Blog
Just hours before the launch of Starship Flight 11, NVIDIA CEO Jensen Huang handed over the DGX Spark (the world’s smallest AI supercomputer) to Elon Musk at SpaceX headquarters!
💻 DGX Spark
- Weight: Just 1.2 kg — the size of a small desktop machine.
- Performance: 1 petaflop of computing power (a trillion operations per second on your desk).
- Processor: Grace Blackwell GB10 Superchip with 128GB unified memory and data transfer speeds 5x faster than PCIe Gen5.
Let’s unpack this carefully 🤔
✴️ Power consumption: DGX Spark uses only 240 watts, with a purchase cost under $4,000. It can run a 200-billion parameter model locally. Compare that to the monthly cost of Claude Max (around $200), plus OpenAI subscriptions — in one year, you’re likely to spend more than the cost of the device just on API usage.
✴️ It offers privacy advantages by enabling full offline operation — no need for the cloud.
✴️ It’s like having a consultation with the Oracle, without plugging into the Matrix, while using the energy equivalent of just 4 lightbulbs.
✴️ For context:
- The first DGX (in 2016) cost around $129,000,
- Weighed over 100 lbs (45+ kg),
- Consumed 3 kilowatts of power.
✴️ Now, DGX Spark can train models with up to 70 billion parameters locally, no data centers or cloud required!!
A historic comparison 💡
Nine years ago, in 2016, Huang delivered the first DGX-1 to Elon Musk at a small OpenAI office. That device helped ignite the revolution that eventually gave us ChatGPT. Today, the same compute power — 1 petaflop — fits in your backpack. 🎒
✴️ DGX Spark democratizes supercomputing, putting it in the hands of researchers, creators, and startups!
It truly is the MacBook of AI computing ~