GPU runners: track alternatives to the RunsOn custom AMI

Aperta
  • Qualcosa non va?
#183 0 commenti 0 reazioni 0 assegnatari Vedi su GitHub

I maintainer di solito rispondono entro 1 giorno

Nessuno ha ancora preso questa issue.

Valutazione

Difficoltà
4/5
Tempo stimato
3-5 giorni
Idoneità per principianti
45/100
Tipo di issue
Funzionalità
Chiarezza
Specificata chiaramente
Stato di attività
Attiva
Stack tecnologico
aws, docker, github-actions, python, shell
Ambito
build-system, ci-cd, cloud, devops

Direzione di ricerca

Review the removed docs/FUTURE-DEVELOPMENT.md lines (linked in the issue) and the current GPU-AMI-SETUP.md. The task involves researching GitHub-hosted GPU runner availability/pricing and evaluating a CUDA container prototype. Start by checking GitHub's documentation for GPU runner availability and pricing, then compare build times and costs. For the container path, examine the existing CPU container setup and the nvidia/cuda base image sizes.

Scritto dal modello di indicizzazione a partire dal testo della issue.

Descrizione

infrastructure

Why

GPU lecture builds (lecture-python.myst) run on RunsOn with a custom AMI, which needs AMI maintenance and Packer rebuilds (docs/dev/GPU-AMI-SETUP.md). Two alternatives were recorded in docs/FUTURE-DEVELOPMENT.md, which #178 removed (source). This issue keeps them tracked.

Alternatives

GitHub-hosted GPU runners

  • No AMI to maintain; the runner is selected with runs-on:.
  • Availability and pricing need checking before this is worth evaluating.
  • Trade-off: likely more expensive per minute than RunsOn, and possibly less choice of instance and GPU type.

A GPU container on an nvidia/cuda base

  • CUDA and cuDNN pre-installed next to the scientific Python stack.
  • Matches the container pattern the CPU builds already use.
  • Still needs a GPU runner (RunsOn or GitHub-hosted) started with --gpus all.
  • Trade-off: an image of roughly 15–20 GB, so pull time could cancel out the setup savings unless layer caching works well.

Related

  • #152 bakes the environment into the current AMI, and is the near-term path.
  • #157 and #158 cover RunsOn Magic Cache and cost right-sizing.

Checklist

  • Check whether GitHub-hosted GPU runners are generally available to the org, and at what price
  • If they are, compare a cold lecture-python.myst build against RunsOn on time and cost
  • Decide whether a CUDA container is worth prototyping, given the image size
Lingua principale
Shell
Stelle
0
Fork
1
Merge medio
20h 3m
PR unite (30g)
48

Preparare l'ambiente

Come iniziare

  1. Leggi tutta la issue e poi la guida ai contributi del progetto.
  2. Commenta sulla issue per dire che te ne occupi tu — evita che due persone facciano lo stesso lavoro.
  3. Fai un fork del repository e lavora su un branch.
  4. Apri una pull request che faccia riferimento al numero della issue.

Altre issue di QuantEcon/actions

Tutte le issue di QuantEcon/actions

Issue simili

Altre issue su Shell/Bash

Ricevi le nuove issue nella tua casella

Un breve riepilogo di issue GitHub adatte ai principianti.