to be clear - I don't do any OC. Putting my boxes together by myself gives me enough thrills and fun. I prefer to tweak my software .

I bought Sapphire HD7950 DualX Boost (925/1250 stock clocks) half a year ago. All of the games ran fine. The only one that had problems was Crysis 3 (that was one of the games I tried first). It randomly crashed with grayish vertical stripes. Only hard reboot helped. I thought it was driver issues at the time and forgot about it, because I didn't like Crysis anyway.

The problems reappeared when I wanted to play Metro LL. It crashes randomly too. Sometimes it's the stripes, sometimes gray screen, sometimes black screen. Only hard reboot works. There are no artifacts, game runs close to 60fps most of the time.

Batman Arkham Origins has the same problems.

The thing is I finished many other demanding games on this card (like FarCry 3) with no problems.

First guess - heat. Nope, that's not it - card stays cool at 60-63°C during gaming. CPU stays at 60-70°C during max stress too.
Second guess - PSU. Not that too. I bought a brand new Lepa G750-MAS - didn't help at all.
Third guess - RAM. Left memtest running for the night - 100% OK. Switched the sticks around - didn't help. Disabled dual channel - no dice.

I tried updating mobo bios, updating/downgrading drivers, reinstalling system, even tried Windows 8.

I logged temperatures/voltages/clocks/fans from all of the sensors and there are no anomalies before the crash. Everything's solid and stable at all times.

I noticed some other strange things too:
- I can throw Furmark, OCCT, whatever at it and it doesn't crash, everything's stable. Well, temps during stress tests are around 70°C (around 10°C more than during gaming) but that's perfectly normal.
- During burn-in in Furmark the core clock is being throttled back to 850 most of the time and the vcore constantly changes oscillating around 1.0V. During gaming vcore is stable at 1.17V. That's probably PowerTune in action, but I thought that it isn't supposed to change vcore, only the clocks.
- The card has two bioses 015.029.000.002.000000 (no light) and 015.030.000.001.000000 (blue light on the bios switch). The strange thing is both bioses have the same clocks (925/1250) but the second one which should be faster has even lower PowerTune limit. The card won't accept any other bios - compatible (dev/subsystem id) or not. I tried all of the 100% compatible ones and a few others. Computer won't boot, no POST, just black screen.

There absolutely no problems with my old GPU (GTX260), so I doubt that some other piece of hardware is at fault.

Trying to RMA is a last resort because Sapphire doesn't allow direct RMA and I lost the receipt. I can try and reason with the guys from the small shop, where I've bought it (I ordered it on-line but picked it up in person, so they have a record of the sale) but I have to be sure it's the card.

My box:
FX-8350
Asrock 970 Pro3
1xHD7950
Lepa G750-MAS
2x4GB RAM (plain kingston sticks, not some extreme OC stuff and whatnot)
1x spinning HDD

If anybody's wondering - the PSU has single 62A (744W) 12V rail and according to some in-depth reviews it's quite good. 750W is more than enough to handle my system.

 

[suraswami]

try increasing the powertune % to 10% and see if it stabilizes.

Check out the mobo VRM temp, it might be heating up. Put a 80mm fan on it or open up the case and try playing your game. Since you already did basic troubleshooting with graphics, let see if the mobo is good, that cpu is known for sucking lot of juice, try to lower its clock and undervolt it to see if the problem goes away.

Powertune option for the graphics card should help.

 

[Blín D'ñero]

Even though you don't overclock, enable Overdive in CCC and set the power slider to 20%. It's a measure to prevent underpower.

The voltage fluctuation you see is normal: it is supposed to be dynamic (the set vcore is just the max value).

 

[Random Murderer]

My guess is your VRMs are overheating, either on your motherboard or GPU itself.

 

[Red_Machine]

MY GTX 580 had a similar problem, which I figured was driver-related. But several revisions later it was still doing it. I cleared all the dust out of it with compressed air and put it back, no further issues. Just try that and see if it helps.

 

[Blín D'ñero]

My guess is your VRMs are overheating, either on your motherboard or GPU itself.

DualX has two fans, plus card isn't overclocked. Overheat is very unlikely. Underpower during peak performance (Crysis 3, Metro): likely yes, if the powerslider is at 0.

 

[pinkeen]

DualX has two fans, plus card isn't overclocked. Overheat is very unlikely. Underpower during peak performance (Crysis 3, Metro): likely yes, if the powerslider is at 0.

I've read that the VRMs have poor cooling (someone slapped thermal pads on them) on Sapphire DualX/VaporX GPUs - I haven't checked if mine has those funky pads yet. Anyway I doubt that it's the VRMs because most people find this card very stable, even during overclocking.

I would rather expect poor ASIC quality --> higher power requirements. It's seems strange that I have to adjust the PowerTune with stock clocks. What's even more strange is that the GPU's stable during stress testing and crashes in games (but not all of them).

