Berth v0.4.1: Go-based GPU resource manager enables live-swapping of CUDA workloads

·2026.11.02 12:36

Key point

The alpha release supports NVIDIA Linux with experimental AMD and Windows support, featuring an OpenAI-compatible API and multi-GPU management.

Berth is a lightweight GPU resource manager written in Go that controls application access via leases, priorities, and a request queue to prevent VRAM conflicts.
Supports live-swapping of CUDA workloads by parking engines in system RAM or using NVIDIA CUDA checkpointing to avoid repeated disk loads.
Provides an OpenAI-compatible API, ComfyUI workflow support, and a web dashboard for monitoring VRAM, temperature, and active leases.
Currently in alpha (v0.4.1) with Linux NVIDIA support, experimental AMD Linux and Windows support, and no cloud account or additional monitoring service required.

This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.

Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.