can it run
No. 8GB of memory can't hold the 12GB working set this model needs. A smaller quant or a bigger machine is required.
weights 9.3GB + KV cache 0.62GB at 4k context vs ~8GB usable → offload-partial. 9.9 GB vs 8 GB usable — partial CPU offload, expect large speed loss
computed roofline: 192.3 GB/s × 0.28–0.50 efficiency window / 9.3GB (Q4_K_M) — bands, never points.
| model file | 9.3GB (GGUF Q4_K_M) |
| minimum memory | 12GB working set |
| this machine | 8GB GDDR5, 192 GB/s |
| params | 14B |
| license | apache-2.0 |
| price | $100 |
easy run: ollama pull qwen3:14b
runs with: ollama · llama.cpp · koboldcpp · lm studio
→ Tesla M40 (24GB, used) — cheapest machine that runs it ($55)
→ every model the Tesla P4 (8GB, used) can run — full list
→ tesla-p100-16gb — the next step up from this machine's GPU