Power Supply Failing? Or Graphics Card?

[HurgyMcGurgyGurg]

System:

Core 2 Duo e 8400
HD 4870
WD 640 GB
2x 2 GB DDR 2 @ 800 MHz
Gigabyte EP45 DS3L

Detailed PSU Specifications:

500 watt Silver Stone SST-ST50F
2 x 12v Rail 18.0A
PSU Rated to supply 2x Hd 4850.
2 PCI-E power connectors
7 Months Old

The first 4 months were all fine. No issues at all, the system ran completely stable with a 500 MHz OC on the CPU and a 25 MHz OC on the 4870 (75 MHz total OC since the card already had a 50 MHz factory OC). I folded on this 4870 for several days at a time without any errors. Then I ran Furmark as a stress test. I had started worrying about my graphics card because it had a PCS+ 3rd party cooler on it. These have great GPU temps (58C while folding @ 30 % fan) but the VRM temps get up to 110 C on it. When I ran the Furmark stress test it triggered the second LED on the HD 4870 to turn on. This one indicates a critical core power fault, my system crashed about a second after this light went on. VRM temps shoot up to 150 C in about 20 seconds I didn't have time to stop it.

On reboot the light was gone, and didn't come back even when the 4870 was folding. I forgot about it (Didn't run Furmark again). Then, about a month later the light came back at power on. I shut it off, turning it on again the light was gone. This continued for about 2 months. Rebooting or changing a BIOS option that required a cold reboot, got rid of it most of the time. With the LED on the system runs perfectly stable as long as no 3D work is being done. It can play flash games, watch a movie, desktop works fine, etc... but try playing a game an it crashes. The interesting thing is I accidentally discovered folding runs perfectly fine with the light on but works about 10 times slower. The CPU can also run at 100% and the system is stable.

Two weeks ago the light began being triggered just by trying to run 3D programs, (Instead of the usual on or off at start). Most of the time this results in system lock up but sometimes I can get task manager up and end the program. Now the weird thing is sometimes running 3D programs won't trigger the light and it works fine. (I'm folding right now, perfectly stable, but two reboots ago folding triggered the light).

If I listen closely, when the light turns on there is an audible high pitched noise that beings, very quiet, but much like the noise of an old CRT television.

So the question is, is my PSU beginning to fail to supply enough power to the HD 4870. The HD 4870 works perfectly fine without the light on. Or is the HD 4870 failing.

Do you think the 4870 is still alright?

What I have tried to fix the problem:

Removed all OC's
Un-plug Re-plug PCI-E connectors
Re-sit the graphics card in the PCI-E x16 slot
Clean the cards fan + Heatsink for dust (Wild guess)

None have worked.

What I can't do but is usually recommended here:

Try a different PSU. (Only have laptops besides this no other adequate PSUs)
Try a different graphics card (Old one fried with my last PSU 7 months ago during a move, NOT my fault, my Dad (He was helping with the move) accidentally plugged it in thinking the 110v would work with 220v where I live now.)
Try a different PCI-E x16 slot (The DS3L only has one)
Try another PCI-E connector (Only 2 with PSU)

I'm toying with the idea of digging through my stash of cables to find a molex to PCI-E connector to see if moving the power draw to a different cord helps.

Any ideas at all?

Thanks

[killadark]

you should try underclocking (gpu) and see if it works if so then the problem would be the psu

[Flying Fox]

So when you see the LEDs light up (or not), does that result in screen artifacts, hangs and/or VPU recover? If not, are you just being a bit paranoid about something that does not show up on screen?

[HurgyMcGurgyGurg]

Whenever I try to start a 3D app with the light on the system locks up completely once the first 3D graphics are shown (As in I can get to the start menu for most games but the level will crash a second after the first graphics are shown). Sometimes I'm able to get task manager up but thats a 1/10. 2D things such as flash games, movies, and the desktop work perfectly fine with the light on, no artifacts. The graphics card does not have ANY issues at all when the light is off, I can even take it back to its old overclock and it still runs stable with no artifacts. I occasionally get VPU recovers due to folding and having a high FLUSH_INTERVAL value, but that's happened before this even started.

I'll try underclocking it for a few days and see if the light shows up.

Anyone else got ideas?

