nvitop v1.3.2

Added

• Add separate implementation for GpuStatsLogger callback for lightning by @XuehaiPan in #114.
• Remove metrics if process is gone in nvitop-exporter by @XuehaiPan in #107.

Full Changelog: https://github.com/XuehaiPan/compare/v1.3.1...v1.3.2

nvitop v1.3.1

Added

• Add Python 3.12 classifiers by @XuehaiPan in #101.

Fixed

• Fix libcuda.cuDeviceGetUuid() when the UUID contains 0x00 by @XuehaiPan in #100.

Full Changelog: https://github.com/XuehaiPan/compare/v1.3.0...v1.3.1

nvitop v1.3.0

Added

• Add Prometheus exporter by @XuehaiPan in #92.
• Add device APIs to query PCIe and NVLink throughput by @XuehaiPan in #87.

Changed

• Use recent timestamp for GPU process utilization query for more accurate per-process GPU usage by @XuehaiPan in #85. We extend our heartfelt gratitude to @2581543189 for their invaluable assistance. Their timely comments and comprehensive feedback have greatly contributed to the improvement of this project.

Fixed

• Fix upstream changes for process info v3 APIs on 535.104.05 driver by @XuehaiPan in #94.
• Fix removal for process info v3 APIs on the upstream 535.98 driver by @XuehaiPan in #89.

Full Changelog: https://github.com/XuehaiPan/compare/v1.2.0...v1.3.0

nvitop v1.2.0

Added

• Include last snapshot metrics in the log results for ResourceMetricCollector by @XuehaiPan in #80.
• Add mypy integration and update type annotations by @XuehaiPan in #73.

Fixed

• Fix process info support for NVIDIA R535 driver (CUDA 12.2+) by @XuehaiPan in #79.
• Fix inappropriate exception catching in function libcuda.cuDeviceGetUuid by @XuehaiPan.

Full Changelog: https://github.com/XuehaiPan/compare/v1.1.2...v1.2.0

nvitop v1.1.2

Added

• Support float number as snapshot interval that >= 0.25s by @XuehaiPan in #67.
• Show more host metrics (e.g., used virtual memory, uptime) in CLI by @XuehaiPan in #59.

Changed

• Move TTLCache usage to CLI-only by @XuehaiPan in #66.

Fixed

• Further isolate the CUDA_VISIBLE_DEVICE parser in a subprocess by @XuehaiPan in #70.
• Respect FORCE_COLOR and NO_COLOR environment variables by @XuehaiPan.

Removed

• Drop Python 3.6 support by @XuehaiPan in #56.

Full Changelog: https://github.com/XuehaiPan/compare/v1.0.0...v1.1.2

The first stable release of nvitop. The APIs (from nvitop import *) and the CLI (nvitop / nvisel) now become stable.

• Repackage project with dual-license (Apache-2.0 and GPL-3.0), see License for more details. Bumping the version.
• Drop Python 3.5 support.

• Add function and method to collect metrics in a background thread in #48 (Resolves #47).

• Add warning messages for corrupted dependencies (Fixes #44).
• Handle "NVML Unknown Error" when failing to get the device handles (Fixes #45).

The last beta version of nvitop. We are waiting for several months of compatibility check the NVIDIA driver and nvidia-ml-py package. The v1.0 stable release will be coming soon if everything goes fine. Feedback is welcome.