7900XTX repair video

For repair, please contact me by following the link in the channel ABOUT page.
Buy me a candy at paypal.me/tonynameless
Tools, schematics, boardview files etc are available here
drive.google.com/drive/folder...

[@borisdg]

VEGA wasn't chiplet design. It was monolithic SoC but using HBM2 memory. Here is chiplet design with regular GDDR6.

[@MrDvneil]

Is kinda the same, both cases have interposer, lines that are not in the board,

[@Revoku]

vega might have been a monolithic gpu die, but it had HBM2 chiplets that aren't replaceable on it that failed in the same way mr pedantic

[@hinz1]

Time to learn reballing chiplets, lol!

[@northwestrepair]

lol If i could do that, i would become GPU repair god.

[@FNGecko]

But you are the GPU repair god! Want some vintage cards donated? 😅

[@NVMDSTEvil]

@@northwestrepair should try taking apart a known bad chiplet gpu for fun to see what it looks like

[@Vo0d0o2009]

you'Re allready a GPU Repair GOD, @@northwestrepair

[@draakevil]

From micro-soldering to nano-soldering :D

[@AladimBR]

I got a Sapphire 7900XTX vanilla during Black Friday. Installed the support, undevolted and limited its boost (running 1440p). I’m keeping hotspot under 75C, not moving desktop at all. No issues so far. If I have to move it in the future, definitely taking the card out.

[@chubby_bear]

Vanilla? Anyway its MBA model? Which Is your limit of boost. My mba touches High junction temps but Is working fine

[@dimmer2211]

My Vapor-X was unstable when it randomly shot up to 3.2Ghz (Core.. not Frontend) at stock. and having two 1440p 280Hz Displays was awesome! just a bit under 100W at idle 24/7...

[@albertovilela9015]

I got a Sappire Pulse RX 7900 XT 20GB 6 month it this grafic card + 7800X3D same temperature 76C Max HotSopt to my PC is all time on is Micro ATX my build ... Sapphire are good atm no problems last grafic card as a Sapphire Nitro+ RX 6800 16GB is a nice grafic card to + Asrock B650M PG Riptide + Ryzen 7 7800X3D + Sapphire Pulse RX 7900 XT 20GB + G.SKill Trident Z5 Neo DDR5 6000MT's CL30 2x16GB=32GB Com Perfil EXPO 6200MT's + Power Suply MSi MPG A850GF 850Watts 80+Gold Full Modular + WaterColler Corsair Hydro H100X 240mm + Case Mars Gaming MC-MPRO + 5fans NOX HUMMER H LEDred

[@AladimBR]

It's the Pulse, the entry level model. I'm limiting to 2200Mhz. I know it can go much higher, but I don't need it and want to keep power output under 300W. If left unchecked, it peaks 400-450W, which is crazy and makes no sense..... I don't need high FPS numbers, I aim at a good average for a smooth gameplay. My 1440p monitor refresh rate is 75Hz, and I wanted the card for the extra memory. I'm no fanboy (my last 3 GPUs were Nvidia, and before was AMD), the way I see it at the moment, I don't want to support Nvidia with their pricing strategy while offering less. I went AMD because it was way cheaper, much more memory and I don't care for ray tracing.@@chubby_bear

[@naomiarmitage8729]

if you need to undervolt and limit boost, im sorry but your gpu is faulty. A stock device is intended to work as is ,without user intervention :/ also 7900 xtx had a funny radiator fail and half of them were DOA cause they reached top temps after a few mins gaming :(

[@mtnmosin2740]

Note to self, don't throw or drop high end expensive stuff, it does not like to be abused.

[@RFC3514]

Who would have guessed, eh? I slammed my Lamborghini into a wall and it stopped working. I took it to a mechanic who fixes a lot of Ferraris and he said "Lamborghinis are crap". It's strange why he's always having to fix Ferraris, though.

[@viigraphics]

True that...

[@johntet]

@@RFC3514 It's chiplet shenanigans . There is no proper car equivalent cause an engine has replaceable parts whereas a chip doesnt. But for the sake of it imagine the engine block being a black box , you know what's inside but you can't touch it. Then imagine having to swap the whole engine cause of a single faulty piston .

[@johanbouwer5737]

It's obvious from the video that the card sustained a really bad drop, AMD or Nvidia that will result in a dead GPU.

[@RFC3514]

@@johntet - Absolutely nothing in this video points to chiplet interconnect "shenanigans". He didn't even test the data lines on the board, let alone any connections in the chip itself.

