[Solved] RX 460 Blown Fuse

Kzoak

n00b
Joined
Dec 1, 2019
Messages
24
Hi everyone!
I'm new here and I come because I hope you can help me.
I have Gigabyte RX 460. Few months ago it stopped working. I didn't have time to check what's really going on.
It doesn't work and isn't detected at all. I checked that fans are still working (if you give them some juice directly). I checked the fuse that is next to PCI - is dead. It may be this one element (or more), though it's a good starting point. But the problem is that I have no idea what kind of fuse is this? Can you help me and tell what fuse should I buy?
My assumptions are that it's 1206 SMD element and because this card is powered only through PCI it's something like 5A or 6,3A (*12V = 60-75W - what in theory PCI can provide). Though I'm not so sure if I'm thinking correctly.
This fuse is called F3 in the picture and there're no signatures on it.

rx460_blownfuse.jpg
 
I would get a 75w plus 10% fuse, (75w/12v=6.25a plus 10% would be around 6.75a) new fuse holder and solder directly to current burnt out fuse with wires. It will now be easily replaceable. Also if you put a loop in the wire to the fuse holder (fuse holder can be anywhere, just wired up) you should be able to take an amp reading. I would not waste time replacing the current fuse unless you feel comfortable in doing that.

If you can RMA the card, it might just be better to do that.
 
  • Like
Reactions: Kzoak
like this
I would get a 75w plus 10% fuse, (75w/12v=6.25a plus 10% would be around 6.75a) new fuse holder and solder directly to current burnt out fuse with wires. It will now be easily replaceable. Also if you put a loop in the wire to the fuse holder (fuse holder can be anywhere, just wired up) you should be able to take an amp reading. I would not waste time replacing the current fuse unless you feel comfortable in doing that.

If you can RMA the card, it might just be better to do that.

Thank you for the answer! The idea with the fuse holder is great. I didn't think about it.
I can't RMA it because it died ~3 months after guarantee period (how convenient). Fortunately I feel comfortable with soldering.

Thanks again!
 
Is it me, or is C155 completely missing? Maybe not needed, I suppose.
 
Thank you for the answer! The idea with the fuse holder is great. I didn't think about it.
I can't RMA it because it died ~3 months after guarantee period (how convenient). Fortunately I feel comfortable with soldering.

Thanks again!
GOing to solder iron or hot air solder?
 
GOing to solder iron or hot air solder?

Iron. I don't have hot air :(
I already soldered it. Checked the connection and it looks fine (beeps ;)) Unfortunately - it's still dead. Fans don't even bother starting. I don't have knowledge nor equipment to push it further. Though that would be an interesting journey to repair it. Shame. I liked this card (though I don't like Gigabyte).

Let me show off my very bad soldering skills with what I had on hand (and I couldn't get any kind of a fuse holder).
 

Attachments

  • IMG_20200214_130722.jpg
    IMG_20200214_130722.jpg
    336.9 KB · Views: 0
Thank you for the answer! The idea with the fuse holder is great. I didn't think about it.
I can't RMA it because it died ~3 months after guarantee period (how convenient). Fortunately I feel comfortable with soldering.

Thanks again!
can still ask if they will repair it. That said, its an rx460, might be time to replace it.
 
Hi everyone!
I'm new here and I come because I hope you can help me.
I have Gigabyte RX 460. Few months ago it stopped working. I didn't have time to check what's really going on.
It doesn't work and isn't detected at all. I checked that fans are still working (if you give them some juice directly). I checked the fuse that is next to PCI - is dead. It may be this one element (or more), though it's a good starting point. But the problem is that I have no idea what kind of fuse is this? Can you help me and tell what fuse should I buy?
My assumptions are that it's 1206 SMD element and because this card is powered only through PCI it's something like 5A or 6,3A (*12V = 60-75W - what in theory PCI can provide). Though I'm not so sure if I'm thinking correctly.
This fuse is called F3 in the picture and there're no signatures on it.

View attachment 223120
That looks like a 3216 package SMD, although I'd have to measure to be sure. I'd guess you want something like a 7A, 24VDC rated part (12V parts don't seem to exist in this size). Plenty of options on Digi-Key and mouser. It's hard to say what an exact replacement would be without seeing the part in person. I'd be surprised if it really has no markings at all, but I can't see any in the picture.