I didn't want to suggest anything but I already tried adjusting PoweTune to +20% yesterday and Metro didn't crash for an hour. Didn't have time to test longer so it's not a definitive fix yet. Usually it crashes after 20-40 minutes but not always. I will do more testing and report back.

Thank you all for the tips! Didn't expect such swift and constructive response .

Even if it works the topic isn't over, because I want to know why it works... :shadedshu

UPDATE:
It just happened again after ~3 minutes of playing metro. Powetune is set at +20%.

 

[ruff0r]

UPDATE:
It just happened again after ~3 minutes of playing metro. Powetune is set at +20%.

Use GPU-Z to check on your VRM Temperatures. Make sure.

 

[pinkeen]

Use GPU-Z to check on your VRM Temperatures. Make sure.

GPU-Z doesn't display VRM Temps. How can I check them?

Update:
I have another theory - people all over internet write that the Sapphire HD7950 VaporX cards have a stock bios that sets 1.5V on the memory. The nominal mvddc should be 1.6V according to the manufacturer's specs. This causes random crashes in games that use a lot of VRAM. This would explain why the card doesn't crash during stress testing (low vram usage) and in most games, but crashes in others. I noticed that Metro has very detailed textures. One of the highest resolution textures I've seen in games.

I contacted Sapphire's support but they aren't helpful. They didn't even acknowledge the tests that I've already done but repeat the same stuff like mantra. Their best advice is to install the drivers from the CD which came with the GPU.

 

[suraswami]

Try HWInfo and log the values.

 

[pinkeen]

Try HWInfo and log the values.

It seems that my card doesn't have the sensors. HWInfo doesn't show VRM temps either.

 

[Random Murderer]

Update:
I have another theory - people all over internet write that the Sapphire HD7950 VaporX cards have a stock bios that sets 1.5V on the memory. The nominal mvddc should be 1.6V according to the manufacturer's specs. This causes random crashes in games that use a lot of VRAM. This would explain why the card doesn't crash during stress testing (low vram usage) and in most games, but crashes in others. I noticed that Metro has very detailed textures. One of the highest resolution textures I've seen in games.

I contacted Sapphire's support but they aren't helpful. They didn't even acknowledge the tests that I've already done but repeat the same stuff like mantra. Their best advice is to install the drivers from the CD which came with the GPU.

You probably saw Ket's thread(s) about it. AFAIK, this only affects the VaporX.
That being said, have you tried both of your card's BIOSes? The general consensus in the VaporX threads was that one BIOS had the correct memory voltage, while the other was undervolted on the RAM. If you believe this to be your problem, try the other BIOS on your card (if you haven't already) with it's powertune raised to its max setting.

Also, when you crashed so quickly in Metro with +20%, did you check the core temp? As for VRM temps, if your card doesn't provide sensors for them, then it'll be very tough to gauge how hot they're getting. You may have just missed them, when monitoring my 7970s in GPU-Z I have to actually scroll down on the sensors page to find the VRM temp sensors.

 

[pinkeen]

You probably saw Ket's thread(s) about it. AFAIK, this only affects the VaporX.
That being said, have you tried both of your card's BIOSes? The general consensus in the VaporX threads was that one BIOS had the correct memory voltage, while the other was undervolted on the RAM. If you believe this to be your problem, try the other BIOS on your card (if you haven't already) with it's powertune raised to its max setting.

I thought that VaporX and DualX are close cousins. I've already tried the second bios but not with PowerTune adjusted. Will check it out, thanks.

Is there a way to check mvddc?

Also, when you crashed so quickly in Metro with +20%, did you check the core temp?

Forgot to enable logging this time but I never saw the temps rise above 72 deg.

As for VRM temps, if your card doesn't provide sensors for them, then it'll be very tough to gauge how hot they're getting.

Definitely don't have the sensors.

BTW Sapphire's support suggested that I test the card in a different PC to isolate the problem. It sounds like not so bad advice, although I'm 90% sure it's the card. The problem is my second box is a bit underpowered for this card (Phenom X2) and the results can be deceiving. When you think about it some more - the results would be deceiving in most cases. The only reliable way to isolate this problem would be to swap each component one-by-one.

I have the impression that Sapphire is stalling.

BTW2 I looked at the VRMs - they don't have thermal pads like on the VaporX but there's an airgap (1-1.5mm) between them and the heatsink. Actually they don't have the solid part of heatsink above but only radiator fins. This might be good because the fan is just above the fins.

 

[Random Murderer]

I thought that VaporX and DualX are close cousins. I've already tried the second bios but not with PowerTune adjusted. Will check it out, thanks.

Is there a way to check mvddc?

Not that I'm aware of if it's not in the GPU-Z sensors tab. (other than multimeter, of course)

BTW Sapphire's support suggested that I test the card in a different PC to isolate the problem. It sounds like not so bad advice, although I'm 90% sure it's the card. The problem is my second box is a bit underpowered for this card (Phenom X2) and the results can be deceiving. When you think about it some more - the results would be deceiving in most cases. The only reliable way to isolate this problem would be to swap each component one-by-one.