[@RFC3514]

0:42 - The "chiplet design" (also used by all current high-end CPUs, and now also adopted by nVidia for their next GPU generation) actually equals _better_ yields, because a single defect on a silicon wafer will only ruin a small chiplet (that can be replaced during fabrication), instead of ruining the whole GPU die. Now, it's not impossible that TSMC (which is the company that actually manufactures those chips) had assembly problems with some specific early model (though I'm not aware of it), but chiplets are very much here to stay, from _every_ manufacturer. Big monolithic dies are dead. And nothing in your diagnosis points specifically to a problem in chiplet interconnects. If that were the case, changing the memory chips wouldn't have changed which memory channels you got errors on. Far more likely there's some damage to the board itself, but you'd have to chase and analyse the actual signal lines (using boardview, an oscilloscope / protocol analyser, etc.) to determine what it is, and it might still not be repairable at all. Another (simpler) diagnostic step would be to put that GPU on an identical (but known to be working) card, or putting an identical (but known to be working) GPU on that card, but for that you'd need to have at least _one_ of those, and it seems that you don't; I guess you don't get sent many of these cards to repair, which is probably a good sign (means they don't fail very often).

[@Azureskies01]

well to be fair monolithic will stay for lower end card at least for a while. the high end stuff is going to be chiplet based for sure though in the next 2-3 generations

[@Lazllb]

While true that the chiplet design approach addresses the issue with cost and yields, the reason NVIDIA is pushing it on their newest chips is because of AI, not cost. They’ve already reached the limits of current HBM/DRAM designs and bringing the memory chips closer to the GPU improves the performance.

[@worldtownfc]

TSMC and other packaging companies don't report on packaging yields. We don't know how much packaging yields eat into better wafer yields with chiplets. Allegedly, Vega 56/64 are known to be fragile due to HBM and interposer issues that develop from wear and tear. Almost seven years later, we can hope that AMD and TSMC figured out a fix for the fragility of earlier designs. Only time will tell to see long-term reliability of RDNA III's chiplet cards (RX 7700 XT/7800 XT/7900 XT/XTX).

[@RFC3514]

@@Azureskies01 - It's down to total die size. Naturally, there's a point where assembling multiple minuscule chiplets is more expensive than losing a few more (equally minuscule) full dies. I'm talking about gaming CPUs and GPUs (meaning mid to high-end), which are the ones people bother to send for repair.

[@RFC3514]

@@worldtownfc - We know it's economically preferable, otherwise their clients (including AMD _and_ nVidia) wouldn't be heading in that direction. Intel has also been using chiplets for years.

[@hmello3250]

AMD is fine as long as you don't drop it

[@brazilpaes]

Just like ngreedia.

[@bonzer111]

@@brazilpaesNONO, dont plug those in, they catch fire!

[@19CD91]

I hear northwest repair saved a few marriages with a good old fashion reball.

[@Stikkzz]

just don't smash your gpu on the ground

[@philr4533]

sick vid, props on the effort put into trying to save the device! 😅

[@northwestrepair]

yeah A for Affort

[@nerdypcbuilder]

I've had my sapphire nitro 7900xtx for a year now and works great. Core stays under 45° and hotspot doesn't go above 60° I love it.

[@derim006]

so why do you need that kind of gpu and do you play in 88k ?? :)

[@adamliz8172]

Man you cant play games, your talking idle temps, theres no 7900xtx that maxes out at 45c

[@naomiarmitage8729]

there is no gpu under 50º or 60ºcelsius under load, period. unless you do watercooling...

[@K31TH3R]

@@naomiarmitage8729 You'd be surprised what a good undervolt and a reasonable FPS cap can do for a Radeon card, you can make them run extremely efficiently and cool with little to no loss in performance, or sometimes even a performance gain if you have a good piece of silicon that undervolts well. The heatsink on the Nitro is absolutely massive, so 45C edge 60 TJ with an undervolt and slightly ramped fan curve does not seem outlandish to me at all. Water does help though. I run a Liquid Devil 6800 XT and if I undervolt to 950mV and run at the stock 2360MHz it will cruise along at 42C all day no problem.

[@thegeekno72]

Moral of the story : GPUs aren't intended to be disassembled to that degree and chiplets are drop sensitive Who could have possibly guessed. As much as I love this channel's repairs, this seems like unnecessary slam on AMD

[@RFC3514]

Nothing in the video indicates a problem with chiplet interconnects. He didn't even check the logical traces on the board, let alone anything in the chip itself.

