March 19, 2026
Elastic ephemeral storage, hardware health details on the host dashboard, and pricing/stability fixes.
Improvements
Section titled “Improvements”- Ephemeral storage no longer restricts which machines a pod can be scheduled to — capacity now expands elastically as needed.
- The host dashboard now shows more hardware detail: multi-GPU interconnect status and disk temperature/health.
- Multi-GPU pods now get the right interconnect configuration applied automatically, improving distributed workload reliability.
- Switched to a more accurate GPU utilization metric for usage reporting.
- Fixed a pricing bug where extra vCPU/RAM charges could be miscalculated for configurations below the GPU-recommended baseline, including a matching display error on the checkout receipt.
- Fixed a bug that could stop a pod again while it was still waiting for reallocation, and stopped repeated emails being sent on every reallocation attempt.
- Fixed several deployment-form issues: selecting vCPU below the recommended minimum, missing default vCPU/RAM values, an inconsistent status-text display, and incorrect framework-version filtering.