You have, potentially, two problems here. One is that the fuse is apparently open. The other is that you may have a short through the board or something elsewhere. Check the major voltage rails for resistance to ground before you go replacing the fuse or applying 12V to the card - this was apparently a semi-common issue on the RX 400 series cards, but if you haven't fixed whatever caused it to fail, you'll just damage the card worse.

If you don't feel like repairing it yourself, I might be interested in buying this card from you. I make threads (and videos for Youtube) about repairing dead graphics cards from time, and I'm always on the lookout for new ones to work on. I'm sometimes even successful.
 
can still ask if they will repair it. That said, its an rx460, might be time to replace it.

I'm not playing games. This card was too powerful for my needs. I used it to calculate stuff. They definitely won't fix it after my soldering :)

The story of my Gigabyte RX460:
There was a card. Graphics card. It didn't have hard time. It played Witcher 3 from time to time and calculated mathematical stuff. It liked Linux and Linux liked it - no drivers necessary (starting with the Kernel of the number I don't remember). But it wasn't hanging only with Linux. There was another bud - Windows. They were hanging together quite a lot.
Everything went fine, new drivers were stable. I "underclocked" it by 10% with Wattman settings and the card never exceeded 40°C and was using very little juice. Very shy and quiet type.
But... because AMD's new drivers weren't able to correctly handle OpenCL 1.2 that I needed. I had to install the latest, but "no so stable" version of drivers that handled OpenCL 1.2 correctly (WHQL 17.9.3 to be very specific). With those drivers my card worked only fine, because my "underclocking" settings defaulted by themselves from time to time with a windows notification about something going wrong about the driver and setting it to default.
So one day I had this task for my card. "You're going to calculate this for me while I'm hanging with my friends". And so it began. Room temperature on this day 28°C. Time to complete the task ~3-4 hours. I was back in about 7h and the screen was blank. The system still worked. Restart... nothing. Look at the card and very fast diagnosis - dead.
What happened? No idea. Is it my fault that I was calculating stuff on it? Probably.

That looks like a 3216 package SMD, although I'd have to measure to be sure. I'd guess you want something like a 7A, 24VDC rated part (12V parts don't seem to exist in this size). Plenty of options on Digi-Key and mouser. It's hard to say what an exact replacement would be without seeing the part in person. I'd be surprised if it really has no markings at all, but I can't see any in the picture.

Maybe it's under the fuse. I would need to desolder it, though hard to do with just an iron solder.

You have, potentially, two problems here. One is that the fuse is apparently open. The other is that you may have a short through the board or something elsewhere. Check the major voltage rails for resistance to ground before you go replacing the fuse or applying 12V to the card - this was apparently a semi-common issue on the RX 400 series cards, but if you haven't fixed whatever caused it to fail, you'll just damage the card worse.

That's a good next step to do. Check resistances.
I couldn't find any really helpful guidance that would inform about what to do / check in the failed GPU.

If you don't feel like repairing it yourself, I might be interested in buying this card from you. I make threads (and videos for Youtube) about repairing dead graphics cards from time, and I'm always on the lookout for new ones to work on. I'm sometimes even successful.

I would love to help you with new "dead bodies" for autopsy, but I'm the kind of person that would spend $100 trying to repair this card, rather than spending $100 on a new one and be done with it :) I'm living my life to learn.
And I'm from Europe anyway. This card would need to fly quite a distance ;)

Because this GPU gave me a reason to actually buy hot air solder I would like to try and repair it. Successfully or not - that's a great way to learn :) I'm into electronics anyway. But programming AVR/ARM isn't really that hard and didn't require me to have SMD soldering skills.

I have some pics of the suspicious spots on this card. I cleaned these places. Now when I think then yeah, probably at least one element need replacement.
4 pins are from the fans connector and 2 pins belong to the capacitor that I guess is 820uF and 25V (or 2,5V).
 

