gavinj.net

From 24GB Blackwell workstation cards to one 128GB DGX Spark, then two Sparks and a QSFP cable: running DeepSeek-V4-Flash-Vision-Exp locally, understanding vLLM, and planning the next step with a 200Gb/s switch.

I started using ShellGPT today with a locally hosted Gemma 4 26B quantized model through Ollama on my RTX PRO 4000 Blackwell workstation. Asking questions in the terminal was fine, but the real value showed up when local AI met the Unix pipeline.

I wanted to know whether a sub-$3k Australian workstation GPU could run useful modern local LLMs. The short answer is yes, if you pick the right model, the right quant, and build around the 24GB VRAM limit instead of pretending it is a tiny cloud.

Google's Gemma 4 models are impressive, but the real story is the licensing. For the first time, a major AI lab has released capable open models under a true Apache 2.0 license. Here's why that changes everything.