[HurgyMcGurgyGurg]

The light is still coming on at boot up most of the time, probably because the under clocking only affects the graphics card once CCC loads.

This morning the system failed to post several times. (Fans spin up, HD is audible, but no post beep and no video feed)

Did the standard isolation process. Removed the DVD Drive, the extra stick of RAM, excess fans, any expansion cards, didn't work. Swapped RAM sticks, nothing. Then randomly on the third or fourth try it worked again. Loaded back on all the standard equipment, still working. Talk about messed up.

On boot the system was stable, however the light was now triggered by watching a movie, the movie played fine but the light went on.

What's the next course of action? This is clearly escalating fast...

[Meadows]

How about getting a proper cooler that doesn't burn your VRM to a crisp?

[HurgyMcGurgyGurg]

I guess your responding to my previous comments on various threads, I've mounted a fan blowing right below the card, before there used to be a hotspot right around the VRMs, now its gone and VRM is below 100 c all the time.

This problem can't be related to VRM, the light indicates a critical core power fault, not VRM.

Anyways, when the light isn't on the Hd 4870 runs without any artifacts at all, no VRM problems.

[MrJP]

Sounds like the card is dying. Did it come with the cooler you describe or did you fit it? If it's still the original cooler, then you should be able to get the card replaced under warranty (ignoring for a second that your minor overclock technically invalidates the warranty).

[HurgyMcGurgyGurg]

Its the stock cooler.

And I will make sure to not mention any overclocks...

However, a quick search of the internet reveals the Furmark problem which caused this in the first place is actually their problem as they have insufficient VRM heatsinks.

You really don't think its the PSU?

As a side note, PowerCooler does not seem to have a website, that's a little ridiculous?

[Contingency]

As a side note, PowerCooler does not seem to have a website, that's a little ridiculous?

A top result on Google for "Power Cooler 4870" is on Tom's. Based on the system specs, it's probably you. Try PowerColor.

[HurgyMcGurgyGurg]

It's not me, but thanks for the link.

I'll try a few more emails then...

[MrJP]

I don't think it's likely to be the PSU because it's a reasonable quality unit with more than adequate power rating for your system. The sequence of events you describe strongly suggest that the GPU VRM was damaged by excessive heat when you ran Furmark, and you're now getting GPU core power supply problems which is exactly what you'd expect to see with a flaky VRM. I think you've got very strong grounds for getting the card replaced.

[mattsteg]

[quote="HurgyMcGurgyGurg]This problem can't be related to VRM, the light indicates a critical core power fault, not VRM.[/quote]
Stop for a moment and consider the function of Voltage Regulator Modules.

I'd also tend to suspect the card much more than the PSU here.

[HurgyMcGurgyGurg]

Sorry, I got my abbreviations confused. I've been meaning VRAM when I've said VRM. GPU-Z lables the VRMs as VDDC slaves so I've usually been calling the VRMs VDDCs.

However, the VRAM and VRM temperatures are about the same through all my test so all the temperatures are still relevant.

That's why I was wondering what VRAMs would have to do with a critical core power fault.

Anyways, so it's the VRMs going whack? Sorry about the confusion. I'm stupid for not realizing this earlier.

Good news is I just got a reply from Power Cooler recommending I start the RMA process for the HD 4870, hopefully this will work out.

[nerdrage]

VRM temps shoot up to 150 C in about 20 seconds

I believe I've read on other forums (XtremeSystems probably) that the max rated temperature for the 4870 VRM is 125C. Since you've FAR exceeded that, it seems likely that you've damaged it. Furmark is known to cause runaway overheating of the VRM on ATI cards -- a situation that does NOT occur when running normal apps like games. ATI has even gone so far as to detect when Furmark.exe is running and slow down the card to try to prevent overheating. Sorry, but I think you're going to need to get a new card.

[HurgyMcGurgyGurg]

Yea, I'm starting the RMA process waiting for them to validate...

It turns out that a lot of Power Cooler PCS+ HD 4870 512 MB are failing due to this problem. It looks like they screwed up when they designed this card.

1. It was the first non-reference HD 4870 so they were definitely rushing.
2. It has insanely high clocks (The reason I bought it) 800 GPU / 950 Mem, I mean 6 months after release there isn't another card selling (At least on Newegg) with clocks anywhere near this)
3. This cooler sets the records for the fastest fails in Furmark.
4. The card has been discontinued its replacement has 20 lower GPU clock and 20 lower Mem clock
(Hmmm I wonder why.....)