Attachments

  • IMG_20190629_183905.jpg
    IMG_20190629_183905.jpg
    671.9 KB · Views: 0
  • IMG_20190629_183918.jpg
    IMG_20190629_183918.jpg
    690.2 KB · Views: 0
  • IMG_20190629_184815.jpg
    IMG_20190629_184815.jpg
    393.6 KB · Views: 0
  • IMG_20190629_184914.jpg
    IMG_20190629_184914.jpg
    444.3 KB · Views: 0
  • Like
Reactions: noko
like this
Just so we're clear, you have maybe a 25% chance of success here, under reasonably favorable circumstances. I don't mean that to sound condescending, but it's just that this mode of failure usually seems to be the result of a short through the PCB itself, which is frustratingly difficult to diagnose, and nearly impossible to repair in any way that leaves the card actually usable, even if you get it "working."

Now... You need to take some resistance measurements between a few spots on the board. I strongly suspect that your new fuse is now blown as well, which suggests that you haven't cleared the short.

I've marked up your photos to indicate where you need to take the measurements. Results of this test will tell us which circuit on the board is actually subject to the short.

460-1.jpg460-3.jpg

1. 12V input across your new fuse. 0 ohms is good here, but I suspect you'll see an open circuit.
2. 12V input to ground - You're looking for a big-ish number here. Most AMD cards I've looked at show a few thousand.
3. GPU VCore - You should see a low number here - something like 10 ohms.
4. Memory - The memory itself has its own power separate from the controller. You're looking for like 50-200 ohms here.
5. Memory Controller - Of the five VRM phases in the row there by display connectors, it looks like the one furthest from the PCI-E slot connector is the memory controller. 40-100 ohms is sane here.
6. 3.3V - You're looking for about 350 ohms here, as I recall. The exact value doesn't matter to us for now, as long as it's not zero.
7. Minor rails - These little 8 pin ICs are often controllers for smaller VRMs that produce various other voltages that the card needs to function. I can't read the part numbers on them, so I can't tell you for sure which pins are the outputs, but it's usually pins 7 and 8. At least one of them is a BIOS ROM chip (looks like U10). We'll assume they're not the issue for now, but you'd be looking for a few thousand ohms on each one if you can figure out which pins are the outputs.
8. 5V minor VRMs - a couple thousand ohms.

Take those measurements and report back. Do not attempt to run the card again until you figure out why the original fuse failed. You risk further damage, if you do. I can't promise I've properly identified all the rails, since I'm just looking at photos, but you should be able to get in the ballpark this way.

In your second two photos, the first looks like a glob of flux residue, but could also be a scorch mark on the board, indicating a short in that area - hard to say from the photo. The one with the fan connector just looks like corrosion, but it's hard to say. Can you post some more photos of that area of the board, from both sides? Was this card used when it came to you?
 
Just so we're clear, you have maybe a 25% chance of success here, under reasonably favorable circumstances. I don't mean that to sound condescending, but it's just that this mode of failure usually seems to be the result of a short through the PCB itself, which is frustratingly difficult to diagnose, and nearly impossible to repair in any way that leaves the card actually usable, even if you get it "working."

Now... You need to take some resistance measurements between a few spots on the board. I strongly suspect that your new fuse is now blown as well, which suggests that you haven't cleared the short.

I've marked up your photos to indicate where you need to take the measurements. Results of this test will tell us which circuit on the board is actually subject to the short.

View attachment 223360View attachment 223361

1. 12V input across your new fuse. 0 ohms is good here, but I suspect you'll see an open circuit.
2. 12V input to ground - You're looking for a big-ish number here. Most AMD cards I've looked at show a few thousand.
3. GPU VCore - You should see a low number here - something like 10 ohms.
4. Memory - The memory itself has its own power separate from the controller. You're looking for like 50-200 ohms here.
5. Memory Controller - Of the five VRM phases in the row there by display connectors, it looks like the one furthest from the PCI-E slot connector is the memory controller. 40-100 ohms is sane here.
6. 3.3V - You're looking for about 350 ohms here, as I recall. The exact value doesn't matter to us for now, as long as it's not zero.
7. Minor rails - These little 8 pin ICs are often controllers for smaller VRMs that produce various other voltages that the card needs to function. I can't read the part numbers on them, so I can't tell you for sure which pins are the outputs, but it's usually pins 7 and 8. At least one of them is a BIOS ROM chip (looks like U10). We'll assume they're not the issue for now, but you'd be looking for a few thousand ohms on each one if you can figure out which pins are the outputs.
8. 5V minor VRMs - a couple thousand ohms.

