Absolutely fantastic. In the late 1970s, I first worked as a repair tech on Moog and Arp synthesizers. It is heartwarming to see people still doing component-level repairs, even more so to see it done without unnecessarily replacing components. I have heard of reballing an IC but have never seen it done before. Tony, thank you.
Oh hell yes! I'm an analog synth guy. You're so lucky, and probably got to meet Robert Moog too! I also do component repairs on my own gear. I'm a big fan of the youtube channel Mr. Carlson's Lab for techniques and inspiration.
@@Nathriel I did in fact meet Robert Moog. He made a presentation at the technical school I attended, and he was talking about the future: digitial synthesis.
One of those random RUclips suggestions that pops-up and keeps me watching until the end! As someone who has done SMD work myself, I can say a very good job! Glad to see that there are stil component-level work being done by specialists in the throw-away society we live in! Great video, thank you!
I don't know how popular the long format videos are but I do know that they show us a slight portion of all the work and effort that you put into every card into every repair Tony. I hope you had a nice lunch or dinner after that 😊
It's not always about the video's length. It's more about if the video keeps your interest and attention during it - so that you don't worry about how long the video is.
@@johnpaulbacon8320 People are less likely to click on a 40min video than on a say 10min video. It's just a larger time investment and the algorithm knows that. It's pretty common knowledge that longer form videos get less attention on YT comparatively. Edit: I personally totally agree with your statement. I just wanted to point out that longer form videos are less viable on YT generally.
@@Bassalicious They see less traffic but tend to perform better. Longer form video's tend to take a larger time commitment, absolutely agree, so those clicking them usually have an idea of the content and are more likely to both watch it through fully and also comment. Shorts for discovery, mid length (the 8-20 min vids) to get people comming back and long form for better revenue and audience retention/engagement. (long form kinda games the system, particularly for premium viewers revenue share)
My ASUS 4090 Strix low voltage light showed up, tried it in two rigs, still the same. I would have taken it back to Microcenter, but knew getting a 4090 replacement was about impossible. I thought I'd roll the dice and send it back into ASUS for repair, big mistake. They wanted to charge me 500 more dollars than the price of the gpu to fix it, 40 dollars for shipping, and 20 dollars for laborto fix it. No negotiation whatsover, called a customer serivice call center that didn't care for my plight or the repair cost, told me basically tough luck, finally contacted the ASUS US CEO, (Reddit research gave me the address) and they replaced my card, They said they fixed it, but sent me a replacement instead. It's working now so theres that. Nice to know that there are people out there fixing GPUS instead of rip off repair centers. (edit on price of repair update lol)
I will never buy anything Asus ever. I have only had to deal with their RMA once but it was hands down the worst experience I've ever had with a company ever. Just an awful company.
A few months ago GamersNexus made a documentary about RMAing the Asus Ally i think, and how badly it went. Reading your comment now comes with no surprise to me. I wish yall better luck with RMAing things.
Sorry to hear your struggle but you didn't give up & won. I've only had to rma one video card & that was a 6800gt gigabyte. Still daily an Evga 9800gtx+ Mint Evga.
@@neonlost Yep I once had to fight months to get my asus gpu back. They somehow "lost it". And btw didnt even inform me about that before I inquired a lot?? So they weren't going to do anything if I dint ask about it lol. I have had many GPUs, both nvidia and amd, and I have had to RMA 2 different GPUs, (only pc part that I ever had to RMA btw) and guess what, both were asus. From now on I avoid asus parts by all cost because I dont need that kind of headache in my life.
Nice repair as usual. I like the small pre-heater, a must for sure to avoid having to blast silly temps to a cold IC. I use my main pre-heater (same as yours, just still in one piece!). I place an aluminium plate on top and the IC goes there. The small one is convenient though.
my FE 3090 survived when i accidentally spilt tea all over my PC a few years ago. had to open it up and washed the pcb in dish soap, then dried it with isopropyl alcohol & replaced thermal pads; surprisingly it still worked even after accidentally knocking off a filter cap. this card has been an absolute champ for 3 years. the old skylake motherboard didn't survive however
@@antonchigurh8343 from 2014 to 2020, cpu advancements weren't very progressive. was planning to upgrade eventually. ended up getting an i9 14900K after the old PC died and paired that with the 3090. the bottleneck for skylake wasn't too noticable
What do you mean CPU advancemens weren't very progressive at that time? Intel's biggest performance leap ever was from 7 to 8th gen when the number of cores increased by 50% from one gen to the other on the i7's
Hello i'm usually not writing comments but I've been watching your videos since you had 2k subs and I really enjoy ur Sorin's style but with GPUS :) . Actually i'm really interested by your tools because i've seen that not only u are choosing good tools but also you have a lot of them that are homemade like the core lifter or the probe with the button to inject voltage only by pressing when we want and even your broom made or small wires that allows you to find traces on the board without the boardview. I would be really interesting if you do a video to introduce them all.
I've never heard of a 3090 memory upgrade mod but I'm sure it's probably possible given it is on just about every other model. The strap configurations are likely similar to the other models that have been modded. It is definitely a mod for the 2080Ti (22GB), 3070 (16GB) and 3080 (20GB). I think I have seen Chinese sellers even selling pre-modded 2080Ti's and 3070's.
Tony, you are a GPU professor, full of knowledge, and you know how to teach and explain. To be clear, there are people who know, but they are not like you, you know how to explain and that is the most important thing. A genius professor specialized in GPU repairs, and probably many more. People like that made me fall in love with electronics.
I've been a subscriber for both but Today i Learned Northridge repair and Northwest Repair are two different channels. I always thought the two were just different techs in the same channel lol
@@Crawdoodle yeah i thought they were two people in one channel. I honestly prefer this over the other guy. He seems a little too angry all of the time
@@kitsumyr9752 Tony here is "real" as in one of us. He tries to make the world better. the other guy is hard on about selling stuff and gives up pretty easy with "no fix" if its not a connector or mosfet failure.
The 3090 has memory on both sides of the board. I fixed this by putting a heatsink and fan on the backplate. I used mine for mining, so I really had to manage those memory temps. In the end, the mining paid for the card and then some. Mine was not a FE, it was a Gigabyte and had a metal backplate.
When trying to find where rubber thermal pads are NOT touching the heatsink, a piece of glass is a great tool. You can see which ones touch the glass just by pressing it against against the board.