[@cowbee8865]

NWR has always been like this praising EVGA cards and slamming AMD. If you followed him for a while this is obvious. He is a great repair tech tho and thats all. I do enjoy watching his vids still regardless of his opinions as most of his statements from time to time are clearly bias.

[@emilbotev3656]

@@cowbee8865 Spot on. Almost like celebrities. You like and admire what they do and maybe not so much who they actually are as a person (character or opinion wise).

[@BrunodeSouzaLino]

It would be impossible for him to detect problems with chiplet interconnects. These connections are done at the factory at die level and not at the socket. A drop strong enough to disconnect the chiplet connections would probably snap the GPU in half.

[@jonnygudman1815]

The card appears to have been treated very badly, and the original sticker doesn't change that. That was the trigger and nothing else, even if chipplets is generally a bad idea.

[@TheHighborn]

It's clearly AMDs fault that there's visible physical damage on the board. What a joke.

[@RFC3514]

It's such a "bad idea" that ever major chip designer / manufacturer is using them. I guess they know nothing about processor design and manufacturing, eh? 😜

[@gusgreen3104]

Dunn Dunnn. “In the Graphic Processing Unit Repair System, the people are represented by two separate, yet equally important groups: The consumer who buys the part, and the Repair Master who prosecute the offenders. These are their stories.”

[@ryanreedgibson]

I hated that show.

[@user78405]

time GOVERNMENT PUT AMD ON TRIAL FOR FRAUD

[@wesleypiper7675]

Bro i love the retro 80s tunes 🤙

[@ResidentWeevil2077]

It's actually synthwave/retrowave music. Karl Casey creates such music royalty-free.

[@umeng2002]

To be fair, Vega had the VRAM as the chiplets (HBM memory, which nVidia already uses in server parts). The new AMD and nVidia GPUs are true chiplet designs were parts of the previous monolithic dies are broken up. In AMD's case, the cache and memory controllers are on the chiplets.

[@JJFX-]

Getting frustrated with AMD cards is fine but I don't understand why you frequently misrepresent common design choices has mistakes. Vega wasn't a chiplet-based GPU, it was a monolithic die with HBM memory instead of GDDR5. This has been common on data center cards for a long time. The 7000 series is the first chiplet-based concept but it's for the GDDR6 memory interface. The GPU core is still a single die in the center. I'm not saying it's without issues but should be more cost efficient than a single die of that size and certainly less expensive than modern HBM.

[@inabstract]

I like this starter, ambient background music.When i discovered this channel was the one used in alot of your videos. It just sets the mood perfectly for some Gpu repair time 👍💪

[@circumcisio10]

Just to let you all know that I liked this video for your effort and diagnostics, not because of the loss.

[@RonnieMcNutt666]

That name tho

[@circumcisio10]

@@RonnieMcNutt666 It is not a name, it is a nick-name.

[@napoficial7123]

2 hours rebowling video with relaxing music! It's a must! ❤

[@ggoddkkiller1342]

1- Take a butter knife 2- Heat it to 200 degrees C 3- Very carefully follow the lines between chiplets..

[@giornikitop5373]

why stop at 200? heat it until its red hot, it will cut much better, almost like butter!!!

[@glintytear]

I find reballing smaller phone CPUs and memories a daunting task already. Cannot imagine myself trying to reball THAT

[@ResidentWeevil2077]

That core is MASSIVE no doubt.

[@andrewgrossman7]

I love your videos. Idk if i am the only one who listens to it for ASMR purposes... But it has to do with your accent, and the music in the background haha

[@cina9218]

NVidia just announced they will be doing this too. GG

[@JusstyteN]

Gg what its a logical thing to do coz if yealds on the chips makes it cheaper

[@hombrepepega3472]

There's no evidence it's ThE ChIpLeT dEsIgN problem. You can destroy connections between a single crystal and its pad and repairman will shrug just like here.

[@RonnieMcNutt666]

Well since Nvidia is doing it its good! Won't have to worry about amd chiplet drivers getting in da way!!!!!!!!!!!