Take those measurements and report back. Do not attempt to run the card again until you figure out why the original fuse failed. You risk further damage, if you do. I can't promise I've properly identified all the rails, since I'm just looking at photos, but you should be able to get in the ballpark this way.

In your second two photos, the first looks like a glob of flux residue, but could also be a scorch mark on the board, indicating a short in that area - hard to say from the photo. The one with the fan connector just looks like corrosion, but it's hard to say. Can you post some more photos of that area of the board, from both sides? Was this card used when it came to you?

That's very helpful! I've read your thread about resurrecting dead GPUs and it is really great, with a lot of useful information! I finally start to understand what's going on.
Because it's late in here, I'm going to get some rest and make measurements tomorrow.
You're right about the new fuse - it's dead too. It was a good fuse so I hope it did its job and protected the circuit. I'm not expecting to fix this card but I want to try it anyway, because as I said - I love to learn and because of this failed GPU I already learned a lot, and there's more. Much more :)
I got the card as brand new. No one used it before me. The air in here is dry/very dry, though I can't ensure that this card didn't get a drop of water while doing something with / next to the PC. Though it's not likely.
I don't have any more photos because I cleaned these spots with alcohol and they look brand new now. These are the only (old) pics I got.

Stay tuned
 
  • Like
Reactions: noko
like this
That's very helpful! I've read your thread about resurrecting dead GPUs and it is really great, with a lot of useful information! I finally start to understand what's going on.
Because it's late in here, I'm going to get some rest and make measurements tomorrow.
You're right about the new fuse - it's dead too. It was a good fuse so I hope it did its job and protected the circuit. I'm not expecting to fix this card but I want to try it anyway, because as I said - I love to learn and because of this failed GPU I already learned a lot, and there's more. Much more :)
I got the card as brand new. No one used it before me. The air in here is dry/very dry, though I can't ensure that this card didn't get a drop of water while doing something with / next to the PC. Though it's not likely.
I don't have any more photos because I cleaned these spots with alcohol and they look brand new now. These are the only (old) pics I got.

Stay tuned
This thread is probably more relevant. The RX460 is an evolution of the 290X, and they share a lot of DNA, including the set of voltage inputs they need.
https://hardforum.com/threads/graphics-card-necromancy-continued-radeon-r9-290x.1982565/
 
Here are my measuring results. These "arrows" that you've drawn really helped.

Measurements:
1. Across the fuse - infinity (of course)

2. 2-3 ohms - sooo... that's not good

3. GPU VCore - ~3 ohms

4. Memory - ~56 ohms

5. Memory controller - OK - ~35 ohms

6. 3.3 V - non-zero (30 ohms)

7. Other:

RT8120D GSP3GJ2N - corner - I don't think I really understand the spreadsheet. Looking at the diagram I would say that only pin 8 is the output and the rest is used to drive it?
Pin 8 - 60 ohms
Pin 7 - Infinity
Pin 6 - ~650
UP0104P SGA51N - I found so many datasheets but all of them are useless and not about this IC o_O
Pin 8 - ~1ohm
Pin 7 - infinity
Pin 6 - ~1k

8. 5V VRMs - UTC 78D05AL - ~620 both


Looking for spreadsheets I found someone who had problem with UP0104P VRM and was looking for replacement for his Radeon 7870.
 
  • Like
Reactions: noko
like this
Here are my measuring results. These "arrows" that you've drawn really helped.

Measurements:
1. Across the fuse - infinity (of course)

2. 2-3 ohms - sooo... that's not good

3. GPU VCore - ~3 ohms

4. Memory - ~56 ohms

5. Memory controller - OK - ~35 ohms

6. 3.3 V - non-zero (30 ohms)

7. Other:

RT8120D GSP3GJ2N - corner - I don't think I really understand the spreadsheet. Looking at the diagram I would say that only pin 8 is the output and the rest is used to drive it?
Pin 8 - 60 ohms
Pin 7 - Infinity
Pin 6 - ~650
UP0104P SGA51N - I found so many datasheets but all of them are useless and not about this IC o_O
Pin 8 - ~1ohm
Pin 7 - infinity
Pin 6 - ~1k

