https://peer.madiator.cloud/videos/watch/02902288-98d9-487e-acdf-9c9c34a93a1b Runpod network volumes are pinned to a single data center which means your storage also picks your GPUs, and in a supply crunch that can mean no GPUs at all. Global volumes remove that pin: create one with no region and no size, and attach it to any GPU running on Secure Cloud. They're backed by object storage and built for data you write once and read constantly like base models, LoRAs, tokenizers, and config files. Network volumes are still the right answer for anything you're writing to constantly: training runs, checkpointing, databases, or anything with heavy concurrent writes. This video covers the difference, how to create a global volume (about ten seconds, in Storage, then New volume), how to attach one at deploy time, and the mount path behavior you'll want to understand before you go looking for files that aren't where you expect. That last part has three cases: a global volume on its own mounts at /workspace; attach a network volume alongside it and Runpod automatically moves the global volume to /workspace-global; but edit the global volume's mount path yourself and the network volume lands on /workspace2 instead. Mount paths are set at deploy time and don't change while the Pod is running. 0:00 The problem with regional volumes 0:59 Global volumes vs network volumes 1:34 Creating a global volume 2:03 Attaching one to a Pod 2:22 Mount paths: the three cases 3:16 Which to use, and the beta Thu, 24 Sep 2026 07:02:00 GMT https://validator.w3.org/feed/docs/rss2.html PeerTube - https://peer.madiator.cloud https://peer.madiator.cloud/lazy-static/avatars/e2927aea-9e0a-454c-9327-68ec87b4da11.png https://peer.madiator.cloud/videos/watch/02902288-98d9-487e-acdf-9c9c34a93a1b All rights reserved, unless otherwise specified in the terms specified at https://peer.madiator.cloud/about and potential licenses granted by each content's rightholder.