[@pesokpesok]

had this happen as well and turned out to be pcb broken trace-signal. Pain in the ass to locate but easy to fix once found. Bet its the same with yours. These new cards weight as much as a tank(for hampsters) and cheap pcbs used plus strain(always mount correctly to minimise - if hanging with only pci-e for suport - you are asking for this to happen) spells trouble

[@RFC3514]

Yep, my guess would be board-level damage. Might be on the surface, might be an internal layer (and not fixable). But without analysing the data lines (or at the very least trying the GPU on another card), there's no way to advance the diagnosis. Claiming "it's the chiplet interconnects" is just random.

[@armatian]

Chiplet makes as much sense on gpus as cpus, but its more difficult. Remember bulldozer the fact that monolithic quad was theorethically superior didnt fix a bad basis. The first dual core was pretty slow at first took a long time to (windows) software to catch up. Im not into first gen of any new, but i trust this will work eventually.

[@10100rsn]

I guess... don't drop your expensive GPU on the floor... or on the table.

[@TheBackyardChemist]

I dont think it is fair to blame AMD for not making the card impact resistant, IMO the real problem is the lack of ability to disable the failed memory channels and keep using the rest of the card.

[@1Rictec]

the problem is NOT the chiplets or AMD look on how that card was used don't drop high tech devices .

[@northwestrepair]

I agree but i've seen worse and core would survive.

[@1Rictec]

@@northwestrepair i do think the chip size also does not help with drops you know the force is not the same across the chip and since some parts are less dense they can break also think they will solve those problems when chiplet designs improve and get to be used on other smaller devices then gpu cards

[@RFC3514]

@@1Rictec - Chiplets have already been in use for several years (by Intel, AMD / TSMC and other brands), including on the best CPUs of the current generation. There isn't much reason to use chiplets on very small devices, because then the extra cost of assembling them into the full chip will be more than the yield benefits from using chiplets in the first place.

[@RFC3514]

@@northwestrepair - You've also seen boards with _less_ physical damage where the "core" (GPU) _didn't_ survive, right?

[@1Rictec]

@@RFC3514 i know that but in the future they will reduce the overall size of the entire die for sure.

[@billchildress9756]

With all the good older cards out there I say go hunt! To those with deep pockets.. let them buy new ones.. If you can afford pricey parts then make sure you spend wisely. I agree that these new cards need to be handled with care and respect! Nice try on this one though but you can sometimes only do so much unless you get some more donors in.

[@Lazllb]

Nice video! I have a 7900XTX and various other GPUs that I use for development purposes. I saw you were using Windows to diagnose the device. Have you ever thought about using Linux instead? It’s much easier to get the information you want from the devices than it is on Windows. Error 43 is just a generic driver failure code and it’s almost useless as to determining the culprit. AMD gpu drivers are open source too. Testing NVIDIA gpus on Linux are a little more complicated but NVIDIA has been improving their support for Linux thanks to the AI boom. Anyways, keep up the great work! Edit: Nevermind I saw you did use some Linux as well after! Nice!

[@lop1652]

And Arch of all distros.

[@the_hate_inside1085]

I'm sure you will be thrilled when your beloved Nvidia releases the Blackwell GPUs, that are a chiplet based design, and the size of the chip is like a small football field. According to their marketing, "the die is the biggest size that is physically possible"...

[@tonict2302]

Awesome video as usual...

[@Del_UK]

Maybe, do what AMD does and disable 1/2 of the chiplet/s in bios? You might have to take some memory off as well. Many, many, many years now, AMD have disabled certain parts of the core of either CPU or GPU and sold it as a lower model. Reason is, the die did not pass tests to be a higher product SKU.

[@northwestrepair]

Disabling channels is possible for 30 series Nvidia but as far as i know, not for AMD. All thanks to encryption.

[@giornikitop5373]

this is not new, it's been done also in the cpu space from intel. disabling cores is done in manufacturing, not possible for a user. disabling vram might be possible but you need a modded bios, but they are encrypted ussually.

[@Del_UK]

@@giornikitop5373 I am sure I have seen over clockers using a BIOS tool. Maybe, AMD, Intel and Nvidia just locked everything down.

[@giornikitop5373]

@@Del_UK yeah, they're not idiots, if something can possibly threaten their profits, they will block it. but tbf, every manufacturer does that.

[@alicebonnet4607]

Really let us down with this one. Nice work regardless.

[@petrifiedhero9610]

can you repair my radeon r9 fury, a resistor was nocked off by the pci connector, and a power component consisting of a copper coil has a casing thats cracked, is it possible you could repair my gpu?