8. 5V VRMs - UTC 78D05AL - ~620 both


Looking for spreadsheets I found someone who had problem with UP0104P VRM and was looking for replacement for his Radeon 7870.
Your GPU and memory resistance looks to be in the range of normal. The 2-3 ohms on #2 is your problem. That's basically a dead short to ground on the card's power supply. The short could be in a number of different places, but the most likely are:
The PCB itself, in the vicinity of one of the high-side MOSFETs
The power MOSFETs on the Vcore rail (We know it's not the memory or controller rails because they're not shorted)
One of the capacitors on the 12V rail

The PCB itself seems to be the most common culprit, in my experience. If this is the case, it's really not worth the effort trying to fix this card, because the PCB is the one component you can't replace. There are a few ways of diagnosing of exactly which one it is, though. The simplest is to supply a current through the board at the operating voltage and amperage, and look for components that get warm. If you find one, you remove it and then recheck your resistance. Keep in mind, it's possible to have more than one failed component. You you need a power supply that allows you to set both voltage and amperage limits, though - an ATX power supply won't work.

Another option is to just start removing the components that are connected to the 12V rail and hope you get lucky. Experience tells me that this doesn't usually work, though. Keep in mind that if you hot air a component off the board, and the board is still hot, most of your measured resistances will be lower than they would be at room temperature. The GPU die will appear to be a dead short until it cools down.
 
  • Like
Reactions: noko
like this
Your GPU and memory resistance looks to be in the range of normal. The 2-3 ohms on #2 is your problem. That's basically a dead short to ground on the card's power supply. The short could be in a number of different places, but the most likely are:
The PCB itself, in the vicinity of one of the high-side MOSFETs
The power MOSFETs on the Vcore rail (We know it's not the memory or controller rails because they're not shorted)
One of the capacitors on the 12V rail

The PCB itself seems to be the most common culprit, in my experience. If this is the case, it's really not worth the effort trying to fix this card, because the PCB is the one component you can't replace. There are a few ways of diagnosing of exactly which one it is, though. The simplest is to supply a current through the board at the operating voltage and amperage, and look for components that get warm. If you find one, you remove it and then recheck your resistance. Keep in mind, it's possible to have more than one failed component. You you need a power supply that allows you to set both voltage and amperage limits, though - an ATX power supply won't work.

Another option is to just start removing the components that are connected to the 12V rail and hope you get lucky. Experience tells me that this doesn't usually work, though. Keep in mind that if you hot air a component off the board, and the board is still hot, most of your measured resistances will be lower than they would be at room temperature. The GPU die will appear to be a dead short until it cools down.

Removing components on the 12V rail to find a failing one doesn't sound like fun. So I would need some proper power supply to look for hot components and then hot air solder to replace them. I was thinking about power supply and hot air for some time now. I will need to look on the Internet if there's some power supply (probably not a laboratory one) + hot-air that I can afford. But it can end up with a dead-end as I could not afford two of those things at once.
 
  • Like
Reactions: noko
like this
Removing components on the 12V rail to find a failing one doesn't sound like fun. So I would need some proper power supply to look for hot components and then hot air solder to replace them. I was thinking about power supply and hot air for some time now. I will need to look on the Internet if there's some power supply (probably not a laboratory one) + hot-air that I can afford. But it can end up with a dead-end as I could not afford two of those things at once.
A lab power supply is exactly what you need. You need to be able to set both the voltage and amperage limit, or you'll just damage the card worse. You can generally make do with a cheap one, but you do need it to do basic lab power supply things. You'd obviously need hot air to remove SMD components. I'd set this card aside until I could get my hands on all of the tools if I were you, but the hot air station is what I'd get first if I had to choose. The power supply is only used for fairly specfic diagnostics.
 
  • Like
Reactions: noko
like this
A lab power supply is exactly what you need.

