Today’s most concrete shift is the arrival of Comma.ai’s Tiny Chestnut eGPU dock, a compact PCIe Gen4 x4‑to‑USB4 adapter that bundles an AMD Radeon RX 9060 with 8 GB of GDDR6 memory. The dock’s open‑source firmware is marketed as a way to sidestep proprietary driver hassles, yet operators should verify that the AMD Linux and Windows drivers are fully supported before committing production workloads. At 8 GB VRAM the unit can comfortably host 6‑B‑parameter models like Llama 3‑6B or smaller, but anything larger will require aggressive quantization or off‑loading.
In parallel, NVIDIA demonstrated that its GB300 NVL72 can serve the 2.4‑trillion‑parameter Qwen 3.8‑2.4T model with configurable reasoning, underscoring that server‑grade GPUs remain the only path for truly massive models. No new server rigs entered the catalog, which stays at 51 verified units, so the Tiny Chestnut may be the only fresh compute offering for edge or lab‑scale teams this week.
If you’re budgeting for a flexible AI workstation, weigh the dock’s low entry price against the inevitable performance ceiling imposed by its memory budget. Expect to see real‑world benchmarks before scaling production workloads.
Composed by the MadCoolStuff editor pipeline · Groq · openai/gpt-oss-120b · 2026-08-13