Only with CPU offload

Llama 3.1 8B Instruct at Q4_K_M needs 6.1 GB but only 5.3 GB is addressable, so about 18% of the layers would stream from system RAM at roughly 60 GB/s. Expect around 25 tokens per second — usable for batch work, painful for chat.

82% on GPU 8K context Q4_K_M · 4.5 GB Llama 3.1 Community Released Jul 2024

Still the most widely deployed local model, with the largest fine-tune ecosystem. Not the strongest 8B any more.

Hardware requirements What hardware do I need to run Llama 3.1 8B Instruct? →

At Q4_K_M with an 8K context: a 10 GB GPU (Arc B570) or a 16 GB Mac (M4). Minimum, recommended, high-quality and long-context tiers, with all 114 devices checked at every quantisation.

Fits instead: Llama 3.2 3B Instruct (3.3 GB) · Llama 3.2 1B Instruct (1.5 GB)

The VRAM budget

weights 4.5 GB
Weights 4.5 GB KV cache @ 8K 1.00 GB Runtime overhead 0.6 GB Over budget 0.8 GB past 5.3 GB

Quantisation ladder

QuantWeightsTotal @ 8KMax contextTok/sQualityFit
F16 15.0 GB 16.6 GB — ~3.0 Reference 11.3 GB over
Q8_0 7.9 GB 9.5 GB — ~7.4 −0.1% ppl 4.2 GB over
Q6_K 6.1 GB 7.7 GB — ~12 −0.4% ppl 2.4 GB over
Q5_K_M 5.3 GB 6.9 GB — ~16 −0.8% ppl 1.6 GB over
Q4_K_M 4.5 GB 6.1 GB 1K ~25 −1.9% ppl 0.8 GB over
Q3_K_M 3.7 GB 5.3 GB 8K 56 −5.4% ppl Fits
Q2_K 3.1 GB 4.7 GB 12K 65 −15% ppl Fits

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

terminal
$ llama-server \
    -hf meta-llama/Llama-3.1-8B-Instruct:Q4_K_M \
    -c 8192 -ngl 26

The engine underneath most of the others. Every knob is exposed. More on llama.cpp.

01Download is 4.5 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.
03If it falls back to CPU silently, drop the context first, then step down a quantisation.
See all models for this rig Compare with another model

Related models on a GeForce GTX 1660 Super

ModelTotalTok/s est.
Llama 3.2 3B Instruct
3.21B · same family
Q4_K_M · 3.3 GB 113 Runs great
Llama 3.2 1B Instruct
1.24B · same family
Q4_K_M · 1.5 GB 292 Runs great
Llama 3.3 70B Instruct
70.6B · same family
Q4_K_M · 42.8 GB ~1.0 CPU offload
Llama 4 Scout 109B-A17B
109B MoE · same family
Q4_K_M · 63.4 GB ~1.5 CPU offload
Gemma 4 E4B
8.0B · similar size
Q4_K_M · 5.2 GB 45 Tight fit
Ling 3.0 Tiny 7.9B-A1.3B
7.9B MoE · similar size
Q4_K_M · 5.1 GB 107 Tight fit
Qwen3 8B
8.19B · similar size
Q4_K_M · 6.3 GB ~22 CPU offload

Every model against the GeForce GTX 1660 Super →

Llama 3.1 8B Instruct on other hardware

DeviceTotalTok/s est.
GeForce RTX 4060 Laptop
7.0 GB usable · 256 GB/s
6.1 GB 34 Tight fit
Arc B570
8.8 GB usable · 380 GB/s
6.1 GB 51 Runs great
GeForce RTX 3060 12 GB
10.6 GB usable · 360 GB/s
6.1 GB 48 Runs great
M4 · 16 GB
10.7 GB usable · 120 GB/s
6.1 GB 15 Runs great
GeForce RTX 4090
22.4 GB usable · 1008 GB/s
6.1 GB 135 Runs great
Radeon RX 7900 XTX
22.4 GB usable · 960 GB/s
6.1 GB 129 Runs great
M3 Max · 64 GB
48.0 GB usable · 400 GB/s
6.1 GB 50 Runs great

All 114 devices, by tier →