The bigger downside to halving the number of chips (the possibility to even do it due to firmware restrictions not withstanding) is the reduction of the bus width, in effect halving the advertised memory bandwidth. And those chips crave memory bandwidth.
Jensen LOVES his customers. It's rare to actually find people who pay for the privilege of being r@p3d and then come back for more, what's not to love? Of course, lube is an added upgrade because rectal bleeding is not considered a bug 'butt' rather a feature!
Measure the power consumption. If excessive, the unit is putting out more heat. Of course, good compression of thermal pads and good wetting of thermal paste is important as well. I work more with audio than computer, but it's not uncommon to have an OP amp drawing abnormally high current and running hot, even though it functions perfectly. You might even invest in a FLIR camera, that would let you single out the hot running chips.
Hey there. I recently subbed. Very good repairs and attention to detail. I am also doing BGA rework professionally, but i must admit, you achieve 5 star results with the basic tools you have(mad skillz). I would recommend that you look into tools like PDR infrared rework stations, Finetech rework stations, and you deffinitely need a Martin mini oven for those bga chips. Your attention to detail and determination is very motivating. All the best!
If you are looking for reballing machines, have a look on the IR8500 (entry level) or a Honton R490/R690 (more “professional”). The IR is not that expensive, the Honton cost a bit more but might be worth for a person doing this for a living. Plenty of videos on youtube about both models. Hope that helps. Kind regards
Nice work as always Tony. I'll post a link to a DIY re-balling station you may or may not have seen on your DC. In respect to the VRAM increase or 2GB modules on the fron...I'm intriqued myself. Great job and thanks.
Forget about the memory, would be a miracle if these pricks at nvidia had forgotten to make it impossible, somehow. Easy in the 1990s, impossible now, mostly because of cooperate greed.
you speak nonsense, u can edit the memory for 3000 series, theres are 3070 with 12gb, 16gb, 3080 with 20 gb, 3080ti with 24gb... if someone finds in bios how to disable the back memory we can do it if it wasn't doing done already
@@Kiburici Yes, with 48GB and Quadro bios, it would be a Quadro RTX A6000. Apparently, flashing GTX to Quadro with resistor modding or something wasn't impossible until not too long ago, GTX690 to K5000, I think, then Nvidia got greedy and decided to do some laser lobotomy, so RTX will never be a Quadro ever again.....
Very interesting video. Had a motherboard with a gpu on it that needed exactly this fix (gpu pulled and re-soldered). A common issue with this mb. Was able to heat gpu and got it working for a short while. ThX for sharing.
Enjoyed this video, I’m going to check my temps using the same method on my sapphire nitro plus 7800xt because I’ve spotted some inconsistent temps between hotspot and memory junction, I did very carefully check the tension on the retention plate 4 screws thingy and got about a millimeter worth of movement out of them, this made a difference of a few degrees on hotspot temps then I adjusted the fan curve to keep things in check, will now check those memory temps using gpuz and furmark and compare them to the hotspot 👍🏻
The 4004 CPU’s with 4K of memory I was repairing back in 79 weren’t quite this complicated. Switching in the paper tape boot loader on the front panel switches was the biggest pain.
When we check piston to valve clearance in an engine we use plasticine and measure how much it squashes, I guess you could glue down little balls of alum foil and measure the height after a dummy fit.
@@AppriI also love the way they talk about measuring the height, rather than the squish width lol - I can see those aluminium balls not squishing and causing some snappage also !
Ahhh!! Another long format video! Much better! Better then that!! Your custom core lifter is baller Dude. I want one. Lol. I saw at one point you tried an ir plate above. Hot air better distribution?
I used to own this GPU, I replaced the pads on memory, which fixed the memory heating up so much. I actually had memory junction temps of 110 degrees celsius! After replacing the pads the core even was able to go into thermal throttle rather than it throttling on memory.
I am about 2 years into self-teaching myself on doing repairs and I have to say I really wish I had the tools used in this video lol. Unfortunately, I am not able to work a normal job right now witch also means I am broke lol and I do not have the money to get tools like you see here and even the tools I do have and can afford I have been really just making the crap tool I have work. Anyways cool video thanks you for the entertainment and wisdom.
Just for reference my 3090 FE, just like in this video, has the Memory running at 90c under load. Thinking this was an issue on my 3090 I found that the memory, has a Max temp of 105c, before its an issue on the 3090. This is the design that NVIDIA did with this generation, I believe due to the GDDR6x chips. So seeing the 90c+ on this 3090 FE, is not a surprise to me. If you want it lower, then its at best memory pads, replacement from the crap (white straw) pads that NVIDIA does at the factory. My card ran fine, and did very well in Microsoft Flight Sim 2020, at max FPS.
Those crappy pads were the sole reason people started trying various thermal pads and even copper shim mods. Especially under backplate, where some individual GPUs had no pads at all (this was during covid, so there were possibly shortages or workers just "forgot" to add them)
Just because they say those temps are within spec doest mean it's safe for longevity of the card life/board its attched too including excess heat on the core from behind. I would always advise to watercool the card on front an back plate for 3090 and 4090 series gpu just because of memory chip temps are way too high with any stock cooler on for my liking ,since watercooling my temps on memory under load are maxed at 39 after long periods at full load
On the 3090s the mem temp being so high is due to half the chips being on the back side of the pcb. Almost no cooling. The sensors will always report the highest temp of all.
ive got a 3090 fe and ive repadded its gone down from 105 to around 86-90 depending on the case its sat in , this is as good as it gets for the 3090 FE. Also i came across something on the interwebs which states theres 2 revisions of the card and depending on which one it is depends on the thickness of the pads required. i used artic tp3's 1.5mm on mine and its been fine ever since :D
It's crazy to think that this wouldn't be covered under Warranty where you are. Here in Australia I think government demands 3-5yr for computer electronics. Use to deal with a lot of these grey pads (dry solder joint) back in PS3/XBOX360 days. This level of repair is rarely worth it unless device is REALLY expensive.
It's just so amazing to me. The simplicity of it all, but how say a piece of foam makes makes all the difference in the world. I didn't realize the alignment of a heat sink was that important.
One problem with some of these boards is that they really need a brace or to be run vertical due to weight. I've seen many GPU boards that have a suspicious amount of flex in them when installed. One old card of mine you could see light between the pad in a couple of places as it had simply pulled away too far.. My son's computer was remedied by putting in a long zip-tie to the chassis to relieve some of the excess weight.
