Yes, just — 0.2 GB spare
Granite 4.1 8B at Q4_K_M fits your Arc A750 entirely on the GPU at 8K context, at an estimated 63 tokens per second. Past 9K the KV cache pushes it over — quantise the cache to q8_0, or step down a quantisation, to go longer.
Fully on GPU
8K context
Q4_K_M · 4.9 GB
Apache 2.0
Released 29 Apr 2026
Matches the old Granite 4.0 32B MoE at a quarter of the size. Fast, Apache 2.0, no reasoning traces.
The VRAM budget
weights 4.9 GB
Weights 4.9 GB
KV cache @ 8K 1.25 GB
Runtime overhead 0.6 GB
Free 0.2 GB of 7.0 GB
Quantisation ladder
| Quant | Weights | Total @ 8K | Max context | Tok/s | Quality | Fit |
|---|---|---|---|---|---|---|
| F16 | 16.4 GB | 18.2 GB | — | ~3.1 | Reference | 11.2 GB over |
| Q8_0 | 8.7 GB | 10.5 GB | — | ~8.7 | −0.1% ppl | 3.5 GB over |
| Q6_K | 6.7 GB | 8.6 GB | — | ~17 | −0.4% ppl | 1.6 GB over |
| Q5_K_M | 5.8 GB | 7.7 GB | 3K | ~29 | −0.8% ppl | 0.7 GB over |
| Q4_K_M | 4.9 GB | 6.8 GB | 9K | 63 | −1.9% ppl | Recommended |
| Q3_K_M | 4.0 GB | 5.9 GB | 15K | 77 | −5.4% ppl | Fits |
| Q2_K | 3.4 GB | 5.3 GB | 19K | 90 | −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.
How to run it
$ llama-server \
-hf ibm-granite/granite-4.1-8b:Q4_K_M \
-c 8192 -ngl 99
The engine underneath most of the others. Every knob is exposed. More on llama.cpp.
01Download is 4.9 GB. Keep it on an SSD — a first load off a spinning disk takes minutes.
02Close anything else holding VRAM. A browser with hardware acceleration can sit on 1–2 GB.
03Only 0.2 GB is spare, so a long system prompt can still push it over. Its real ceiling here is 9K context.