[@ThePrimaFacie]

The music in the first big ff repair was pretty cool. The transition from repair to blank screen is jarring maybe have a name of the test bench or something in a corner so instead of someone just seeing a black screen then the sped up "AMERICAN MEGATRENDS" show a little something so the avg or new watcher knows whats going on a little cue. Also if you can help it maybe lose the "soft edge" or maybe keep like a 1/4 of it? For me it makes it look like I need to wipe my eyes to see it clearer, or that you have to much flux on the lens. Thanks for the vid and the info. Have a good one

[@Azureskies01]

If chiplets are bad you better tell that to nvidia, they just announced a dual GPU chiplet AI chip they are going to sell for a shit load of money! they gotta know! you gotta tell um!

[@hinz1]

That stuff gets ridiculous, like 200 billion transistors, god know how many interconnects, 50'000 microBGA balls for the chiplets, 5'000 BGA balls. And if one single important one of these fails, dead GPU.

[@northwestrepair]

@@hinz1yep 😐

[@Azureskies01]

@@hinz1 they clearly don't know how to make GPUs, just like AMD /s

[@maklogetrich2378]

​@@Azureskies01tell that to nvidia? good luck with that 😂

[@RFC3514]

Yeah, I'm sure he can school nVidia's engineers (and TSMC, which is who actually _makes_ both nVidia and AMD chips) about *chip design.* 🙄

[@Suresh_Pillai]

Sugar!! Missed the sweet spot! But great NVIDIA design though.. Tony, you showed your appreciation well.. hope its not going to deaf years in NVIDIA.. 😆😆

[@EinSwitzer]

I had one of the first prototypes 6800 rx xfx gpus it was a single chiplet made of chip pieces like memory controllers split left side of chip right side of chip

[@stewiegriffin6503]

have you tried to restart PC ?

[@TeslaExplorer]

calling it a day requires a skill set of its own

[@Rmm1722]

Good work 💯

[@xavierjiang7112]

If anything, it tells you to treat these even more carefully than you would. Not unlike CPUs. IHS, heavy duty reinforced heatsink bracket and everything. And perhaps a new form factor not relying on a single bracket on one end for support.

[@coladict]

Oh, btw, this could have been an early batch where AMD had QC issues with the vapor chamber leading to inadequate cooling. That's a known problem for the reference model. Later batches have this fixed, but you can't know which one it is when you're buying it. That might be what killed the card.

[@vacannot]

would you do a "gpu tierlist" of the different manufacturers build / cooling quality of the 4000 series?

[@farfromerease]

Amazing video as usual. Even if you repeated the same commentary on every video but on a different card you fix I'd watch it because your voice reminds me of your sense of humour and the jokes u make some times during editing 😭😂

[@Cyb3rn3tiX]

I love these videos, definitely informative, and expensive GPUs break just as much as cheaper ones. So what's your thoughts on a decent GPU in this time of age 🤔🙂

[@Dandan-tg6tj]

Not comfortable to ask you this but did you test the data lines on the board and the connections to the chip itself? Thank you.

[@TechLevelUpOfficial]

Vega had HBM memory on board, the Navi 31 cards have chiplet design but still rock a normal GDDR6 modules on the PCB. It's not the same thing.

[@Arokhantos]

I thought my 7900 XTX was dying but it was just a driver issue that everyone could reproduce in Dying light 2 during one very specific cutscene where skullface gets introduced along with cool gas and smoke effects that produce an RGB laser show now fixed on 24.3.1 after a lot of nagging.

[@RFC3514]

If it was just in one game and in a specific cutscene, maybe it would have been more productive to ask the game developers to fix it. Modern graphics drivers actually include a lot of exceptions and workarounds to deal with game bugs (that's why Intel ARC cards still have so many problems - it's not actually Intel's fault, it's just that they need to work around game bugs and badly optimized game code), but neither AMD nor nVidia are very likely to make a priority out of fixing some problem that affects _one_ cutscene in _one_ game.

[@Arokhantos]

@@RFC3514 i asked both, and it was not one cutscene, i reported it in 1 cutscene that happens to be very accesable tho and easy to check

[@utrak]

Donating that card was a chad move by the owner

[@DerKopfsammler666]

I was considering to buy one of these cards for an upgrade, but seeing how well they die... I'd rather wait for next gen or something... 😅

[@SaHaRaSquad]

Well on this channel you only ever see the broken ones

[@covante4822]

1:29 I've got one and it's been great for me. My first amd card ever, been nvidia since the 660 but I've had a better experience with this than any nvidia card. Probably wasn't very good for the CS players who got banned for no reason or this guy whose card is in the video who dropped it and broke something "We have a sign of an impact".