When you are constantly trying to make things cheaper stuff like this is bound to happen. Like the pads that are about as thick as a piece of tissue paper in the name of saving a few pennies on copper. They clearly are designed more for one time soldering than repairability but the fact that they are pulling off the boards merely from uneven mounting pressure shows they are not even sufficient for that.
To change the memory sizes on the GPU, you must have access to a schematic from the board (not public available) and be able to reconfigure the jumpers on the board (0 ohms resistors or not, I saw previous models worked by percentages in drop voltages) and may be some cases changing the memory space inside the GPU bios.
very detailed as always Tony, thx to You I have learned to repaire grafic cards and boards in general. Been asking in last few videos, what kind of new software for memory are You using ?
10:18 you can add a 3d printed impeller and diverter to your output to achieve the spread of hot air. A changeable diffuser nozzle would allow for directed heat versus diffused heat for specific applications.
Speaking from experience, the completely working gpus you get sent will likely trace back to a failed PCI-X socket. Had one of those in the other day. These stupid cards or the MB/case's they put them in put too much stress on the slots, and the slots fail.
3090 memory getting super hot is so common, passive backplate was never really enough cooling. When I had a founder's edition 3090, I put a waterblock on each side and that's what it took to get decent memory temps.
Why are you reballing? The issue is most likely between the die and BGA package, not the package and the PCB. The thermal stress from reballing might solve the problem temporarily, but you're charging the customer for a fix that probably won't last.
3090's memory temps are always insanely high when in use. they can even go up to 100c bcs there is no active cooling on the back of the card, where the other memory modules are
I redid all the pads on my 4 4080s. I can not remember the sizes but you can find them on many crypto mining sites. The new pads made a huge difference in memory temps for me. like 15-20 degrees under full load.
Great video that was full of helpful information. So thank you. Do you know where I can buy power harnesses like yours to power the card? Also, that tool that you have to check power with two screens, did you make that and is it something I can buy? I really am a beginner and I'd like to get set up on a bench properly.
I've seen enough reballs to know that you would be almost "out of business" if they made sockets for the chip on the GPUS like on CPU mobos. And gpu vendors would probably starve to death.
@@ariewijaya1679 yeah he's underestimating how many people would still break something. Or the fact people are terrified of opening up their 2000+ dollar gpus themselves
I've seen a GPU case where it would 100% pass furmark with zero artifacts or crashes but random polygons would disappear in games, it was annoying AF. All memory was OK. I traced the anomaly to vCore fluctuating like crazy. Unfortunately the card died with a probe slip before I could fix it or even determine if it was defective core or vrm. vrm controller swap didn't help.
Great Details-Thanks As long as you're there, a few strips of soft copper on those mem chips could fill the gap and conduct better than extra thermal tape ?
@@fixeuskeane ...and occasionally blow up customer hardware by injecting 3V into a laptops CPU and charging the customer a fee for him destroying it live on stream, then going on a banning spree for anyone pointing that out 🤡
Even though they don't seem to likely cause any contact where it shouldn't occur, when you switch off the little hearing plate there still seem to be little traces of solder paste inbetween the points the should be at, almost looking like dendrites (14:44, left corner, a bit earlier also right corner)
Looks great, I have a 1060 GTX card that is acting up in a similar way. Normal 2D works fine, but if you run a 3D-intensive game, like the System Shock Remake for about 15 minutes or so, the whole system hangs. It's not a driver issue, because the fault experienced is exactly the same whether I run Linux or Windows.
You probably have said before but what do you set the heat bed to and the overhead heat temp when removing and replacing the cpu? Thanks for your time and videos
People send you working cards? It may seem like they work properly, but they don't always do. I have one of those. Works for a few weeks, then starts acting up and becomes unusable. Take the card out of the PC and use the integrated graphics for a while. Put card back, seems to work again. Starts acting up after a few weeks again. Had to send it back for warranty twice. Hopefully they'll fix it this time.
You probably have a motherboard issue with high resistance on the pcie slot contacts. When you pull the card and reinsert it improves teh connections for a while. Have you used contact cleaner and lube on the slot and on the card contacts? Same if you have a modular power supply, the connectors from the PSU to cables and for the power to the card could be causing the same, so clean those too.
Wow. I think you just talked me out of replacing my old 3080FE with a used 3090. Those "within spec" temps are insane. A 24 hour Furmark stress test has my mem temps at 78-80c unmodded, and I thought that was bad... No way those cards can survive long term at those temps. I'd like the extra VRAM, but it's just not worth it.
This may be dumb but i have always wondered this. What are the gold pins at the top of the board for? Not the regular pcie slot gold pins but the ones on the other side, at the top of the card when its on the preheating station. Im slightly tech savy and learning new things daily so i was just wondering.
@@danielhulan3058It is obsolete but yeah was still included on 20 and 30 series despite it not being useful for any games any more, technically it is called NVLink. I guess it still had workstation applications?
Initially 3090 wasn't designed with 24 memory ICs, but memory companies wasn't ready by the launch date with their 2Gb mem IC's that's is why nvidia has to urgently redesign the board and add another half of memory behind the card. This GDDR6X is very hot due to new 2-bit eye technology. But it becomes more problematic when you want to cool it on both side in sandwiched configuration. If you have money to buy 3090 u should buy also a double-sided waterblock. My watercooled 3090 has maximum 63C T degrees. I cant even imagine T when running on stock...
@@SlyNine well. Firmark. Firestrike And in general gaming Gaming its obvious very cool. Firmark is every I used to down clock my memory to a point where it was in the 80s max. I have a few profiles on MSI. one profile is minus 500mem. And like minus 500 mhz on the core. I use that's of the time. Because I normally play easy games. Like valhiem, diablo 2 re master... lately deep rock survivor And Pal World with my gf I pump it up when I play cyberpunk, etc
Yeah, this is why I buy branded known to be good graphics cards as opposed to those made by NVidia themselves. A lot of the "issues" of framerate differences can be contributed to cost cutting factory practices at the expense of quality control.
hello, im searching a kind of study to make the same job as you. Do you have recomendation ? Exept 5 year of ingenior, there is simple formation. What kind off study you recommand. Thanks
I really like your videos, i can relax while watching and learn something i'm never gonna use in my life lol. What was the problem at 20:06? one of the circles was grey, while the other ones where silver? was there a contact point coated or something?
