MI Developer Buys NVIDIA Grace Hopper with 960GB RAM for €7,500
- An AI developer secured a high-performance workstation equipped with NVIDIA Grace Hopper superchips for €7,500, a significant discount from its estimated market value of tens of thousands of...
- AI developer David Noel Ng recently acquired a workstation featuring NVIDIA Grace Hopper superchips at a remarkably low price of €7,500 (approximately $8,100 USD as of December 16,...
- The workstation's core comprises two NVIDIA Grace Hopper superchips, paired with two Grace processors and two Hopper H100 GPUs.
“`html
AI Developer Acquires NVIDIA Grace Hopper Workstation for a Fraction of Its Value
Table of Contents
An AI developer secured a high-performance workstation equipped with NVIDIA Grace Hopper superchips for €7,500, a significant discount from its estimated market value of tens of thousands of euros. The acquisition required substantial technical expertise and effort to restore functionality.
The Unexpected Find
AI developer David Noel Ng recently acquired a workstation featuring NVIDIA Grace Hopper superchips at a remarkably low price of €7,500 (approximately $8,100 USD as of December 16, 2025). This is a substantial discount considering comparable systems typically cost tens of thousands of euros. Ng described the purchase as akin to “finding a piece of gold in a trash can,” but acknowledged the significant work required to make the system operational.
Inside the Workstation: A Powerhouse of Components
The workstation’s core comprises two NVIDIA Grace Hopper superchips, paired with two Grace processors and two Hopper H100 GPUs. A key component driving the system’s value is the 960 GB of LPDDR5X RAM, often referred to as “golden” memory due to its high performance and cost. This configuration delivers substantial computing power, making it ideal for demanding AI workloads.
| Component | Quantity | Description |
|---|---|---|
| NVIDIA Grace Hopper Superchip | 2 | Integrated CPU-GPU architecture for high-performance computing. |
| NVIDIA Grace Processor | 2 | Arm-based CPU designed for data center applications. |
| NVIDIA Hopper H100 GPU | 2 | High-end GPU optimized for AI and HPC workloads. |
| LPDDR5X RAM | 960 GB | High-bandwidth, low-power memory. |
Restoration Challenges and DIY solutions
The workstation wasn’t immediately functional upon purchase. Ng reported extreme challenges, including a GPU temperature reaching a reported “16 million degrees Celsius” (likely hyperbole indicating a severe overheating issue) and the necessity of manually soldering components. He humorously suggested that a sponsored post likely wouldn’t allow such candid descriptions of the restoration process.
Ng employed a resourceful approach to cleaning and debugging, utilizing isopropyl alcohol, four refurbished Arctic AiO liquid coolers, custom-milled copper parts created on a CNC machine, and a significant quantity of 3D-printed components (kilograms worth). This extensive work highlights that the price difference wasn’t simply a lucky find, but rather a “compensation” for the considerable skills and effort invested in bringing the system back to life.