I Doubled My GPU’s Memory…

The Problem: Cloud AI's Grip on Personal Computing [0:00]

The speaker highlights the increasing reliance on cloud services for computing, where personal data is vulnerable and AI tools are often costly. This has fueled a surge in interest for local AI solutions and self-hosting, prompting the creation of a powerful home machine.

"in short the man wants you to forget that the p in pc stands for personal and a lot of people are getting fed up about it"

The Solution: Doubling VRAM on RTX 4090s [1:00]

The project involves upgrading two NVIDIA RTX 4090 graphics cards from 24GB to 48GB of VRAM each, totaling 96GB. This significant increase allows for the running of large AI models that would typically require professional-grade, much more expensive hardware.

"i got insane deals on a pair of 4090s and i'm going all the way to michigan to upgrade them not replace them upgrade them from 24 gigs to 48 gigs each giving me 96 gigs of vram across both cars"

Expert Hands at GPU Lab [2:00]

The upgrade process is undertaken at GPU Lab, with the speaker entrusting the delicate work to Greg, the expert behind the operation. The initial step involves preparing the cards by drying them to remove moisture.

"greg the man behind gpu lab will be doing all of the hard work so greg where do we start"

The Upgrade Process: Memory Module Transplant [4:00]

The core of the upgrade involves removing memory modules from donor boards (sourced from China) and transplanting them onto the user's PCBs. This intricate process requires precision, starting with flattening the solder pads on the target boards.

"we're taking boards from china adding some memory modules from china bought in bulk on a tray we're taking the memory off of my boards taking the the core off of my board these are two different cards there's like a founder's edition and i think it was a gigabyte board and then transplanting all of that onto that that chinese pcb"

Precision Soldering Under the Microscope [6:00]

Each memory module is meticulously placed under a microscope, ensuring all solder bumps are perfectly aligned and flattened. This step is crucial for a successful connection and requires careful attention to detail.

"we want to basically apply thin layer flux and then put the memory modules one by one under a microscope i look under a microscope just to make sure all the solder bumps are flattened"

Reflow and Cooling [8:00]

Once the memory modules are in place, they are soldered onto the circuit board using a BGA rework station. The process involves a specific reflow curve to ensure proper adhesion, followed by a controlled cooling phase.

"this is our bga rework station this is used to apply memory modules to our new circuit boards technically in china they do it one by one you could do it on the same machine that we would extract memory models from but here i have special nozzles that basically allow it so i could solder on 12 new memory modules all at the same time under the same reflow curve"

Sourcing Memory: The Micron Shortage [10:00]

A challenge in the process is the scarcity of Micron memory chips, which are no longer being produced. This has led to increased prices for available stock, necessitating the use of tested memory modules from boards with faulty cores.

"but micron isn't making those anymore no about december they stopped the whole market dried up and it was very chaotic all the third-party sellers of new old stock completely clutched all of their memory modules and they were listening for ridiculous prices"

The Speaker's Turn: Placing Memory Modules [12:00]

The speaker gets hands-on, attempting to place memory modules under the microscope. Despite initial slowness and difficulty, they show improvement and successfully place several modules.

"i can't see anything uh so adjust these like this wow that's not the easiest great so if you want to take some flux like that yeah just kind of like sparsely like i did not like yeah it's like the spotty yep that's perfect"

Core Removal and Preparation [15:00]

After the memory modules are attached, the next critical step is removing the main GPU core from the donor board. This is done using the BGA rework station, carefully lifting the chip once the solder balls reach their liquidus state.

"so now we're going to take off the core and basically i got to change out the nozzles these ones were for applying membrane modules so this is the gigabyte card the gigabyte card was given to me by mr dank pods wade will uh we'll see if it survives"

Reballing the GPU Core [18:00]

The removed GPU cores are then prepared for reballing. This involves cleaning the existing solder bumps, applying flux, and then using a stencil to precisely place new solder spheres onto the pads before reflowing them.

"we have like one that like merged with another ball so now i just sucked up that double ball and i'm going to clean it up and i'm going to put another few set of spheres on it because it's still hot i could still rework with it and it'll be fine"

Final Assembly and Testing [22:00]

With the memory modules and GPU cores upgraded and reballed, they are reattached to the new PCBs. The speaker then installs a cooler, and the card is plugged in for testing using Vulcan MemTest to ensure functionality and check for errors.

"so this is the display port cable that goes to this monitor right here this is our waiting for vga led the orange one once that goes away we're going to get an image oh let's go posting now around windows"

Success and Future Services [28:00]

Both upgraded 4090s pass testing, confirming the success of the modification. Greg mentions ongoing work on other projects, including 170HX mods and potential A100 conversions, and promotes his website, GPU Lab.

"they have to sit in the oven for 12 hours so i'm not actually going to be taking my cards home greg will be shipping them to me later but that's it we're done they both work you did 99.9 of it but how'd i do"