On memory arrangement change it depends entirely on the GPU. In many GPUs all the memory chips are arranged in parallel to increase the bus width. In this case you can't. However, if you have two chips per a channel instead of one, you MIGHT be able to, depending on the pinning of the chip type and the ability of the board firmware to recognize the new arrangement. A FrankenGPU with workstation memory but a gaming card core is not possible because workstation card memory is typically parity memory and thus uses and extra bit line to ensure data integrity, at the expense of a tendency for slightly higher latencies.
you need to change the memory Strap resistors to tell the GPU with what ram its dealing with, same with the resistors for the PCI-e ID, so far i know you dont need to change the bios just these values, the PCIe ID mod was called tesla mod years ago, it was the way to get a professional card for cheap, today you can change the pcie id by modding the driver, in other videos you had a fancy layout of the board, may look for "STRAP" in these?
honnestly would looove to see if you do such things in a video i got the schematics for my notebook and im keep thinking to upgrade its GPU the same way
To answer the Q about the RAM swap, Yes you could put double capacity on one side, but the card will report ram failures for the missing chips... But alternatively, swap ALL the chips for twice the capacity for double the onboard RAM, and then as the RAM isn't pushed to the limit it would run cooler that way.
Wouldn't there be any other issues based on the configuration of the board? In other words, if the board was designed for 1GB RAM chips, mainly based on power delivery, but you are using 2GB chips, would the 2GB chips be using the same exact power or would they demand more?
What an interesting idea, never actually thought about that.. Swap out the memory & bump up capacity? All I know is that if there is Anything BIOS related.. Ask the Stilt's (easily found online =)
I know in the old days when you used to overclock video memory you would get artifacting because the transistors would be overloaded. So yeah I would suspect VRAM.
Hello! There was a modded 3090 with 48GB of VRAM, so yes, its possible to get 48GB of VRAM. Regarding using all the chips in a single side, I could be wrong, but I think you end up running with half the memory bus size, unless front and back chips share the same traces on the PCB.
Do you have a link for that (48GB mod)? I saw a video from Russian repair shop - they tried this and it didn't work. As they are quite experienced it would be really nice to see how somebody actually did that.
@@piotrsennov7392 hey! Spent multiple days searching for it to no success, only think I remember is that it was a chinese job done to make the card suitable for larger ML models.
15:15 - "i've got a good feeling about that that it will work..." 20:02 - failer failer 34:15 - the pads met the chips :D 36:36 - maybe EVen the EVga had pad pads :)
Im selling my 3090 and went back to my old 1080ti or who is my "ole reliable" power draw and heat on the 1080ti was great. Hoping battlemage is on par with the 3080 or so with less issues in regards to heat and power draw.
I would say that creating a 48 GB version has a higher chance of working properly, electrically, because leaving the pads empty on the back may lead to reflections and other interference on the very high speed data lines. But, the big problem is firmware and driver lockdown, which may make it impossible to use the extra capacity.
Absolutely fantastic. In the late 1970s, I first worked as a repair tech on Moog and Arp synthesizers. It is heartwarming to see people still doing component-level repairs, even more so to see it done without unnecessarily replacing components. I have heard of reballing an IC but have never seen it done before. Tony, thank you.
You are very wise not to be trying to repair them today... a world of hurt. Especially after teh owner has "had a go at recapping it".
@@paulperry7091 i live for this
Oh hell yes! I'm an analog synth guy. You're so lucky, and probably got to meet Robert Moog too! I also do component repairs on my own gear. I'm a big fan of the youtube channel Mr. Carlson's Lab for techniques and inspiration.
@@Nathriel I did in fact meet Robert Moog. He made a presentation at the technical school I attended, and he was talking about the future: digitial synthesis.
Pretty cool!
One of those random RUclips suggestions that pops-up and keeps me watching until the end! As someone who has done SMD work myself, I can say a very good job! Glad to see that there are stil component-level work being done by specialists in the throw-away society we live in!
Great video, thank you!
I don't know how popular the long format videos are but I do know that they show us a slight portion of all the work and effort that you put into every card into every repair Tony. I hope you had a nice lunch or dinner after that 😊
It's not always about the video's length. It's more about if the video keeps your interest and attention during it - so that you don't worry about how long the video is.
@@johnpaulbacon8320 People are less likely to click on a 40min video than on a say 10min video. It's just a larger time investment and the algorithm knows that. It's pretty common knowledge that longer form videos get less attention on YT comparatively.
Edit: I personally totally agree with your statement. I just wanted to point out that longer form videos are less viable on YT generally.
@Bassalicious Hello. I know that. For me subject matter is more important then length. Length is next. Thanks for replying and keep up the good work.
@@Bassalicious They see less traffic but tend to perform better. Longer form video's tend to take a larger time commitment, absolutely agree, so those clicking them usually have an idea of the content and are more likely to both watch it through fully and also comment.
Shorts for discovery, mid length (the 8-20 min vids) to get people comming back and long form for better revenue and audience retention/engagement.
(long form kinda games the system, particularly for premium viewers revenue share)
My ASUS 4090 Strix low voltage light showed up, tried it in two rigs, still the same. I would have taken it back to Microcenter, but knew getting a 4090 replacement was about impossible. I thought I'd roll the dice and send it back into ASUS for repair, big mistake. They wanted to charge me 500 more dollars than the price of the gpu to fix it, 40 dollars for shipping, and 20 dollars for laborto fix it. No negotiation whatsover, called a customer serivice call center that didn't care for my plight or the repair cost, told me basically tough luck, finally contacted the ASUS US CEO, (Reddit research gave me the address) and they replaced my card, They said they fixed it, but sent me a replacement instead. It's working now so theres that. Nice to know that there are people out there fixing GPUS instead of rip off repair centers. (edit on price of repair update lol)
ASUS repair centers are blackholes.
I will never buy anything Asus ever. I have only had to deal with their RMA once but it was hands down the worst experience I've ever had with a company ever. Just an awful company.
A few months ago GamersNexus made a documentary about RMAing the Asus Ally i think, and how badly it went. Reading your comment now comes with no surprise to me. I wish yall better luck with RMAing things.
Sorry to hear your struggle but you didn't give up & won. I've only had to rma one video card & that was a 6800gt gigabyte. Still daily an Evga 9800gtx+ Mint Evga.
