Flabbergasted GPU repair wizard highlights dangers of liquid metal after leak kills entire RTX 5070 Ti — user-applied TIM spread to every crevice of the PCB, physically cracking and shorting out the core

Liquid metal damage on an RTX 5070 Ti
(Image credit: NorthridgeFix on YouTube)

Most GPUs today still ship with conventional thermal paste, but some higher-end cards — such as the Founder's Edition RTX 5090 — use liquid metal to maximize the cooling potential on the core. Unfortunately, if the application isn't perfect from the factory, it can lead to devastating outcomes, which is exactly what happened with an RTX 5070 Ti that landed on the desks of repair wizards Northridge Fix.

When liquid metal comes into contact with tiny surface-mounted componentry like capacitors, it will create microbridges that can short those parts. The leak can eventually reach the main power rails in worst-case scenarios, too. Not only that, but some liquid metal TIM can slowly eat away at metals like aluminum and weaken solder joints. Eventually, ICs in the affected area might even get knocked out of place.

Article continues below
Worst GPU Mess I’ve Seen RUINED by Liquid Metal - YouTube Worst GPU Mess I’ve Seen RUINED by Liquid Metal - YouTube
Watch On

In the video, we see that the liquid metal has somehow reached into every nook and cranny of the PCB, and it's impossible to spot without a microscope. The owner said they cleaned off as much as they could, but clearly, there was enough left on the board to frustrate the technician. About halfway through the cleaning process, the repairman declared that only a miracle can save this GPU at this point.

Liquid metal damage on an RTX 5070 Ti

(Image credit: NorthridgeFix on YouTube)

From the looks of it, liquid metal had gotten underneath the GPU core as well, which shorted it internally, as we can see a physical edge crack in the video. This is unfixable because you'd need to replace the chip, and that's just one part of the problem. The memory modules are affected, too, and the 1.8V power rail is shorting to ground, which means that liquid metal reached a ground pad and destroyed critical circuitry as well.

Even if the GPU core hadn’t been directly contaminated by liquid metal, the short on the 1.8 V power rail would have destroyed it regardless. So, in the end, this becomes a cautionary tale about letting enthusiasm override judgment. If the owner had just used regular, silicone-based thermal paste, he'd have a working GPU — applying liquid metal to a card not designed for it, even if you can create your own seal, is too risky.

Liquid metal damage on an RTX 5070 Ti

(Image credit: NorthridgeFix on YouTube)

Even large manufacturers producing $4,000 GPUs sometimes struggle with getting the application right. Besides, non-factory-applied liquid metal jobs rarely help lower the temperatures in a meaningful way in the first place. You'll certainly lose your warranty if you try to do it anyway, which is why Asus refused to accept the RMA request for this broken RTX 5070 Ti. Unfortunately, not even NorthridgeFix could fix it.

Google Preferred Source

Follow Tom's Hardware on Google News, or add us as a preferred source, to get our latest news, analysis, & reviews in your feeds.

Hassam Nasir
Contributing Writer

Hassam Nasir is a die-hard hardware enthusiast with years of experience as a tech editor and writer, focusing on detailed CPU comparisons and general hardware news. When he’s not working, you’ll find him bending tubes for his ever-evolving custom water-loop gaming rig or benchmarking the latest CPUs and GPUs just for fun.

  • PEnns
    Insane!

    But hey, having "Liquid Metal" in the GPU sounds so cool....
    /s
    Reply
  • flytrap23
    So it was user error.
    Reply
  • abufrejoval
    I've come to a similar conclusion with much less drastic results.

    I got myself a mobile-on-desktop Mini-ITX board as an experiment from Aliexpress some years ago, which came with a shim applied to enable a normal socket 1700 cooler to be used on top of the "naked" mobile chip.

    Benchmark results weren't great, HWinfo reported nearly constant heat related throttling, even if the measured temperatures were only a arount 60°C, well within what's silicon's comportet comfort zone.

    Tried Noctua paste and coolers instead, didn't do much to improve the situation.

    Then, since the die is naked by default and surrounded by a die carrier totally devoid of any SMD devices, but sports a bit of an elevated basin at the edge, I decided to try liquid metal, both between the die and the shim's bottom, as well as the shim's top and the CPU cooler. The board resided in one of those almost cube compact Mini-ITX cases, popular in the pre-NUC period, which has the mainboard horizontal.

    Wow, did it allow me to clock that Alder Lake i7-12700H to near desktop performance, near 100 Watts sustained without heat throttling was where I left it, since beyond that point the mainboard components were simply not designed to take the abuse.

    Adding a B580 GPU meant moving the mainboard into a small tower with a vertical mount, because I needed a beefier PSU, but I soon noticed that the CPU throttling had come back, while temperatures still weren't truly alarming.

    Taking things apart I noticed that the liquid metal had escaped, not to the point were it damaged any other components, but it evidently couldn't do the heat-transfer job any more, leaving a void behind it was supposed to fill.

