Skip to content
can-i-run-this-llm

Community leaderboard

Fastest rigs, ranked by real benchmarks

Find the fastest measured hardware for each model and the most decode throughput per dollar. Rankings use community medians, with the model and quantization kept visible.

Add your benchmark →Full benchmark table →
prelim means one or two reports; three or more confirms the community median. Reports can use different runtimes, contexts and run setups. These are measured decode speeds, not capability scores or matching-run comparisons.

Most tokens per dollar

Measured decode tok/s per $1,000 of indicative device cost. Only rigs with a known price qualify. Catalog prices are not a complete ownership-cost estimate; model size and quantization differ between entries.

  1. 1.
    NVIDIA RTX 4090 (24GB)

    43.4 tok/s per $1k

    Llama 3.1 8B Q4_K_M

    130.2 tok/s$2,999 device1 reportprelim

Fastest rig per model

Each model has its own ranking, ordered by community median decode speed. Every measured quant remains a separate entry.

Llama 3.1 8B

  1. 1.
    NVIDIA RTX 4090 (24GB)
    130.2 tok/sQ4_K_M1 reportprelim
  2. 2.
    MacBook Pro / Mac Studio M2 Max (32GB)
    49.3 tok/sQ4_K_M2 reportsprelim

3 benchmarks · 2 devices · 1 model · full benchmark table →