@@neonlost Yep I once had to fight months to get my asus gpu back. They somehow "lost it". And btw didnt even inform me about that before I inquired a lot?? So they weren't going to do anything if I dint ask about it lol. I have had many GPUs, both nvidia and amd, and I have had to RMA 2 different GPUs, (only pc part that I ever had to RMA btw) and guess what, both were asus. From now on I avoid asus parts by all cost because I dont need that kind of headache in my life.
Nice repair as usual. I like the small pre-heater, a must for sure to avoid having to blast silly temps to a cold IC. I use my main pre-heater (same as yours, just still in one piece!). I place an aluminium plate on top and the IC goes there. The small one is convenient though.
my FE 3090 survived when i accidentally spilt tea all over my PC a few years ago. had to open it up and washed the pcb in dish soap, then dried it with isopropyl alcohol & replaced thermal pads; surprisingly it still worked even after accidentally knocking off a filter cap. this card has been an absolute champ for 3 years. the old skylake motherboard didn't survive however
You must think you’re cool using 666. I hope you know what that means.
Why would you pair a 3090 with a Skylake era CPU anyways, that would be the Thanos of bottlenecks lol
@@antonchigurh8343Maybe OP is planning to upgrade CPU but lack cash for MBD and CPU upgrade. Cash bottleneck is more common than PC bottleneck.
@@antonchigurh8343 from 2014 to 2020, cpu advancements weren't very progressive. was planning to upgrade eventually. ended up getting an i9 14900K after the old PC died and paired that with the 3090. the bottleneck for skylake wasn't too noticable
What do you mean CPU advancemens weren't very progressive at that time? Intel's biggest performance leap ever was from 7 to 8th gen when the number of cores increased by 50% from one gen to the other on the i7's
Hello i'm usually not writing comments but I've been watching your videos since you had 2k subs and I really enjoy ur Sorin's style but with GPUS :) .
Actually i'm really interested by your tools because i've seen that not only u are choosing good tools but also you have a lot of them that are homemade like the core lifter or the probe with the button to inject voltage only by pressing when we want and even your broom made or small wires that allows you to find traces on the board without the boardview. I would be really interesting if you do a video to introduce them all.
I agree with you and I hope @northwestrepair sees your comment
I've never heard of a 3090 memory upgrade mod but I'm sure it's probably possible given it is on just about every other model. The strap configurations are likely similar to the other models that have been modded.
It is definitely a mod for the 2080Ti (22GB), 3070 (16GB) and 3080 (20GB). I think I have seen Chinese sellers even selling pre-modded 2080Ti's and 3070's.
if the 3070 & 3080 mods works well on games & stuffs oh boy! that's a very tempting attracting value cards right there!!!
but there are firmware cap though iirc, nvidia being nvidia as usual
why aren't the memories replaceable using the new JEDEC format, its ridiculous nvidia keeps capping memory and selling it so much more overpriced
recently heard of 4090 surfacing in China with 48gb to
I use cans of freezer spray to help pinpoint faults. Has served me well for years.
Good vid. Like your practical steady hand skills.
Tony, you are a GPU professor, full of knowledge, and you know how to teach and explain. To be clear, there are people who know, but they are not like you, you know how to explain and that is the most important thing. A genius professor specialized in GPU repairs, and probably many more. People like that made me fall in love with electronics.
I've been a subscriber for both but Today i Learned Northridge repair and Northwest Repair are two different channels. I always thought the two were just different techs in the same channel lol
It is hard to say they even do the same work when you watch how they go about repairing stuff.
Nope the other one is ego driven that will make customers pay more when he knows they have money.
This guy is pretty humble.
@@Crawdoodle yeah i thought they were two people in one channel. I honestly prefer this over the other guy. He seems a little too angry all of the time
@@kitsumyr9752 Tony here is "real" as in one of us. He tries to make the world better. the other guy is hard on about selling stuff and gives up pretty easy with "no fix" if its not a connector or mosfet failure.
what a hell of a repair. awesome work! what we'll learn is : almost everything what gets broken could be fixed
The 3090 has memory on both sides of the board. I fixed this by putting a heatsink and fan on the backplate. I used mine for mining, so I really had to manage those memory temps. In the end, the mining paid for the card and then some. Mine was not a FE, it was a Gigabyte and had a metal backplate.
When trying to find where rubber thermal pads are NOT touching the heatsink, a piece of glass is a great tool.
You can see which ones touch the glass just by pressing it against against the board.
How does that help if boards have multiple levels?
@@juliusvalentinas use a small piece of glass, like a microscope slide
The bigger downside to halving the number of chips (the possibility to even do it due to firmware restrictions not withstanding) is the reduction of the bus width, in effect halving the advertised memory bandwidth. And those chips crave memory bandwidth.
no, because 3090 use clamshell configuration, the chip on the front and the back shares the same 32 bit bandwidth
I'm sure Jensen Huang will back his product.. as long as you bring the vasoline...
Jensen LOVES his customers. It's rare to actually find people who pay for the privilege of being r@p3d and then come back for more, what's not to love? Of course, lube is an added upgrade because rectal bleeding is not considered a bug 'butt' rather a feature!
AMD FANBOY WANTING LISA SU FISTING DETECTED
AMD PEASANT BUDGET GAMER BURGER FLIPPER DETECTED!@@georgehubbs1747
@@cesaru3619 I'm sorry, I couldn't here you. Jensen left something on your chin. 🖕
@@cesaru3619Hey femdom can be pretty fckin hot! 😂
Measure the power consumption. If excessive, the unit is putting out more heat. Of course, good compression of thermal pads and good wetting of thermal paste is important as well. I work more with audio than computer, but it's not uncommon to have an OP amp drawing abnormally high current and running hot, even though it functions perfectly. You might even invest in a FLIR camera, that would let you single out the hot running chips.
i think they are all covered up while running so as they have one combined heatsink it would not be FLIR useful?
I did not expect to spend nearly 40 minutes of my early morning watching a GPU repair video, but I was enthralled. Awesome stuff.
I love your attention to detail. Excellent video. Thank you for posting.
Hey there. I recently subbed. Very good repairs and attention to detail. I am also doing BGA rework professionally, but i must admit, you achieve 5 star results with the basic tools you have(mad skillz). I would recommend that you look into tools like PDR infrared rework stations, Finetech rework stations, and you deffinitely need a Martin mini oven for those bga chips. Your attention to detail and determination is very motivating. All the best!
