Yes — with 32.6 GB to spare
GLM-4.7-Flash 30B-A3B at Q4_K_M fits your CPU only · DDR5 dual-channel entirely on the GPU at 8K context, at an estimated 9.9 tokens per second. There is room for its full 198K window.
Fully on GPU
8K context
Q4_K_M · 17.5 GB
MIT
Released 20 Jan 2026
MIT-licensed 30B-A3B tuned for agentic coding, with a DeepSeek-style latent KV cache. 60–80 tok/s reported on a 4090.
The VRAM budget
weights 17.5 GB
Weights 17.5 GB
KV cache @ 8K 0.41 GB
Runtime overhead 0.6 GB
Free 32.6 GB of 51.2 GB
Quantisation ladder
| Quant | Weights | Total @ 8K | Max context | Tok/s | Quality | Fit |
|---|---|---|---|---|---|---|
| Q8_0 | 30.9 GB | 31.9 GB | 198K | 5.6 | −0.1% ppl | Long context |
| Q6_K | 23.8 GB | 24.8 GB | 198K | 7.3 | −0.4% ppl | Long context |
| Q5_K_M | 20.6 GB | 21.6 GB | 198K | 8.5 | −0.8% ppl | Long context |
| Q4_K_M | 17.5 GB | 18.6 GB | 198K | 9.9 | −1.9% ppl | Recommended |
| Q3_K_M | 14.2 GB | 15.2 GB | 198K | 12 | −5.4% ppl | Long context |
| Q2_K | 12.2 GB | 13.2 GB | 198K | 14 | −15% ppl | Long context |
Quality is the published perplexity delta against f16 weights. Max context assumes an f16 KV cache; q8_0 roughly doubles it. This model uses multi-head latent attention, so its cache is a compressed latent rather than full K and V.
How to run it
$ llama-server \
-hf zai-org/GLM-4.7-Flash:Q4_K_M \
-c 8192 -ngl 99
The engine underneath most of the others. Every knob is exposed. More on llama.cpp.
01Download is 17.5 GB. Keep it on an SSD — a first load off a spinning disk takes minutes.
02With no GPU, thread count matters more than clock. Start at one thread per physical core.
03There is room to go to the model's full 198K context on this card.