[@meltinsr]

​@SaHaRaSquad I have more than a year the 7900xtx and its super but this video got me worried

[@RFC3514]

So you came to a *repair* channel that showed *one* unit of a specific card model that had failed after being *dropped,* and you concluded they must die very easily? Great logic.😆

[@setadoon]

7900xtx is baller. Don't treat your fragile, expensive electronics like basketballs and you'll be fine.

[@mleise8292]

Chiplets are inevitable right now to get the numbers of transistors up and cost down, as much as it is a pita for repairs. As for the RX Vega, I've never thought about what happens when an HBM2 chip dies. It's a total loss, right? To be fair, I think most that developed memory issues were mined on 24/7 and the number of repairs does not reflect what the average customer playing games experienced.

[@buffhorses]

Dang I just bought the Sapphire Nitro Version of this card. I hope I don't get screwed.

[@setadoon]

It's an amazing card. Best 7900xtx made 👍

[@coladict]

That Nvidia chiplet design is specifically for their datacenter cards, because it costs an estimated $6000 to manufacture. That's before Nvidia's own margins. Don't expect it to trickle down to consumer cards any time soon, but you could expect more datacenter customers who are willing to pay top dollar for that repair. Oh, and they're using HBM3 memory, which is also made part of the chip package, meaning you can't remove and reball it like you do with GDDR.

[@manISnoGOD]

Hello my friend can you make a video of your set-up and what to need to go in repairing gpu's great learning videos GOD bless you

[@Swiss4.2]

Unfortunately chiplets are coming to Nvidia as well; they are far more cost-effective and suitable for larger dies at the expense of repairability.

[@user78405]

nvidia is very different designed ...it use intel substrate idea vs amd dumb idea to use not substrate for their chiplet cache ...cpu would be fine but extra cache will not when they solder entire on top cpu core...bad idea when intel use substrate for extra caches

[@RFC3514]

There's nothing "unfortunate" about it. AMD and Intel have been using chiplets for years (the best CPUs all use chiplets), and both AMD and Nvidia chips are made by the same factory (TSMC).

[@patx35]

Fun fact, the previous VEGA chiplet design was AMD's attempt at making a fast GPU, but it was a failure because they have too many dead HBM memory chiplets at the factory. The new chiplet design is a cost saving measure. The memory controllers are seperated from the main compute core. That way, AMD can use cheaper manufacturing methods for the memory controller.

[@GOBLeo]

Goddamn! After all this work not getting anything in return would be a crime.

[@RFC3514]

First step should have been to put the GPU on an identical card that he knew was working. That would immediately tell him if the problem was the GPU or the card. But I guess he doesn't have another one.

[@hoseinqadam]

I dont think its fair to hate on chiplet design because someone dropped the card. That is like hating having windows on your car door because it breaks when you hit it with a rock (sometimes it breaks and sometimes it does not).

[@BasicComputerRepair]

Lots of knowledge and skill. Yet AMD manages to make that obsolete. Good work sir!

[@SilverBullet27188]

What GPU would you recommend? You have pointed out flaws in every card that has come through your repair shop. I went with AMD this gen because I didn't want to spend 1600 dollars nor possibly get a melting connector.

[@RFC3514]

He only gets sent cards that fail, so look for models that he _doesn't_ show on his channel - those probably fail less. 😁

[@t1e6x12]

@@RFC3514 Also have to keep in mind some models will be more popular than others. As an example if you see 50% of the cards coming in are ASUS, that doesn't mean ASUS sucks if they account for 50% of the AIB market.

[@RFC3514]

@@t1e6x12 - Of course. They suck mainly because of Asus Armoury Crate. 😜

[@K31TH3R]

The universal answer to the question of "which GPU would you recommend?" is "just get a Sapphire or PowerColor 7800 XT." The 7800 XT is to GPUs as Brother is to printers. "Want something that prints? Get a Brother. Want something priced reasonably and just works? Get a Brother. Want something that makes weird sci-fi noises and perfectly replicates your important documents? GET a Brother. Want something that won't piss you off? GET A BROTHER. Want something that takes up desk space and collects dust for 6 years but still works when you need to print something for a court case against you? BROTHER. Want to print off grandma porn to hang up on your wall to make people who invade your space uncomfortable? Bruh. But also Brother."