If you are looking for reballing machines, have a look on the IR8500 (entry level) or a Honton R490/R690 (more “professional”). The IR is not that expensive, the Honton cost a bit more but might be worth for a person doing this for a living. Plenty of videos on youtube about both models. Hope that helps. Kind regards
Nice work as always Tony. I'll post a link to a DIY re-balling station you may or may not have seen on your DC. In respect to the VRAM increase or 2GB modules on the fron...I'm intriqued myself. Great job and thanks.
You are amazing! Your repair skill is second to none.
Krisfix Germany is about on par, also a GPU specialist. But yeah, these people are wizards, literally.
Forget about the memory, would be a miracle if these pricks at nvidia had forgotten to make it impossible, somehow.
Easy in the 1990s, impossible now, mostly because of cooperate greed.
you speak nonsense, u can edit the memory for 3000 series, theres are 3070 with 12gb, 16gb, 3080 with 20 gb, 3080ti with 24gb... if someone finds in bios how to disable the back memory we can do it if it wasn't doing done already
@@Kiburici Yes, with 48GB and Quadro bios, it would be a Quadro RTX A6000.
Apparently, flashing GTX to Quadro with resistor modding or something wasn't impossible until not too long ago, GTX690 to K5000, I think, then Nvidia got greedy and decided to do some laser lobotomy, so RTX will never be a Quadro ever again.....
@@hinz1 you can still do it with RTX cards what are u speaking about
@@Kiburici 20 and 24GB 3080/3080Ti are actually made on 3090 PCB, you can't use 16Gbit GDDR6X on 3080/3080Ti core
I have old gpus with upgradable socketed memory
Very interesting video. Had a motherboard with a gpu on it that needed exactly this fix (gpu pulled and re-soldered). A common issue with this mb. Was able to heat gpu and got it working for a short while. ThX for sharing.
Enjoyed this video, I’m going to check my temps using the same method on my sapphire nitro plus 7800xt because I’ve spotted some inconsistent temps between hotspot and memory junction, I did very carefully check the tension on the retention plate 4 screws thingy and got about a millimeter worth of movement out of them, this made a difference of a few degrees on hotspot temps then I adjusted the fan curve to keep things in check, will now check those memory temps using gpuz and furmark and compare them to the hotspot 👍🏻
The small PCB of 3090 FE makes the memory temp situation even worse, throttling the GPU clock down the drain. This card begs for a water-block.
The 4004 CPU’s with 4K of memory I was repairing back in 79 weren’t quite this complicated. Switching in the paper tape boot loader on the front panel switches was the biggest pain.
Great job finding the fault. It was a bit surprising that you decided to fast-forward through the discovery of it though.
did not know about MSI's artifacting feature, thanks dude! I like to run these tests myself, just to make sure :D
Fell asleep watching your videos and had a dream that I cracked the board on my 3090 haha
Lol
When we check piston to valve clearance in an engine we use plasticine and measure how much it squashes, I guess you could glue down little balls of alum foil and measure the height after a dummy fit.
plastic gauge is far more commonly used for good reason no? It won't scratch or damage.
@@AppriIt's more for the even squishing properties. Things that would scratch or damage would not squish so well.
@@MSFSFreeware Both true I suppose lmao
@@AppriI also love the way they talk about measuring the height, rather than the squish width lol - I can see those aluminium balls not squishing and causing some snappage also !
The quadro version of that gpu does indeed have 48 GB VRAM
Yes. The RTX A6000. The Quadro name was dropped. The RTX A40 has the same GA102 chip with slower GDDR6 and passive cooling.
@@Ranguvar13 make a ramdisk and run windows off that xD
nah rather use RAM as cache for ZFS@@SArthur221
another great job. if you can't fix no one else could.
Ahhh!! Another long format video! Much better! Better then that!! Your custom core lifter is baller
Dude. I want one. Lol. I saw at one point you tried an ir plate above. Hot air better distribution?
totally awesome work there. it is always cool to watch someone like you work. get some food.
I used to own this GPU, I replaced the pads on memory, which fixed the memory heating up so much. I actually had memory junction temps of 110 degrees celsius! After replacing the pads the core even was able to go into thermal throttle rather than it throttling on memory.
What was the temp on mem afterwards?
@@dtiydr around 85-90
@@arromger Can not understand how the memory can get that incredibly hot.
LOVE the longer videos, people can skip around if they like shorter videos...
You do this stuff couse you enjoy doin it, no rush.
I am about 2 years into self-teaching myself on doing repairs and I have to say I really wish I had the tools used in this video lol. Unfortunately, I am not able to work a normal job right now witch also means I am broke lol and I do not have the money to get tools like you see here and even the tools I do have and can afford I have been really just making the crap tool I have work. Anyways cool video thanks you for the entertainment and wisdom.
Just for reference my 3090 FE, just like in this video, has the Memory running at 90c under load.
Thinking this was an issue on my 3090 I found that the memory, has a Max temp of 105c, before its an issue on the 3090.
This is the design that NVIDIA did with this generation, I believe due to the GDDR6x chips.
So seeing the 90c+ on this 3090 FE, is not a surprise to me.
If you want it lower, then its at best memory pads, replacement from the crap (white straw) pads that NVIDIA does at the factory.
My card ran fine, and did very well in Microsoft Flight Sim 2020, at max FPS.
Those crappy pads were the sole reason people started trying various thermal pads and even copper shim mods. Especially under backplate, where some individual GPUs had no pads at all (this was during covid, so there were possibly shortages or workers just "forgot" to add them)
The higher temp in 30 series is due to Nvidia using Samsung node if they were gone with tmsc it would have been lower
Just because they say those temps are within spec doest mean it's safe for longevity of the card life/board its attched too including excess heat on the core from behind. I would always advise to watercool the card on front an back plate for 3090 and 4090 series gpu just because of memory chip temps are way too high with any stock cooler on for my liking ,since watercooling my temps on memory under load are maxed at 39 after long periods at full load
On the 3090s the mem temp being so high is due to half the chips being on the back side of the pcb. Almost no cooling. The sensors will always report the highest temp of all.
ive got a 3090 fe and ive repadded its gone down from 105 to around 86-90 depending on the case its sat in , this is as good as it gets for the 3090 FE. Also i came across something on the interwebs which states theres 2 revisions of the card and depending on which one it is depends on the thickness of the pads required. i used artic tp3's 1.5mm on mine and its been fine ever since :D
20:08 is the time you are looking for
It's crazy to think that this wouldn't be covered under Warranty where you are.
