My Experience Buying a Used GPU

I bought this used RTX 3090, a Palit GameRock edition specifically, from CeX earlier this month. By the fact that it’s sitting here on this desk and not happily purring away inside my PC should tell you how that experience went. I want to talk through what happened, what I tried, and what’s happening now not to completely ward you off buying used GPUs, but perhaps as a warning of what can happen, and what to look out for, and what to try too. So, here goes…Right off the bat something was up with this card. It reeked of cigarette smoke, yet looked pretty clean at least on the outside. That almost certainly means that it has had an attempt at cleaning, but only surface level. It hasn’t been taken apart to be cleaned. I stuck it in my PC anyway, and as soon as I loaded up a game, the fans were running at 100%. The performance improvement from the RTX 3060 I had been using wasn’t much greater either, which surprised me a fair bit, and my god the fan noise was unbearable. It was blow-your-house-down levels of noise and airflow – and then it happened. My displays shut off. I could still hear audio from a video that was playing on my second monitor, so I knew it wasn’t the system locking up or anything, it was the GPU. I hard reset the system and immediately opened HWInfo. I should have done this the second I heard the fans spinning full blast, but anyway I was shocked to find that the GPU core temperature was up at 88°c, and the GPU hotspot temperature was over 108°c, and the card was throttling like crazy. No wonder it shut itself down! This thing needs a repaste – stat! So, that’s exactly what I did. I opened up the card, carefully cleaned off the thermal paste, then put nice fresh paste in – specifically Corsair TM30, as that’s all I had on hand. I put the card back together and ran a quick Furmark test on it in the test PC I have upstairs. Temps were down to very manageable levels, frame rates were up where I’d expect them. Profit! I stuck the card back in my PC and away I went. It only took a day for that very liquid TM30 to pump its way out of the die area, and the high temps shot back up, as did the black screen crashing. I bought some Gelid GC Extreme, took the card back out and re-repasted it, making sure to cover the whole die area thoroughly. Again, the card went back together, tested with Furmark in the test PC and everything looked good, so back in my PC it went. Within three days it started crashing again, although this time not exclusively under load. Sometimes just sitting watching a film on Plex, it’d black screen. I checked the temperatures and it wasn’t always that that caused it. In fact, I noticed something strange. The card was sitting at 120W of power draw at idle – even with no core usage. While I do have three displays connected, two of which are adaptive sync displays, this seems pretty crazy. Looking at NVIDIA SMI it looks like the “minimum power” level is set at 100W, which seems weird. An RTX 3070 I’ve got also has that set at 100W, despite drawing considerably less than that at idle so perhaps that’s a red herring. Anyway, a third of the total power draw limit being pulled at idle is crazy, and means the card runs fairly hot at all times. So, back out the card comes, and I take it to the test PC to isolate if it’s the card, the drivers, Windows, or what. Before I stick it in the PC though I notice that the heatsink feels loose, so I take the card apart again, and find that the paste has migrated away from a specific point on the die – both on the die side and the heatsink side. It’s made a circle around this point, and to me that says that this part of the core is failing and running insanely hot. If this card was thoroughly abused and sold to CeX when it had an intermittent issue like this, CeX would likely never know – well if they’d actually tested it they’d know it ran at insane temps but still – and they’d happily sell it on to a schmuck like me. To me though, it’s clear this card is faulty and not something that can be repaired. Oh, and I did also update the VBIOS to the Resizable BAR supported version just to check if that helped – and it did a tiny bit, before the card assumedly started overheating again and went back to frying itself alive.Luckily I bought this from CeX, so as soon as I’m done filming, I’m heading to my local CeX store to return it. CeX have an astonishing five year warranty on items sold by them – now that isn’t quite as good as it sounds, within the first 30 days it’s a bit more no-questions-asked, within the first six months it’s a full refund, and after that it’s whatever the current CeX sale price would be, meaning if you paid, say £650 for a graphics card in 2025, but they’d sell that card for £400 two years later when it fails, they’d only give your £400 back. That’s still exceptional value compared to buying from ebay or marketplace of course, there there’s very little in the way of protections, which is the first learning point I’d hope you’d take away from my experience. Buy with the most amount of security you can – whether that’s buying from a place like CeX, or even just using a credit card or PayPal to pay for it so you have an extra layer of protection is a great idea.I’d also highly recommend as soon as you get the card in to test it fully. Load up HWInfo and Furmark to make sure the temps are sky-high, and that it doesn’t crash. The sooner you can raise an issue, be that with CeX, Ebay or the seller, the better. I’d also be careful with opening up the card as that can be a problem with some sellers – even if it’s a noble cause of repasting the card in the hopes that that saves it. To be clear though, this is likely a somewhat rare occurrence. The majority of GPUs aren’t sold because they are on their last legs, they are sold because their original buyers upgraded to something better, and there’s nothing wrong with the card. It doesn’t hurt to test it out before committing to it though.