[@proluxelectronics7419]

It's amazing the lead free solder would be earth friendly, Must be when it goes into landfill..😆😆

[@RFC3514]

That is literally why lead (and mercury, and other heavy metals) are banned from mass-produced goods: because people just threw them into the trash, the metals would leak into the ground and would contaminate the food chain. They're still allowed for repairs, because that actually helps reduce the amount of waste. But as long as people buy tons of crap and throw it into the trash every day, the only way to prevent everyone from being poisoned is making sure those substances aren't there in the first place.

[@MaddJakd]

It is an element afterall....

[@proluxelectronics7419]

@@RFC3514 Yes, but lead free and thermal stress don't mix, lead is ductile.

[@RFC3514]

@@proluxelectronics7419 - I think I haven't had a single electronics failure caused by cracked solder joints in the last 10 or 15 years (compared to several caused by dead capacitors, for example), so that really doesn't seem to be a big issue. It took some industries a couple of years to figure out the best temperatures, the best fluxes and and the best lead-free formulations (there isn't a single type of lead-free solder, just like there wasn't a single type of leaded solder), but it works fine now. I'll take my water and my seafood with less heavy metals, thank you.

[@Elimunary]

Well, this is only just the beginning the next nvidia gpu series will be like this too because it is the way the tech advances just look up the nvidia most recent new ai chip announcement.

[@tilburg8683]

Nothing that a reball doesn't fix. Flat tire?, just reball the core. Cpu pins broken? You guess it reball the core. South bridge of the motherboard broken? Reball the core.

[@brnmcc01]

You forgot flash the BIOS.

[@maklogetrich2378]

brainware error? reball

[@RFC3514]

When the tool you know best is a hammer, everything looks like a nail.

[@Steven-hq3go]

It's nice he donated it instead of listing on eBay as broken giving someone else the hope of fixing it. I'm guessing maybe you didn't charge the customer for repair attempt as they donated it in the end

[@drerbrerard130]

I watch this channel for the music. I have no idea what he's doing

[@piotrkurek79]

AMD is better than nVidia.

[@inou2222]

I take that he paid you with the dead card instead of a repair attempt fee.

[@MaxIronsThird]

sheet commentary, it's AMD's fault dude dropped a GPU on the ground

[@RonnieMcNutt666]

Duh it's the drivers. ... non- drop proof drivers .. AMD BAD

[@northwestrepair]

No its not AMDs fault that it was dropped. AMDs fault is at producing something that cannot be dropped. Nvidias drop all the time. ripped pads, broken solder joint, cracked PCB, thats it. If damage is not extensive, its repairable but in this case, damage is minimal and its already 100% toast.

[@South_0f_Heaven_]

@@northwestrepairyeah it’s sad top of the line AMD GPU is junk because of minor damage but could also be the nozzle just isn’t big enough to cover the entire core and not heating it evenly. Same issue with the 80/90 series cards the cheap Amazon nozzles need to be 44mm-46mm to properly heat the cores.

[@joshjlmgproductions3313]

​@@northwestrepairWoosh, my guy.

[@dampintellect]

It may be more fragile but people aren't buying GPUs based on how hard you can throw them and still use them.

[@johncanning6428]

can you tell me what program u are using to test AMD memory I'm assuming its a modified Mats and mods thx

[@gamingbeast-mh6ds]

Hey just wanted to know if I could send you my gpu to fix it. I took it apart today to clean it and I couldn’t get the light up part to come out so I put it back together and it was working fine for about 5-10 mins, started playing Mx bikes and I started to see stuttering and after that my screen turned black

[@JusstyteN]

I want you to put this core on a known good board.

[@northwestrepair]

Dont have one.

[@teddp]

And shouldn't bother because 99.9% we know it's the core.

[@RFC3514]

@@teddp - You should sell your "telepathic board diagnosis" skills. Changing memory chips resulted in a different error, but you're 99.9% sure it's the core. Impressive.

[@user-bi7cs2rq1i]

Could you lonk the track playing at 2 min mark?

[@easyhardenduro9847]

It's funny to see these old classic memes😂

[@MostafaTalaat3953]

nice

[@TrueStrike0_0]

was gddr6 or gddr6x used?

[@FirstLast-ds7xu]

200C reball preheater can be not chiplet friendly

[@d.i.p7043]

Btw, Ever new GEN Radeon GPU suffer silly problem like Nvidia? like PCB cracking and melting power connector?

[@joshuascholar3220]

I've got a vega datacenter GPU. It's weird. It will pass memory test as long as you want, but every now and then under load I get an ECC failure error from it.