Here in Australia I think government demands 3-5yr for computer electronics.
Use to deal with a lot of these grey pads (dry solder joint) back in PS3/XBOX360 days.
This level of repair is rarely worth it unless device is REALLY expensive.
It's just so amazing to me. The simplicity of it all, but how say a piece of foam makes makes all the difference in the world. I didn't realize the alignment of a heat sink was that important.
One problem with some of these boards is that they really need a brace or to be run vertical due to weight. I've seen many GPU boards that have a suspicious amount of flex in them when installed. One old card of mine you could see light between the pad in a couple of places as it had simply pulled away too far.. My son's computer was remedied by putting in a long zip-tie to the chassis to relieve some of the excess weight.
I remember my PS3 came with a factory defect... They told me to send it in for a replacement... 17 years later im still waiting for that replacement.
Oh no Tony's bringing the heat!
When you are constantly trying to make things cheaper stuff like this is bound to happen. Like the pads that are about as thick as a piece of tissue paper in the name of saving a few pennies on copper. They clearly are designed more for one time soldering than repairability but the fact that they are pulling off the boards merely from uneven mounting pressure shows they are not even sufficient for that.
Good video, Nvidia's quality control is great as we can see.
Nice video. I like your professionalism a lot.
To change the memory sizes on the GPU, you must have access to a schematic from the board (not public available) and be able to reconfigure the jumpers on the board (0 ohms resistors or not, I saw previous models worked by percentages in drop voltages) and may be some cases changing the memory space inside the GPU bios.
You should change the name of the channel to "GPU Repair GOAT"
You make reballing look so easy.
Wow, it's like they made a special "rub off here for no returns" pad! ;)
SPEEHLESS. Simply stunning ......
Hello firts thank you gor squematics and show this defect in thr GPU NVIDIA nice day and greetings from Mexico
very detailed as always Tony, thx to You I have learned to repaire grafic cards and boards in general. Been asking in last few videos, what kind of new software for memory are You using ?
Man that thumbnail drew me in bravo!!!!
10:18 you can add a 3d printed impeller and diverter to your output to achieve the spread of hot air. A changeable diffuser nozzle would allow for directed heat versus diffused heat for specific applications.
Speaking from experience, the completely working gpus you get sent will likely trace back to a failed PCI-X socket. Had one of those in the other day. These stupid cards or the MB/case's they put them in put too much stress on the slots, and the slots fail.
Thanks for the class time friend!
3090 memory getting super hot is so common, passive backplate was never really enough cooling. When I had a founder's edition 3090, I put a waterblock on each side and that's what it took to get decent memory temps.
Why are you reballing? The issue is most likely between the die and BGA package, not the package and the PCB. The thermal stress from reballing might solve the problem temporarily, but you're charging the customer for a fix that probably won't last.
3090's memory temps are always insanely high when in use. they can even go up to 100c bcs there is no active cooling on the back of the card, where the other memory modules are
I redid all the pads on my 4 4080s. I can not remember the sizes but you can find them on many crypto mining sites. The new pads made a huge difference in memory temps for me. like 15-20 degrees under full load.
Great video that was full of helpful information. So thank you. Do you know where I can buy power harnesses like yours to power the card? Also, that tool that you have to check power with two screens, did you make that and is it something I can buy? I really am a beginner and I'd like to get set up on a bench properly.
I've seen enough reballs to know that you would be almost "out of business" if they made sockets for the chip on the GPUS like on CPU mobos. And gpu vendors would probably starve to death.
Reball still get the job because the pins will get damaged by a careless users 😂
@@ariewijaya1679 yeah he's underestimating how many people would still break something. Or the fact people are terrified of opening up their 2000+ dollar gpus themselves
Great repair!!! I feel the struglem but that"s the way!!!
I've seen a GPU case where it would 100% pass furmark with zero artifacts or crashes but random polygons would disappear in games, it was annoying AF.
All memory was OK. I traced the anomaly to vCore fluctuating like crazy. Unfortunately the card died with a probe slip before I could fix it or even determine if it was defective core or vrm.
vrm controller swap didn't help.
Great Details-Thanks
As long as you're there, a few strips of soft copper on those mem chips could fill the gap and conduct better than extra thermal tape ?
nah this dude its a next level fixing god
A few levels above the "NO FIX" guy...
@@Nordic-sm3urthat "no-fix" guy does not fix only replace. 😂😂😂
@@fixeuskeane ...and occasionally blow up customer hardware by injecting 3V into a laptops CPU and charging the customer a fee for him destroying it live on stream, then going on a banning spree for anyone pointing that out 🤡
Even though they don't seem to likely cause any contact where it shouldn't occur, when you switch off the little hearing plate there still seem to be little traces of solder paste inbetween the points the should be at, almost looking like dendrites (14:44, left corner, a bit earlier also right corner)
Looks great, I have a 1060 GTX card that is acting up in a similar way. Normal 2D works fine, but if you run a 3D-intensive game, like the System Shock Remake for about 15 minutes or so, the whole system hangs. It's not a driver issue, because the fault experienced is exactly the same whether I run Linux or Windows.
You probably have said before but what do you set the heat bed to and the overhead heat temp when removing and replacing the cpu? Thanks for your time and videos
Awesome video, with good commentary.
People send you working cards? It may seem like they work properly, but they don't always do. I have one of those. Works for a few weeks, then starts acting up and becomes unusable. Take the card out of the PC and use the integrated graphics for a while. Put card back, seems to work again. Starts acting up after a few weeks again. Had to send it back for warranty twice. Hopefully they'll fix it this time.
You probably have a motherboard issue with high resistance on the pcie slot contacts. When you pull the card and reinsert it improves teh connections for a while. Have you used contact cleaner and lube on the slot and on the card contacts? Same if you have a modular power supply, the connectors from the PSU to cables and for the power to the card could be causing the same, so clean those too.
@@bbahunter6436 Wasn't that. The card is fixed now (they didn't say what it was, though).
Hello great video absolute wizardry what temps do you use to remove a core?
Depends
Brill repair, thanks for the info.
Wow. I think you just talked me out of replacing my old 3080FE with a used 3090.
