can it run
Comfortably. 40GB of memory against a 5GB working set leaves headroom for context, the OS, and anything else you're running.
| model file | 3.3GB (GGUF Q4_K_M) |
| minimum memory | 5GB working set |
| this machine | 40GB HBM2e, 1.55 TB/s |
| params | 4B |
| license | gemma |
| price | $5900 |
easy run: ollama pull gemma3:4b
runs with: ollama · llama.cpp · koboldcpp · lm studio
no alternatives — every machine in the catalog can run this one.
NVIDIA (used market) ↗ · Amazon ↗