I have the impression that Sapphire is stalling.

It's worth a shot. I assume you have steam, so install it on the Phenom rig, toss the 7950 in, and play some Metro.

BTW2 I looked at the VRMs - they don't have thermal pads like on the VaporX but there's an airgap (1-1.5mm) between them and the heatsink. Actually they don't have the solid part of heatsink above but only radiator fins. This might be good because the fan is just above the fins.

Are you sure you're not looking at the chokes? The chokes are the squarish bits and the VRMs are short and behind the chokes. For a better example, I uploaded a pic(credit goes to w1zz for the original picture):

Spoiler

The chokes are circled in RED and the VRMs in GREEN.

 

[pinkeen]

Are you sure you're not looking at the chokes? The chokes are the squarish bits and the VRMs are short and behind the chokes. For a better example, I uploaded a pic(credit goes to w1zz for the original picture):

Spoiler

http://img.techpowerup.org/131031/front_full.jpg

The chokes are circled in RED and the VRMs in GREEN.

You're right, I was looking at the chokes.

I have another idea - I've got an old GTS250 lying around. I'll toss it in and try playing some games...

I suspect that something's wrong with the mobo. Many people over the net complaining that asrock's Pro3/Extreme3 mainboards freeze a lot. Especially when paired with FX-8350.

I noticed a weird thing - the default (@ auto) memory voltage is 1.585V. I forced it to 1.5V and will se what happens. Is it possible that CPU-Z/HWinfo report the memory timings incorrectly? Because they both say RAM is at 4-5-5-15 (667Mhz). That's crazy because I'm pretty sure the computer wouldn't post with these timings. I tried to force them in BIOS to match the SPD but software still reports 4-5-5.

 

[vega22]

mem volts should be 1.6v not 1.5v that being said i would bet they will do stock clocks on 1.5v too as mine will do 1500 on 1.6v.

have you tried to see if the sensors read right on the other bios switch?

have you tried to flash the bios of the card to a newer one?

the way it is crashing on crysis sounds like it is undervolted/clocked too high to me, from an overclockers perspective, but i would try dropping the core clock back with something like trixx or afterburner and seeing if that helps it any too.

if yours is like mine (idk tbh) the black plate under the fans and heatsink should be cooling the ram and vrm so that shouldnt be an issue.

aida64 reads every sensor on these cards too.

 

[Blín D'ñero]

[...] Is it possible that CPU-Z/HWinfo report the memory timings incorrectly? Because they both say RAM is at 4-5-5-15 (667Mhz). [...]

Yes CPU-Z can be wrong about actual multiplier, actual core speed, actual ram timings... just trust what you put in in BIOS yourself.

 

[pinkeen]

The strangest thing happened. I cleared the CMOS (jumper & battery).
Now I can't run anything graphically intensive. Metro LL crashes few secs after launch. OCCT GPU test crashes the display as soon as hit the "ON" button. It worked a few times (after clearning the CMOS) but now it crashes all the time.

Even windows crashed a few times and wouldn't boot a few times.

Now I am clueless.

Even stranger thing is that I can run Unigine Valley under Linux without any problems.

UPDATE:
Tried booting to linux once again - crash. This time the computer didn't boot. No fans, no nothing, just the power light and the noise of hdd spinning up. I put the gts250 in and everything seems to be working, so the radeon is pretty dead. 400USD down the drain (yes, pc hardware is a bit more expensive here than in US). What happened?

mem volts should be 1.6v not 1.5v that being said i would bet they will do stock clocks on 1.5v too as mine will do 1500 on 1.6v.

According to SPD and manufacturer it should be 1.5V.

have you tried to flash the bios of the card to a newer one?

There is no newer bios. I tried to flash every compatible bios but not one will boot.

the way it is crashing on crysis sounds like it is undervolted/clocked too high to me, from an overclockers perspective, but i would try dropping the core clock back with something like trixx or afterburner and seeing if that helps it any too.

Tried underclocking the core & mem to no avail.

 

[pinkeen]

The card isn't fried.

The computer started shutting down, rebooting and all of the weird stuff even with the old card. I took it apart and carefully removed all the dust with compressed air. Put back the 7950 in and it's working. I think that some dust has gotten into the empty RAM slots while I was clearing the CMOS.

I haven't checked if the problem with games still persists, but I would guess it does. The recent crashes were probably unrelated.

I noticed a strange thing - one of the memory chips has no cooling. The other ones are cooled by the main heatsink but the one on the side is completely naked. Is this right? It would explain why only the games using much VRAM crash. What do you think?

I found a picture on sapphire's website - apparently it is naked by design:

BTW The VRMs have a dedicated heatsink so they should be fine.