Question GPU not working and then caught on fire

Jul 27, 2021
5
1
15
The story: I got home after a few days away with the PC being unplugged. It doesn't start. I disassemble and unplug everything, connecting one component at a time to deduce where the problem lies. Everything works (PSU, MB, SSD, CPU) until I connect the GPU, at which point the PC won't boot.

Next, I keep the GPU inserted into the MB, but unplug the power cable that goes into the GPU from the PSU, and start the PC again. Now, it does boot, but a component on the GPU catches on fire. I obviously quickly shut the PC down.

I removed the fan from the GPU to assess the damage on the board, this is what it looks like:
Overview:
QWF3hoC.jpg


Close-up:
qynzBdb.jpg


The card is a GTX 1070, specifically a MSI GeForce GTX 1070 8GB ARMOR OC. As can be seen, the component that blew is labeled Q37, which seems to be a MOSFET transistor.

I have obviously not tried to boot with the card inserted since. The card was bought in 2016, so it's out of warranty.

Any info or input on options (primarily feasibility of repair) you can give would be appreciated.

Specs:

GPU: MSI GeForce GTX 1070 8GB Armor OC
CPU: Intel Core i7 6700K 4.0 GHz 8MB
MB: Gigabyte Z170-HD3
RAM: Corsair 32GB (2x16GB) DDR4 2133MHz CL13
PSU: Corsair CX650M 650W
OS: Windows 10
 
Last edited:
These graphics cards are not meant to be user serviceable. You would need a highly skilled technician to replace the burned components.

the burned chip is a Magnachip MDV3605 which is a MOSFET responsible for power modulation and voltage control. You may be able to find one for sale and someone skilled enough with solder to replace it for you, but I would not be too hopeful

The MOSFET being so hot that it actually caught fire suggests the card was subjected to very high voltage which it could not control. In such cases it is rare for only the MOSFET to be damaged. Likely, several of the chips on the card are damaged, possibly including the GPU itself, but that MOSFET is the only one that caught fire.

In which case you could spend a long time chasing down and replacing every broken component on the card before finding that the GPU itself is fried and Nvidia don’t sell those as spares.

It’s probably worth a gamble replacing just that one chip, but if that doesn’t work (which is the most likely outcome) then don’t go any further. Sell the card as broken for parts on eBay.
 
  • Like
Reactions: jay32267
How does a GPU blow like that?😗 Is it to do with a too low-powered PSU?

I sure wish I knew. I've used this GPU and PSU combo for over 4 years, so it'd seem strange to me if it had to do with it overloading when simply starting it (as opposed to running a graphics intensive application, for example).

However, also note that the GPU blew when I didn't have it plugged into the PSU, which I have no idea what it means.
 
Also, don’t plug any more graphics cards into that PC until you have replaced the PSU with a high-quality, tier A or tier B PSU.
What power supply was he using?
I see a gtx1070 power is 150 they recommended a 450 PSU.
They do mention the possibility of them catching fire, but never heard of someone,s catching fire
.I just looked it up on Utube. looks like someones has cached fire.
And it seems there were problems with the 1070 and 1080 the vrm thermal pads
wow.😵
View: https://youtu.be/WAbl0fLY06U?t=7
 
Last edited:
How does a GPU blow like that?😗 Is it to do with a too low-powered PSU?

Hard to tell from the description of what happened and not being there in person, but seemingly it was plugged in without the ATX 12V cables so was trying to draw 300W from the PCI-E socket, on what sounds like a possibly shorted system that wasn’t powering on to begin with. Possibly due to a PSU short.

Best guess is that the PSU sent too many watts down the the VRM that controls power from the PCI-E socket, it couldn’t handle the load and that MOSFET is part of that VRM.
 
What power supply was he using?
I see a gtx1070 power is 150 they recommended a 450 PSU.
They do mention the possibility of them catching fire, but never heard of someone,s catching fire
.I just looked it up on Utube. looks like someones has cached fire.
And it seems there were problems with the 1070 and 1080 the vrm thermal pads
wow.😵
View: https://youtu.be/WAbl0fLY06U?t=7

I have updated the initial post with my specs (should have included those from the start). The PSU is a Corsair CX650M.
 
I have updated the initial post with my specs (should have included those from the start). The PSU is a Corsair CX650M.

Update: I have tested two other cards in the PC now. Although they are both old (a GTX 460 and a Radeon 4670), neither of them have encountered any issues.