My fault. By saying "not a laboratory one" I put my thoughts without explanation that I don't have money for a power supply like this. Even cheap ones are expensive.
But as they say - Necessity is the mother of invention. You said ATX power supply won't suffice. BUT I've found a great step-down module based on XL4016 chip. It can be driven by ATX power supply. It can do up to 8A easily. 1,5-32V no problem. Very good efficiency and on top of all of that it doesn't cost $100, but $15 - for a better circuit with Voltage and Amperage reading and ability to set them both. Too good to be true? I've read about it and it's real :) I think I could get a working setup soon ;)
 
  • Like
Reactions: noko
like this
For a somewhat cheaper CC/CV power supply, I'd recommend this one: https://www.aliexpress.com/item/4000282551930.html
The RD6006 + S06A case + 60V power supply shown on that page is a good combo. It's not linear, but is good enough for this type of work.
Comes to about $120 US total for all 3 parts. You assemble yourself - it's easy. Since you are already a techie, you'll find it fun. I did.
 
FANTASTIC THREAD!

Some of us would ditch the card but fixing it does give a lot of knowledge and seeing it in action is very cool! Usually after an electrical component potentially failed as in a piece of controlled equipment. I try to smell everything I can, touch everything I can (of course de-energized, capacitors discharged etc.), look for anything burnt. This usually can lead to what is causing the issue and any subsequent collaqual damage hopefully. The fuse should have protected most components. Actually supplying current flow to unknown cause other than a power rail going to ground somewhere I've never done before, which hopefully leads to the faulty component(s) becoming obvious without causing any further damage.
 
You need to be able to set both the voltage and amperage limit

I have the equipment, though I'm not certain what's the best way to supply this card. Should I buy some PCI raiser and mod it to have ability to put in 12V in there? Or is there some equipment I couldn't find that I can plug in the GPU and test it?
I could always do it dirty way and solder + and - to the power supply ;)

When I'll get down to test the card. I suppose I should start with 12V and give it as little amperes as possible at the beginning. Have the circuit shorted with a fuse "plugged in" and give it some time to warm up. What's the good starting amount of Amperes? 500-1000 mA?

FANTASTIC THREAD!

Some of us would ditch the card but fixing it does give a lot of knowledge and seeing it in action is very cool! Usually after an electrical component potentially failed as in a piece of controlled equipment. I try to smell everything I can, touch everything I can (of course de-energized, capacitors discharged etc.), look for anything burnt. This usually can lead to what is causing the issue and any subsequent collaqual damage hopefully. The fuse should have protected most components. Actually supplying current flow to unknown cause other than a power rail going to ground somewhere I've never done before, which hopefully leads to the faulty component(s) becoming obvious without causing any further damage.

Yes, I've already learned a lot and I think there's still more coming :)
I would love this card to work - just for science of the whole operation.
 
  • Like
Reactions: noko
like this
I have the equipment, though I'm not certain what's the best way to supply this card. Should I buy some PCI raiser and mod it to have ability to put in 12V in there? Or is there some equipment I couldn't find that I can plug in the GPU and test it?
I could always do it dirty way and solder + and - to the power supply ;)

When I'll get down to test the card. I suppose I should start with 12V and give it as little amperes as possible at the beginning. Have the circuit shorted with a fuse "plugged in" and give it some time to warm up. What's the good starting amount of Amperes? 500-1000 mA?

I think the best way to do this is to solder at least the positive lead directly to the card. The pad for the fuse is a good place. It's best to solder the ground as well, but you can use an alligator clip on the IO plate if you can't find a place to solder it to, keeping in mind that it needs to be a pretty big pad. Don't be afraid to remove components from the card when you do this test and solder the leads to their pads. The 12V caps are a good option, although if you use through-hole pads, putting the capacitor back becomes a pain if you can find and clear the short.

You will likely also need some sort of indicating medium to actually tell where the short actually is. You can use isopropanol for this, but obviously need to be careful because it's quite flammable. The proper tool is refrigerant cold spray, but I actually think isopropanol works a little better. When it works, this method makes the short pretty obvious, but I'd say that you have about a 75% chance of the short being in the board itself, and not a component. If that's the case, you're going to do all this, and you will not obviously have a hot spot on the board. If that happens, it's time to give up and use this card for parts to fix others.
 
Let me give you some update. So I've searched for hot spots on the card but didn't find anything.
Probably I'll really need lab power supply - because this buck converter probably can't do what I want ;d (I power it with a laptop power supply because they're usually ~19V and 3-5A).
At 2A the core have gotten warm. I couldn't find anything warm nor hot. I tried with isopropanol to see something but no luck. At 4A I could hear the card squeaking and it didn't have any cooling so I didn't want to push it too far.
 

