Building a High-End AI Desktop

kept by eddie dec 11, 25

An engineer bought a discounted Grace-Hopper server, converted it to water cooling, and now runs 235B parameter models locally for under €9,000.

Scaling Karpathy's Autoresearch: What Happens When the Agent Gets a GPU Cluster
Giving Karpathy's autoresearch agent access to 16 GPUs enabled 9x faster optimization by running experiments in parallel and enabling factorial grid searches.
also on H100, #developer-tools
Unsloth Studio | Hacker News
Unsloth launches Unsloth Studio, an open-source, no-code web UI for locally running, training, and exporting 500+ open models with 2x faster training and 70% less VRAM.
also on Llama.cpp, Nvidia, #open-source, #ai, #developer-tools
AI At Home Part 2: Multi GPU Drifting
The author boosted Deepseek V4 Flash inference from ~10 to ~20 tokens/sec on four e-waste AMD V620 GPUs by using layer parallelism and a speculative draft model.
also on Llama.cpp, #hardware, #open-source, #developer-tools