F

fal Compute

Infrastructure service

by Fal

Provides dedicated GPU instances for training and fine-tuning workloads.

Find alternatives to fal Compute

What it does

• GPU cloud
• Model training

Features 3

• Dedicated H100 instancesDedicated GPU instances in 1xH100 and 8xH100 SXM sizes, billed per hour at fixed rates.
• Multi-node InfiniBand8xH100 instances placed in the same sector are linked over InfiniBand for distributed training.
• Full SSH accessUsers SSH into the running instance and control the machine directly.

Sources

Pricing

Pricing page · 14 Oct 2026

"are billed per-hour at fixed rates"

Description

Official documentation · 14 Oct 2026

"dedicated GPU instances that run continuously"

What it does

Official documentation · 14 Oct 2026

"Dedicated GPU instances for training, fine-tuning"

Constraints

Trust centre · 18 Oct 2026

Derived from the security facts on this page, from the same source.

Security

Trust centre · 18 Oct 2026

Company-wide fal trust center (script-rendered; read in the in-app browser 2026-10-18): lists 'SOC 2 Type II', same source as fal-serverless and fal-agent.

Sold within

Official documentation · 14 Oct 2026

"The two products serve different workload profiles:"

Name

Official documentation · 14 Oct 2026

"fal Compute gives you"

URL

Official documentation · 14 Oct 2026

"fal Compute gives you"

Deployment

Official documentation · 14 Oct 2026

"A fal.ai account with Compute access"

Also from Fal

• fal Model API

  Hosted API that runs image, video and audio generation models, billed by compute time.

• fal Serverless Infrastructure service

  Runs a customer's own model or app on fal's GPU fleet, billed by the hour rather than per generation.

• fal Agent AI agent

  Subscription agent that plans and runs image and video generation across fal's hosted models, from first frame to final delivery.

• fal Sandbox Application

  Browser workspace that runs one prompt through several hosted image and video models side by side and shows each result with its model, latency and cost, so users can compare models before building on them.

Open Fal in the directory

Something wrong here? Send a correction — quote this product id: fal-compute.