[@mrhappy8966]

I thought about getting a rx 7900xt to upgrade my rx 6800 but decided not to its still doing well for me.

[@myne00]

In theory AMD/NV could release software that allows you to disable nonfunctional parts and get a working card with lower performance. They just wont because there's no incentive.

[@floser75]

I can sense the Top Gun vibe at 1:51

[@MrGeneralScar]

At the beginning of the video, it was noted there was damage to the back of the PCB, what if this damage was causing some issue. Still interesting nonetheless. I also dont believe nvidia's new AI cards cost really that much more to manufacture than thier top end gaming GPUs. They just charge huge amounts for them because they know the big companies will pay it.

[@CNC295]

Over a $1000 for that 1 card what a hit

[@FranklinClinton1]

❤

[@kamilhorvat8290]

I was expecting chiplet reballing. Any plans to start with it?

[@user-nl2ho9gk3y]

Make a GPU brands comparison video in term of build quality 😊,or even specific models

[@RFC3514]

He only gets cards that are defective or damaged, so the ones with the best build quality are precisely the ones he *doesn't* get.

[@user-nl2ho9gk3y]

@@RFC3514that the reason, he get card for repair which mean he have better knowledge of quality and electronic used in GPU .

[@RFC3514]

@@user-nl2ho9gk3y - Again: he only gets cards that have failed in some way. So the best ones, that fail less often, he will rarely (if ever) get, because they don't need to be repaired. It's like asking a doctor "who are the healthiest people". Those don't need to go to the doctor, so he probably won't know them.

[@tafadzwamusekiwa659]

Lol what a coincidence I repaired one yesterday and XFX but had faulty vram

[@fredygump5578]

Tell us how you really feel?

[@northwestrepair]

like crap. but am glad its over.

[@ibeebzz9804]

Interesting, did that card have Hynix or Samsung memory?

[@northwestrepair]

hynix.

[@ibeebzz9804]

Early card then, most are Samsung now

[@konstantinosvakouftsis4086]

Could someone explain why this PCB is bad I really don't get it

[@Nick_With_A_Stick]

That last part of the video 💀

[@DaleC.]

I wonder if Sapphire or PowerColor make better quality products than AMD. Or are they all basically the same inside?

[@Creep4EverMV588]

From what I saw from few repair videos, amd reference models have the worst quality pcbs. Sapphire and PowerColor look more solid, even my own 7800 xt hellhound... my old psu popped and killed the motherboard or the cpu socket specifically but the gpu survived without any issues

[@DaleC.]

@@Creep4EverMV588 Thats good news. Thanks for the response. 👍

[@Turco949]

Tony, where do you buy your chips from?

[@FunkyTechy]

Chiplets are like disc breaks, you’ll come to love them and when you learn how to effectively repair like you have monolithic dies it’ll probably be easier to

[@vishensivparsad]

probably need very expensive equipment to work with the tiny interconnects

[@RFC3514]

He doesn't repair chips at all (with or without chiplets), nor does any normal repairperson. He just desolders and resolders them (and other board-level components, like BIOS chips, memory chips, connectors, capacitors, etc.). Repairing the inside of a chip is in a completely different league (it's almost never cost-effective, and in many cases it's completely impossible).

[@Morpheus-pt3wq]

Vega did NOT have chiplets, Vega had HBM memory. And most of their failures were related to heatsinks not being properly made to accomodate the change of height between HBM and core - or in case of Asus, were wrongly torqued. AIBs often don´t give a ****, when it comes to making their own AMD designs. Do your research please, before spouting out nonsense. Chiplets probably are the future, since chips becoming too big will greatly affect the yield from wafers.

[@kendoty2463]

. . . Expensive card made hands sweat; board salty & slippery . . .

[@tanjem]

why wont they use a socket for the GPU and made it upgrade features for it like AMD CPU ??

[@RFC3514]

Two mains reasons: 1. Different GPUs generally have different memory configurations, different power requirements, etc., so each card would only be compatible with 1 or 2 GPU models (unlike motherboards, that are compatible with 10 or 20 CPU models). 2. GPU coolers are much harder to install and remove, because they cover the whole card, so it's not easy for normal consumers to replace just the GPU.

[@tanjem]

@@RFC3514 time to do some power standards for GPU ...it is running wild ...

[@kendoty2463]

. . . Avoid using unreliable parts on an expensive module. . .