    Cleaning things up showed pits on the shim's top, where its unknown metal composition, that of the cooler on top and the liquid metal itself might have had some active chemical discussions, but it also showed a small pit on the CPU die, that couldn't be good... I developed an error theory, where that small pit on the CPU die caused a thermal gap, which heated so fast to throttling, the chip's temperature sensors didn't even report the overtemp condition before the throttle took hold... Especially since it only seemed to affect P-core #2 but Alder Lake is reported to throttle globally, not per-core, on temperature.

    I tried graphene thermal pads next, hoping it might fit better into those pits and bumps. While there were no functional failures, I saw that the persistent heat issue with P-core #2 remained. Since my low-profile Noctua cooler was incompatible with the mainboard's backplate, I suspected the pads also weren't getting the pressure they needed to work properly. Inspecting the situation had the graphene pads tearing pretty near immediately so I went back to paste short-term and perhaps try a phase change material and a different cooler compatible with the backplate later.

    Got a 150 Watt rated cooler that wasn't Noctua, but seemed very much inspired by them: Most importantly it worked with the backplate installed, used Der 8auer's best paste but results stayed the same, at 45 Watts the system ran into a throttle wall.

    I finally did the phase-change transition in the midst of winter a few weeks ago and had pretty near fantastic results again, initially...

    ....only to suffer the now typical drop back to 45 Watts of effective performance, heat throttling almost constantly signalled while temperatures still never went beyond 60°C.

    But I did notice, that the BIOS had reset when I did that work, so I reset the BIOS again... and got full power!

    And the end of that long story turned out to be that I kept a applying a "FLEX override" in the BIOS, which sounded like something cool I'd want, but might have done just the opposite.

    I usually don't do a ton of manual BIOS settings, but some things like enabling resizable BAR, VMs, and XMP RAM profiles are just plain necessary, while I also tend to disable ports I don't use, keep LED and num-lock stuff off and enable advanced settings on entry. Somewhere along the search of "sensible settings", I must have come across "enable FLEX override" and it sounded like a good thing to do...

    Things I learned:

    Treat liquid metal like water cooling: not for me!

    Most likely I could have just stuck with the original paste, because it was the BIOS setting that gave such upsetting performance. Simply getting a better paste, when you have to re-apply after taking things apart, was both my original approach and what I have returned to: yonder lies silly land ...at least until you're into into several hundreds of Watts, liquid cooling and whatnot, areas were I do not tread... except if I should have to repaste a GPU.

    I left the phase change stuff in that MoD Alder-Lake, above and below the shim; can't say that it does better than paste, but it might last longer, if I can keep my fat fingers away: it doesn't even promise multiple use, just longer endurance. I might revisit phase change for a GPU, probably not before.

    I've put a big sticker on the Alder Lake chassis, to remind me to never change that "FLEX" setting again. System is doing fine at around 90 Watts sustained without throttling and then never exceeding 90°C, either. Typically it runs at much lower temperatures and Wattage, very low idle power but with a bit of muscle when needed is what I wanted it for, after all.
    Reply
  • Notton
    I never understood the appeal of liquid metal.
    It's high risk, high investment, and low reward.

    Just use PTM7950 and call it a day.
    Thermalright TFX, or Iceberg FuzeIce plus also lasts a long time on direct die.
    Reply
  • Eximo
    Only time I have de-lidded I used Thermal Grizzly kryonaut under the lid. The real gains came from reducing the TIM gap (during that period where Intel wasn't soldering)
    Reply
  • Gururu
    is it even worth to send sub 1k cards to Northridge? What is their ballpark hourly?
    Reply
  • EzzyB
    Never tried it, but I had no idea that, in addition to shorting contacts if it got to them, it is actually corrosive to the board itself if it gets on it. Full pass, I'm simply not that precise and I know my limitations! :D
    Reply
  • hotaru251
    people think like typical paste you need a lot of liquid metal...and hat is the issue.

    you NEVER want a lot of LM on your surface. if it looks wet you have too much. you apply a VERY thin bit to the surface of the die/heatspreader and the cpu/gpu cooler....they when put together form a single layer via surface tension & there is very little risk of it ever leaking.

    for laptops you want more security (as they are banged around which doesnt happen in desktops) and why they come w/ padding or other ways to hopefully "contain" any leakage.
    Literally watch someone like der8aur to see how little is actually used.
    Reply
  • bit_user
    hotaru251 said:
    when put together form a single layer via surface tension & there is very little risk of it ever leaking.
    I'm not sure about that. I think it's still susceptible to pump-out.

    As for what happened in this case, the extent of the leakage suggests the user applied grossly excessive amounts of LM, rather than what you describe.
    Reply
  • SkyBill40
    That's a painful and costly lesson to learn which, had they not chosen to go down a risky path, could have been avoided.
    Reply