I'm looking for a replacement card now, but I'm obviously wary. I feel quite confident that it was the card and not any other part that failed, but what, if anything, can/should I do to deduce whether or not that's true? Or should I just plug it in and cross my fingers?
 
Last edited:
There was a short that blew the Mosfet and most likely a capacitor or 2 around it, but where that short is from is very hard to find. It could come from a VRM chip, memory chip that has gone, you would need a microscope camera, voltage feed and a heat camera etc, etc etc.....to see what might be the cause. Very few shops can repair this sort of thing and with a RTX 3090 it might be worth the cost but a 1070, not really worth it.

The PSU, while not the cream of Corsair it is a decent one and unlikely the cause of your GPU dying.
 
  • Like
Reactions: david slayer
What power supply was he using?
I see a gtx1070 power is 150 they recommended a 450 PSU.
They do mention the possibility of them catching fire, but never heard of someone,s catching fire
.I just looked it up on Utube. looks like someones has cached fire.
And it seems there were problems with the 1070 and 1080 the vrm thermal pads
wow.😵
View: https://youtu.be/WAbl0fLY06U?t=7
Those were EVGA specific issues when they skimped on thermal pads.
 
He pulled the atx plug from the GPU and it tried to pull all of its juice directly from the PCIe x16 header. I am going to be surprised if the GPU voltage regulator is the only thing burnt. The Card pulls power from both sources together, so when you pulled the high amperage cable it over pulled the PCI, blowing the regulator. There is a good chance you zipped your motherboard. Those channels aren't fused, which is why it smoked. Never hot swap (unless the device is directly rated) and never turn anything on without all cables plugged in.
 
He pulled the atx plug from the GPU and it tried to pull all of its juice directly from the PCIe x16 header. I am going to be surprised if the GPU voltage regulator is the only thing burnt. The Card pulls power from both sources together, so when you pulled the high amperage cable it over pulled the PCI, blowing the regulator. There is a good chance you zipped your motherboard. Those channels aren't fused, which is why it smoked. Never hot swap (unless the device is directly rated) and never turn anything on without all cables plugged in.

I've tried to research what happens when you start without power plugged into the GPU. While information is scarce, the majority of answers I've seen have said that essentially, the PC wouldn't start at all and give an indication about insufficient power. But considering the circumstances, I guess exceptions occur. It seems unlikely to me that my result would be typical though. I can imagine that power cables being forgotten happens at least somewhat regularly when PC's are built or maintained, and if this invariably lead to the destruction of a new and very expensive piece of hardware, it'd surely be considered a severe design flaw.

Anyways, I tested my PC with an old card to make sure it was still fine, which it seemed to be. I then bought a 1080 second hand for 2800 SEK (~£230) which seems decent considering the shortage and other prices I've seen. So far, it is running without issue.
 
  • Like
Reactions: david slayer
I've tried to research what happens when you start without power plugged into the GPU. While information is scarce, the majority of answers I've seen have said that essentially, the PC wouldn't start at all and give an indication about insufficient power. But considering the circumstances, I guess exceptions occur. It seems unlikely to me that my result would be typical though. I can imagine that power cables being forgotten happens at least somewhat regularly when PC's are built or maintained, and if this invariably lead to the destruction of a new and very expensive piece of hardware, it'd surely be considered a severe design flaw.

Anyways, I tested my PC with an old card to make sure it was still fine, which it seemed to be. I then bought a 1080 second hand for 2800 SEK (~£230) which seems decent considering the shortage and other prices I've seen. So far, it is running without issue.
Having built pc's since the 80286 days personally and professionally I can assure you that there are a number of things that can be done (or undone) while building a machine that can cause smoke to roll out the side, especially when you first start it up. For any experienced builder it only happens once or twice because you become very aware of the consequences. Over the years the cabling has become much more difficult to do it with, but even some dummy plugs can be put in backwards. Just know that in cases where there are multiple power cables they usually augment the feed coming from the PCI slot, not necessary replace it entirely. It sounds like it just blew the power regulator on the GPU, so that is good.
 
The very same thing just happened to me.
Difference is - I was playing some games, PC restarted. When rebooting, my 1070 decided to catch fire all of a sudden.
This time, nothing was unplugged. Dust was below average, and it worked fine for over 3 years with the same PSU and everything else.

I actually was surprised when I saw you had the same model as me, and that it happened only a few months apart