Engineering notes.
Thoughts on software, hardware, and building things that scale.
The eGPU Myth: Why a ~$300 Dock Won't Turn Your GPU Into an AI Workstation
We benchmarked an RTX 3090 over USB4. The engineering is brilliant. The numbers aren't there yet.
Local LLM Inference: The 24GB VRAM Wall
Why the 24GB VRAM limit is the hardest bottleneck in consumer AI, and how quantization techniques are attempting to bypass it.
RTX 3090 vs 4090: A Deep Dive for AI Researchers
Is the 4090 worth the premium for local inference, or is the 3090 still the reigning champion of value?