So odds are they will probably take it back no questions asked, because they do not want the bad press. So far I haven't found complaints of people getting rejected with this problem, looks like they recognize it is their's.

Now I have to deal with the whole, "I live in Asia but I have a valid address in the US so don't give me trouble about warranties, and don't ask me to ship it overseas to get it fixed thing"

[HurgyMcGurgyGurg]

Well at least it's validated, now to ship the thing overseas, 2 weeks, 3-4 weeks for RMA process, 2 weeks back, that's 2 months. Mobo doesn't have integrated graphics, time to start shopping for a cheapo temporary card...

Think the HD 4350 is to cheap? Any low end Nvidia worth getting?

[MrJP]

Buy another 4870 then Crossfire when yours come back??

Seriously, it depends how low a performance level you're prepared to tolerate for a while. Whatever you buy will depreciate in the meantime, just by less if you buy a cheaper card. I might have missed it, but have you mentioned what resolution screen you're running? Assuming you're at the sort of level that justified a 4870 in the first place (i.e. 1680x1050 or higher) and you're a really keen gamer, then you might find the really cheap cards to be too limiting even for a short time. so should look towards something like the 4670 or 4830. Perhaps even look on it as opportunity to try out the competition, pick up one of the current good deals on a GTX 260, then resell either that or the 4870 when it comes back depending on your preference. Whatever you do, try to hold out for a week or so to see what falls out from ATI's new introductions - pricing of some of these options may well be affected.

[HurgyMcGurgyGurg]

I wish I had the money right now, anyways I kinda scoffed at CrossFire as this mobo only has 1 16x PCI-E and a 500 watt PSU isn't going to cut it for two HD 4870s!

I'll survive for two months some how on a cheapo graphics card. I doubt the low end will shift much in price in the coming weeks.

It's a 1680x1050 monitor, considering a Radeon 9200 managed to power it satisfactorily on XP a few years ago (It was my last gap card when my other graphics card failed) I think pretty much anything even the ultra budget HD 4350 will cut it for now.

Plus, odds are it wouldn't hurt to have a low power card in the PC for two months, probably will save its cost on electricity in that time compared to the HD 4870 .

[HurgyMcGurgyGurg]

Erm, new info, the 4870 is now working stable even at load at the minimum clock speeds allowed in CCC 475 MHz GPU and 500 MHz MEM.

The light still on but it works at this speed, is that still likely to be failing VRMs or PSU underpowering?

Sorry for the hesitation but with a 2 month turnaround time to get this back I'm trying to be as sure as possible this isn't the PSU.

[MrJP]

Get a cheap Watt meter. Plug in between PC and wall. Run your choice of software with the low clocks on the graphics card, and try again with the standard clocks which cause the crashing problem. In each case record the maximum power usage. With your system this will be well under 300W, so PSU output will only be about 50%-60% of its rated maximum load. Unless it's seriously faulty, the problem is almost certainly not with your PSU. You could also use some monitoring software to log voltages to check for signs of voltage dipping significantly under load.

However I'd still guess that the issue is the damaged VRM supplying a lower than normal voltage to the GPU. This is fine if you underclock the GPU far enough, but it's not enough to run the GPU at full speed. If this was an old card and you had no other option left to try, you could try modifying the card BIOS to up the core voltage, which might get the VRM to up its output back to the original level. However since you've still got a warranty and the manufacturer have already admitted there's a problem, I'd just send it back.

In reality the only way to be really sure would be to test with a spare PSU or graphics card, but if you don't have these handy then you don't have many options. If you were already going to pick up a cheap card to stand in while you sent the 4870 back, then at least you can test with this before you send off the 4870.

[HurgyMcGurgyGurg]

Buying some low end card around the $50 range today. I know there is the whole argument that $20 more doubles your performance but to be honest I just don't have the money right now.

I'll make sure to test the new card before I ship it off.

[HurgyMcGurgyGurg]

Running stable with an HD 4550, god it sucks in games but I'll live with it for two months...

Time to ship out the HD 4870.

And you know they're going to fix it no questions asked when they never even asked for an invoice or proof of purchase.