◆ What
Personal GPU research infrastructure
128GBvram
3.4xngcSpeedup
◆ How
NVIDIA DGXCUDA 13.0NGC ContainersPyTorchOllama
◆ Why
The Lab + The Stack
Hero: Flagship projects with major impact
Overview
NVIDIA DGX Spark with GB10 Blackwell and 128GB VRAM. Personal ML research lab enabling experiments that would otherwise require expensive cloud compute. NGC 25.09 containers provide 3.4x speedup vs baseline PyTorch.
The Journey
2024-09-15•The Lab Gets Real
DGX Spark arrives. 128GB VRAM. A personal GPU research lab rivaling small teams.
↳ Hardware meets ambition. Now anything is trainable.
2024-12-01•The Distributed Lab
Mac + DGX + Pi working together. Vector indexing on DGX. Trading bots on Pi. Development on Mac.
↳ One person, three machines, infinite leverage
Key Metrics
128GB
vram
3.4x
ngcSpeedup
$0
perExperimentCost
Technologies
NVIDIA DGXCUDA 13.0NGC ContainersPyTorchOllama
Evidence & Links
Related Writing
Connected Projects
→Enables