Attachments

  • IMG_20200228_160254.jpg
    IMG_20200228_160254.jpg
    419 KB · Views: 0
  • IMG_20200228_160302.jpg
    IMG_20200228_160302.jpg
    218.9 KB · Views: 0
  • IMG_20200228_162025.jpg
    IMG_20200228_162025.jpg
    612.9 KB · Views: 0
If the core is getting warm, that's not good. It means your short is likely one of the high side mosfets, which is now directly connecting the 12V input to the GPU VCore rail. You can check resistance with a multimeter to confirm this. I bet you see 3 ohms or less.
 
Do you use the IPA to look for it boiling/bubbling somewhere? Neat trick. I'm in the semiconductor industry and have used lock-in thermography before for find shorts. https://www.optotherm.com/software-lock-in.htm Basically infrared camera on a microscope. Most won't have access to that, though.

Are the high-side mosfets the VRMs or are they somewhere else?
 
Do you use the IPA to look for it boiling/bubbling somewhere? Neat trick. I'm in the semiconductor industry and have used lock-in thermography before for find shorts. https://www.optotherm.com/software-lock-in.htm Basically infrared camera on a microscope. Most won't have access to that, though.

Are the high-side mosfets the VRMs or are they somewhere else?
The high-side mosfets are a part of the VRM. They're used to switch the 12V input power on and off, so that, when smoothed out by the coils, you get an average of whatever the Vcc for that circuit is - usually 1.0 to 1.5v on a graphics card. When they fail, you end up with basically a wire going straight from the 12V input to output of the VRM, such that you have 12V being sent to GPU, which is about 8X what it's supposed to have. When this happens, it usually kills the core.
 
Hi guys!

I had the same problem with my RX 460 from Gigabyte and I want to share my experience.
I got the card working by swapping high-side mosfet that was shorted and by replacing a blown fuse which is the origin of this topic.
It was running fine for 2-3 days and then suddenly today my PC shut itself off on idle and the issue is back. The same mosfet is shorted and the fuse is blown.
I'm guessing the real cause of the problem is hiding somewhere else - maybe one of the ICs? I have one last try left with this GPU because I bought 2 fuses and 3 mosfets.
What do you guys think gone bad?
There might be a similar situation with this GPU.
 
  • Like
Reactions: Kzoak
like this
Hi guys!

I had the same problem with my RX 460 from Gigabyte and I want to share my experience.
I got the card working by swapping high-side mosfet that was shorted and by replacing a blown fuse which is the origin of this topic.
It was running fine for 2-3 days and then suddenly today my PC shut itself off on idle and the issue is back. The same mosfet is shorted and the fuse is blown.
I'm guessing the real cause of the problem is hiding somewhere else - maybe one of the ICs? I have one last try left with this GPU because I bought 2 fuses and 3 mosfets.
What do you guys think gone bad?
There might be a similar situation with this GPU.

Is it the same phase that's failing every time? I would guess a failed bootstrap capacitor, but low voltage from the power supply could also conceivably cause that.
 
Is it the same phase that's failing every time? I would guess a failed bootstrap capacitor, but low voltage from the power supply could also conceivably cause that.
Yes, exactly the same phase is failing. PSU is not the problem here as the first time it happened in a different PC. Where is a bootstrap capacitor located?
 
Yes, exactly the same phase is failing. PSU is not the problem here as the first time it happened in a different PC. Where is a bootstrap capacitor located?
https://www.renesas.com/us/en/produ...e-output-buck-controllers/device/ISL6545.html

If you look there, you'll see a pretty common single phase VRM design. The bootstrap capacitor is CBoot. One side is connected to the phase output (between the MOSFETs and the coil), and the other is connected to the controller's bootstrap pin.

Remember that the core VRM on this card has four phases, each with its own phase controller, so you'll need to figure out which pin on the phase controller is the bootstrap pin and use that to determine which capacitor nearby is the bootstrap capacitor.
 
bootstrap.png

I found a bootstrap pin in the datasheet of the gate driver on my card. Nothing there on a problematic phase, where at the other phases there is around 2.4 ohm. So that capacitor probably is dead or missing, now I have to figure out where it is.
 
