NVIDIA a dit NON, la Chine l'a fait : doubler la VRAM de ses RTX !

NVIDIA a dit NON, la Chine l'a fait : doubler la VRAM de ses RTX !

NVIDIA said NO, China did it: doubling the VRAM on its RTX cards!

Title history
  • NVIDIA a dit NON, la Chine l'a fait : doubler la VRAM de ses RTX ! — observed on 20261105-1655 UTC (YouTube)

by MAKERTRONIC TECH · View the channel on PSI

Applied Sciences & Engineering Computing & Cybersecurity UKComputer hardwareUKRComputer and information technologies hardware: Maintenance and repairs
🎙 MAKERTRONIC TECH 👥 6K 📅 November 5, 2026 ⏱ 17 min 👁 196 📄 expert opinion 🧭 2026-11-05
Available in: English (current) Français

Keywords

VRAMGPUShenzhenNVIDIALLM

Summary

The video explores the growing market of GPU VRAM modification in Shenzhen, China, where workshops upgrade consumer graphics cards like the RTX 2080 Ti, RTX 3080, and RTX 4090 with significantly more memory than NVIDIA officially offers. The presenter explains that NVIDIA deliberately limits VRAM on consumer cards to push users toward expensive professional models, creating a demand for unofficial modifications. The technical process involves desoldering memory chips (BGA), replacing them with higher-capacity ones, adjusting hardware straps (resistors) to configure the memory controller, and flashing a modified vBIOS. For the RTX 4090, a more radical approach is used: the GPU die is removed and placed on a custom PCB with 24 memory slots in a clamshell configuration, achieving 48 GB. The video evaluates performance for local LLM inference, comparing prices and capabilities: a 2080 Ti with 22 GB costs 300-450€, a 3080 with 20 GB around 480-550€, and a 4090 with 48 GB around 2000-2400€. While these cards offer considerable value for running AI models, the presenter highlights several risks: unreliable BGA soldering, recycled memory chips, overheating of backside memory in clamshell mode, potential driver issues, and lack of warranty. The conclusion is cautious, weighing the cost-effectiveness against the high chance of failures and the lack of official support.

209 words

Critical Evaluation

Value of the Information & Strength of the Argument

The video provides valuable insights into a niche hardware market, explaining both the technical feasibility and practical implications. It argues convincingly that NVIDIA’s segmentation policy creates a demand for gray-market solutions, and the reasoning for the modifications is well-structured. The valuation of these cards is supported by concrete price comparisons and performance expectations for LLM inference. However, the argumentation is largely anecdotal, lacking quantitative benchmarks or official data. The presenter relies on personal experience and hearsay, which weakens the overall persuasiveness but still offers a credible overview for enthusiasts.

Scientific Rigor, Source Quality, Title Accuracy

The video demonstrates a good level of technical accuracy in describing the modification processes and memory specifications. However, it does not cite any external sources, academic papers, or official documents; the information appears derived from personal experience and market observations. The title accurately reflects the content and does not exaggerate the scope. Scientific rigor is moderate: while the technical details are plausible and consistent with known practices, the lack of verifiable references and the absence of reproducible data limit its reliability. The creator’s credibility is enhanced by practical knowledge of hardware and AI, but the lack of citations prevents a higher reliability score.

207 words

Title / Content Match

The title effectively conveys the video's focus on Chinese workshops circumventing NVIDIA's VRAM limitations, thoroughly matching the content.

Quality & Reliability

7/10

The video provides a detailed technical explanation of GPU VRAM modding, including BGA soldering, strap adjustments, and vBIOS flashing, along with market prices and risk assessment. However, it relies on personal experience and general observations without verifiable sources or independent benchmarks, which limits its scientific rigor.

Chapters

Contribution & Novelties

This video contributes by demystifying the hardware modding scene in China, particularly the techniques and market dynamics behind GPU VRAM upgrades. It provides a practical guide for enthusiasts considering such purchases, including pricing and risks. The originality lies in connecting technical details with economic incentives and real-world usage for AI.

Pour aller plus loin :

  • Ball grid array — Explains the BGA soldering method used to attach memory chips.
  • Video BIOS — Details the role of vBIOS in GPU configuration and potential modification.
  • Graphics processing unit — Overview of GPU architecture and memory management, useful for understanding VRAM limitations.

99 words

Radar Profile

The radar profile suggests a video with high technical depth and moderate quantity of information, but lower reliability due to unsourced claims. The balance between quantity and quality is good, while the technical level stands out, and global reliability lags slightly due to missing citations. This is typical of expert opinion content without formal validation.

Reliability 6/10