# XiongjieDai/GPU-Benchmarks-on-LLM-Inference

Multiple NVIDIA GPUs or Apple Silicon for Large Language Model Inference?

Repository: https://github.com/XiongjieDai/GPU-Benchmarks-on-LLM-Inference
Canonical: https://ross.abutalabs.com/products/gpu-benchmarks-on-llm-inference
Language: Jupyter Notebook
License Family: other
Last push: 2024-05-13T00:29:05+00:00

## Health v2 (maintenance only)
Score: 28/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 0, release rhythm 35, longevity 81
- inputs: {"age_days": 1142, "days_push": 843, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases, no_license
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1935, forks 76 (observed 2026-08-28T04:05:56.991630+00:00)

## What it is
A curated benchmark dataset comparing LLM inference speeds (tokens/s) across many NVIDIA GPUs and Apple Silicon chips using llama.cpp on LLaMA-family models. Results are presented as tables and Jupyter notebooks covering quantized and full-precision model sizes from 8B to 70B parameters.

## Use cases
- which gpu should i buy for running llama locally
- compare inference speed of rtx 4090 vs a100 for llm
- can a macbook m1 max run llama 70b
- how fast is llama 3 8b on a 3090
- best gpu for local llm inference on a budget
- how many gpus do i need to run a 70b model
- apple silicon vs nvidia for llm inference

## When to choose
- choosing hardware for local LLM inference with llama.cpp
- comparing consumer and datacenter GPUs for token generation speed
- checking whether a given GPU has enough VRAM for a model size and quantization

## When to avoid
- benchmarking training performance rather than inference
- comparing inference frameworks other than llama.cpp
- needing up-to-date results for the newest GPU generations

## Facets
- artifact type: dataset
- maturity: maintenance
- function: benchmarking, llm-inference, gpu-computing
- domain: large-language-models, gpu-computing, hardware, performance
- platform: -
- tags: llama-cpp, nvidia-gpus, apple-silicon, tokens-per-second, hardware-comparison, runpod, gpu, macos, linux

## Member repositories
- XiongjieDai/GPU-Benchmarks-on-LLM-Inference (main) score 28

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:05:56.991630+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T03:08:27.217322+00:00, confidence not recorded.
  - readme: https://github.com/XiongjieDai/GPU-Benchmarks-on-LLM-Inference (fetched 2026-08-28T04:05:56.991630+00:00, sha 34fce38c3d75)
- Data as of 2026-08-30T08:39:29.467469+00:00.