Those "within spec" temps are insane. A 24 hour Furmark stress test has my mem temps at 78-80c unmodded, and I thought that was bad...
No way those cards can survive long term at those temps. I'd like the extra VRAM, but it's just not worth it.
It would probably drop temperatures down as far as migrating all the chips to the front. But it would probably reduce memory bandwidth.
my guy is playing battlefield 1, you got my like
Since 3000 series the memory temps are a joke.
In like 5 years we will possibly see a lot of them actually die from silicon ageing.
This may be dumb but i have always wondered this. What are the gold pins at the top of the board for? Not the regular pcie slot gold pins but the ones on the other side, at the top of the card when its on the preheating station. Im slightly tech savy and learning new things daily so i was just wondering.
Oh wait is that for sli? I thought that was obsolete.
@@danielhulan3058It is obsolete but yeah was still included on 20 and 30 series despite it not being useful for any games any more, technically it is called NVLink. I guess it still had workstation applications?
@@danielhulan3058Nvidia calls it NV-link now, but still possible on xx80 and xx90 cards
The amount of work you put into it is worth more than the card probably
Initially 3090 wasn't designed with 24 memory ICs, but memory companies wasn't ready by the launch date with their 2Gb mem IC's that's is why nvidia has to urgently redesign the board and add another half of memory behind the card. This GDDR6X is very hot due to new 2-bit eye technology. But it becomes more problematic when you want to cool it on both side in sandwiched configuration. If you have money to buy 3090 u should buy also a double-sided waterblock. My watercooled 3090 has maximum 63C T degrees. I cant even imagine T when running on stock...
I just run a slight undervolt and about minus 80 on the ram. And its plenty cool, as well as out preforming stock settings.
I have an ABI card not FE
@@fepethepenguin8287what were you running to test it?
@@SlyNine well. Firmark. Firestrike
And in general gaming
Gaming its obvious very cool.
Firmark is every I used to down clock my memory to a point where it was in the 80s max.
I have a few profiles on MSI. one profile is minus 500mem. And like minus 500 mhz on the core.
I use that's of the time. Because I normally play easy games. Like valhiem, diablo 2 re master... lately deep rock survivor
And Pal World with my gf
I pump it up when I play cyberpunk, etc
Thermopro BGA Reworking Soldering Center is a lifechanger. Worth the money! And try liquid thermal pads.
Yeah, this is why I buy branded known to be good graphics cards as opposed to those made by NVidia themselves. A lot of the "issues" of framerate differences can be contributed to cost cutting factory practices at the expense of quality control.
hello, im searching a kind of study to make the same job as you. Do you have recomendation ? Exept 5 year of ingenior, there is simple formation. What kind off study you recommand. Thanks
I really like your videos, i can relax while watching and learn something i'm never gonna use in my life lol. What was the problem at 20:06? one of the circles was grey, while the other ones where silver? was there a contact point coated or something?
On memory arrangement change it depends entirely on the GPU.
In many GPUs all the memory chips are arranged in parallel to increase the bus width. In this case you can't.
However, if you have two chips per a channel instead of one, you MIGHT be able to, depending on the pinning of the chip type and the ability of the board firmware to recognize the new arrangement.
A FrankenGPU with workstation memory but a gaming card core is not possible because workstation card memory is typically parity memory and thus uses and extra bit line to ensure data integrity, at the expense of a tendency for slightly higher latencies.
As a note, you CAN upgrade the RAM in many smart devices, although most require you to mod the firmware to recognize it.
you need to change the memory Strap resistors to tell the GPU with what ram its dealing with, same with the resistors for the PCI-e ID, so far i know you dont need to change the bios just these values, the PCIe ID mod was called tesla mod years ago, it was the way to get a professional card for cheap, today you can change the pcie id by modding the driver, in other videos you had a fancy layout of the board, may look for "STRAP" in these?
honnestly would looove to see if you do such things in a video
i got the schematics for my notebook and im keep thinking to upgrade its GPU the same way
To answer the Q about the RAM swap, Yes you could put double capacity on one side, but the card will report ram failures for the missing chips... But alternatively, swap ALL the chips for twice the capacity for double the onboard RAM, and then as the RAM isn't pushed to the limit it would run cooler that way.
Wouldn't there be any other issues based on the configuration of the board? In other words, if the board was designed for 1GB RAM chips, mainly based on power delivery, but you are using 2GB chips, would the 2GB chips be using the same exact power or would they demand more?
What an interesting idea, never actually thought about that.. Swap out the memory & bump up capacity? All I know is that if there is Anything BIOS related.. Ask the Stilt's (easily found online =)
I know in the old days when you used to overclock video memory you would get artifacting because the transistors would be overloaded. So yeah I would suspect VRAM.
Hello! There was a modded 3090 with 48GB of VRAM, so yes, its possible to get 48GB of VRAM.
Regarding using all the chips in a single side, I could be wrong, but I think you end up running with half the memory bus size, unless front and back chips share the same traces on the PCB.
Do you have a link for that (48GB mod)? I saw a video from Russian repair shop - they tried this and it didn't work. As they are quite experienced it would be really nice to see how somebody actually did that.
@@piotrsennov7392 hey! Spent multiple days searching for it to no success, only think I remember is that it was a chinese job done to make the card suitable for larger ML models.
furmark 2 has a artifact scanning tool, don't know how well it works fr though
yes but it no longer supports custom resolution
15:15 - "i've got a good feeling about that that it will work..."
20:02 - failer failer
34:15 - the pads met the chips :D
36:36 - maybe EVen the EVga had pad pads :)
i got rid of my 3090 because of the memory heat
replace the awful stock thermal pads with 1.5mm Thermalright pads - MJT rarely goes above 100c
any 3000 series with GDDR6X run that hot. I have 3080 FE similar memory temps when maxed out.
Im selling my 3090 and went back to my old 1080ti or who is my "ole reliable" power draw and heat on the 1080ti was great. Hoping battlemage is on par with the 3080 or so with less issues in regards to heat and power draw.
I would say that creating a 48 GB version has a higher chance of working properly, electrically, because leaving the pads empty on the back may lead to reflections and other interference on the very high speed data lines. But, the big problem is firmware and driver lockdown, which may make it impossible to use the extra capacity.
Why is it running at 1350mhz? That's a fail safe clock speed
Because it’s furmark. All GPUS run at “fail safe” clock speeds when running this benchmark in order to protect the core.