Skip to main content

Can the RTX 5080 Run Llama 3.1 8B?

Runs

Yes. Llama 3.1 8B at Q4_K_M is a 4.9 GB download (Hugging Face file listing for bartowski/Meta-Llama-3.1-8B-Instruct-GGUF), and the RTX 5080 has 16 GB of VRAM, leaving about 11.1 GB for the KV cache and runtime. Published runs put generation at a median of 88.5 tokens per second across 2 cited measurements. Fastest cited run: Local AI Master.

Model file4.9 GBQ4_K_M
VRAM16 GBRTX 5080
Headroom11.1 GBafter weights
Median speed88.5 tok/s2 cited runs
Bandwidth ceiling~195 tok/s960 GB/s ÷ 4.9 GB

Published runs for Llama 3.1 8B on the RTX 5080

GenerationQuantRuntimeContextSourceDate
132 tok/s q4_K_M ollama Local AI Masterlabelled "llama3.1:8b" 2026-01-01
44.9 tok/s Q4_K_M llama.cpp LocalScorelabelled "Meta Llama 3.1 8B Instruct" 2025-06-01

More about this pair

How much VRAM does Llama 3.1 8B need?

The Q4_K_M file is 4.9 GB. Add roughly 1.5 GB for the runtime and an ~8K-token context, so about 6.4 GB of VRAM holds it comfortably. Longer contexts grow the KV cache and need more.

What is the fastest Llama 3.1 8B can generate on the RTX 5080?

Generation reads every weight once per token, so the RTX 5080's 960 GB/s of memory bandwidth divided by the 4.9 GB file sets a ceiling near 195 tokens per second. Real runs land below it; SpecPicks rejects any published figure above it.

Check another pair

See the full matrix → · All RTX 5080 benchmarks →

Llama 3.1 8B on other GPUs

Other models on the RTX 5080

How the checks work

The fit is the Q4_K_M file size against the card's VRAM, with 1.5 GB kept for the runtime and an ~8K context. A published run counts only if the pair fits, the run is Q4-class, its source does not call it an estimate, it is under the bandwidth ceiling above, and it is not one figure pasted across many cards. The whole matrix is downloadable as a CSV under CC BY 4.0.

As an Amazon Associate, SpecPicks earns from qualifying purchases.

This page is editorial synthesis based on publicly available information. No independent first-party benchmarking is reported.