Llama 3.1 on NVIDIA GeForce RTX 4070 · 12GB VRAM

✅ Runs (tight) — Llama 3.1 8B @ Q8_0
Llama 3.1 8B at Q8_0 needs ~9.5 GB (weights 8.0 GB + KV 1.0 GB + overhead 512 MB @ 8K ctx) of your 10.8 GB usable VRAM — fits, but little headroom — close other apps or trim context. Expect ~20–35 tok/s (fast).
context @ Q8_0: runs to its full 128K
quantneedsspeed
✓ Q8_09.5 GB~20–35 tok/s (fast)
✓ Q6_K7.6 GB~30–50 tok/s (fast)
✓ Q5_K_M6.8 GB~30–55 tok/s (fast)
✓ Q4_K_M6.1 GB~35–60 tok/s (fast)
$runlocal install llama3.1:8b
⚠ NVIDIA support is best-effort in v0.1 — verify before relying on it.
runlocal verdict card
download card share on x share on reddit drop it in your model-drop post
Llama 3.1: Runs (tight)
for your own README — links back here

Same model, other machines: Apple M1 · Apple M1 Pro · Apple M2 · Apple M3 Pro · Apple M2 Pro · Apple M4 Pro

Also runs on a NVIDIA GeForce RTX 4070: Qwen3 · DeepSeek-R1 (distill) · Qwen2.5 · Gemma 4 · Mistral 7B

Best models for a NVIDIA GeForce RTX 4070 · 12GB VRAM →

Architecturally similar

These share Llama 3.1's KV-cache geometry — they size memory the same way, so here's how they fit on a NVIDIA GeForce RTX 4070 · 12GB VRAM too:

modelsharesverdict here
Qwen2.5-Coder 8kv/128hd ✅ Runs (tight)
Llama 3.2 8kv/128hd ✅ Runs comfortably
Llama 3.3 8kv/128hd ❌ Won't fit
Mistral Small 3 8kv/128hd ⚠️ Partial GPU offload
Mistral Nemo 8kv/128hd ✅ Runs (tight)
Codestral 8kv/128hd ⚠️ Partial GPU offload
Llama 3.2 Vision 8kv/128hd ✅ Runs comfortably
Mixtral 8x7B 8kv/128hd 🐢 CPU-only (slow)

…and 7 more — see the Llama 3.1 page.

Check any combo yourself: open the checker →