  • Like
Reactions: Kzoak
like this
View attachment 227457
I found a bootstrap pin in the datasheet of the gate driver on my card. Nothing there on a problematic phase, where at the other phases there is around 2.4 ohm. So that capacitor probably is dead or missing, now I have to figure out where it is.

If you have zero ohms between BST and the phase output in the circuit shown in your picture, I think there's at least as good a chance of a failed phase control IC as there is a failed capacitor.

Edit: If you have a short to ground on that pin, it's almost certainly the IC, and not the capacitor.

Edit2: Looked again. If you have zero ohms between BST and the phase output, try removing the bootstrap capacitor and see what you get. If you still have zero, it's the IC, not the cap.
 
Last edited:
If you have zero ohms between BST and the phase output in the circuit shown in your picture, I think there's at least as good a chance of a failed phase control IC as there is a failed capacitor.

Edit: If you have a short to ground on that pin, it's almost certainly the IC, and not the capacitor.
No short to ground on that pin and also no connection between BST and phase output.
 
How bad would it be to desolder this one mosfet and try to run the card on 3 phases?

EDIT:
From what I've learned trying to get more information about how to repair a GPU, looks like failing MOSFETs is the second most common problem after failing capacitor (at least it looks that it's like this).
I'll try to check MOSFETs on my card once I get more time to do so. Anyway I've seen people desoldering mosfets and running their cards "just fine".
 
Last edited:
Are the zero ohm resistors shown in the schematic actually on the board?
There are resistors and capacitors on the board close to every driver but they are not connected to BST. I checked the resistance between Drain and BST again and I realized I was wrong - for every phase even broken one there is ~256 ohms. BST is also not shorted to ground as I mentioned before.

How bad would it be to desolder this one mosfet and try to run the card on 3 phases?

EDIT:
From what I've learned trying to get more information about how to repair a GPU, looks like failing MOSFETs is the second most common problem after failing capacitor (at least it looks that it's like this).
I'll try to check MOSFETs on my card once I get more time to do so. Anyway I've seen people desoldering mosfets and running their cards "just fine".
I saw people doing that too. It will probably work, but I think you would need to underclock and undervolt the card after that.
 
How bad would it be to desolder this one mosfet and try to run the card on 3 phases?

EDIT:
From what I've learned trying to get more information about how to repair a GPU, looks like failing MOSFETs is the second most common problem after failing capacitor (at least it looks that it's like this).
I'll try to check MOSFETs on my card once I get more time to do so. Anyway I've seen people desoldering mosfets and running their cards "just fine".

Short answer is probably not, for a couple of reasons.

First, many of the GCN designs have a voltage controller that is smart enough to tell that one phase isn't running at all. Exactly what it does when this happens, I don't know, but there's a decent chance that it will just not work at all. At best, it's problematic.

Second, with only four phases, you're down 25% of your power supply to the core. If you don't do a lot of compute heavy tasks with this card, you may get away with it, but it sort of defeats the purpose of having a powerful GPU if you can't use it, and you can obviously forget about overclocking.
 
I removed the dead mosfet from the GPU and started checking for shorts and I found that most of the capacitors on the VRM are shorted.
SHORTED.png
 
I removed the dead mosfet from the GPU and started checking for shorts and I found that most of the capacitors on the VRM are shorted.
View attachment 228382
This just means you still have A short somewhere. How did you determine which MOSFET it was that needed to be removed?

If by shorted, you mean that you have 0 ohms to ground between the positive and negative side of those capacitors, I would try removing the corresponding low side MOSFET on the suspect phase as well. Failing that, you need to hunt for the short, which is hard to do without pretty sophisticated lab equipment.
 
This just means you still have A short somewhere. How did you determine which MOSFET it was that needed to be removed?

If by shorted, you mean that you have 0 ohms to ground between the positive and negative side of those capacitors, I would try removing the corresponding low side MOSFET on the suspect phase as well. Failing that, you need to hunt for the short, which is hard to do without pretty sophisticated lab equipment.
I determine which mosfet needs to be removed by checking for a gate-source short.
I will try removing low side mosfet.
 
Back
Top