April 24, 2026

The RTX 3070 gets a second life - doubling VRAM with a bare-metal mod

The RTX 3070 gets a second life - doubling VRAM with a bare-metal mod
What the mod actually involves
Why Nvidia tried to stop it
Does it actually work?
 
Article
What do you do when you need more VRAM but can't afford a new GPU? If you're the YouTuber Vex, you grab a soldering iron and do it yourself — Nvidia's wishes be damned.

In a nearly hour-long video, Vex walks through a full hardware modification of an RTX 3070, upgrading its onboard GDDR6 memory from 8GB to 16GB. The result is a card that punches well above its original spec — and a repair odyssey that is part engineering challenge, part comedy of errors.
What the mod actually involves
The process isn't as simple as swapping chips. It requires removing the existing VRAM modules with a hot-air rework station, cleaning and re-tinning the pads, reballing new GDDR6 chips using a stencil and solder balls, and then reflowing them onto the PCB with precision. Any misalignment or cold joint can render the card completely dead. Vex documents all of this across two days of work, complete with setbacks, emergency plans, and multiple rounds of rework.
Why Nvidia tried to stop it
The title is a nod to Nvidia's practice of shipping the same GPU die in multiple configurations, artificially limiting memory capacity to segment the market. The 3070's GA104 silicon is capable of more — Nvidia simply chose not to fully populate it. This kind of mod exposes that gap and lets savvy users reclaim hardware capability they technically paid for in silicon, even if Nvidia would prefer they buy upward instead.
Does it actually work?
After a second round of rework, multiple boots, and a war of attrition with the reballing process, Vex gets the card running. The final section of the video covers real-world performance testing, showing what a 16GB RTX 3070 can do — particularly in scenarios where 8GB would otherwise cause stuttering or VRAM overflow. The video is a worthy watch for hardware enthusiasts and anyone frustrated by artificially tiered GPU specs. It's messy, honest, and genuinely impressive — the kind of project that reminds you what's possible when you're willing to risk a GPU on a soldering iron. Watch it here: YouTube
AI was used to help create this content.
Written by Marius L
The creator and owner of Hashrate.no goes by the alias r0ver2. With years of hands-on experience working with GPU hardware, he started building and configuring his own systems in 2017 — gradually scaling from a home setup to a larger multi-GPU operation, gaining deep technical knowledge of hardware management, power delivery, thermals, and system stability along the way.
Last updated: April 23, 2026