Voyager SDK v1.6.1 is Here with Windows Perf and M.2 Max Power Control

Voyager SDK v1.6.1 is a patch on top of v1.6.0. No huge new feature set, but useful cleanup, real Windows performance gains, closed-loop power control on the upcoming Metis M.2 Max, and a proper docs portal at docs.axelera.ai.

Windows is Faster

That’s not something you hear about Windows very often, is it? But this is referring to Voyager on Windows, so that might explain it. 😋

The PCIe driver and libaxldev have both had performance work, closing the gap with Linux, which the Windows fans out there will appreciate. Logging and tracing infrastructure is also improved, with a utility for viewing traces.

By the way, before we go too much further, this release requires Metis Driver 1.3.2 or newer on Windows. v1.3.1 and earlier are no longer supported, so v1.6.1 ships with driver 1.3.11.

Metis M.2 Max Power Control

Closed-loop power control on M.2 Max has been improved ahead of its launch:

• No more unexpected performance degradation.
• The limit set via axdevice --set-power-limit now matches what axmonitor reports at runtime.
• Default is now 8.5 W (down from 11 W), biased towards average workstation hosts. On hosts meeting PCIe SIG M.2 Spec Rev 4.0+ you can experiment up to 11 W in 0.1 W step increments. On power-constrained embedded SBCs, you can actually drop as low as 4 W.

More Metis Hardware Supported

Metis PCIe Rev.2 boards with 2 GB and 8 GB DDR variants are now supported.

Pipeline Builder

• axstreamer VideoDecode now supports color_format=auto, letting the decoder pick its native output format and allowing colour-convert to fuse into resize operators.
• OpenCL Resize and Colour Conversion plugins now support I420-formatted images downstream.

Notable Fixes

• Multi-stream video output to disk no longer drops throughput.
• Face-recognition pipeline no longer fails with "Failed to set kernel argument 7".
• Segfault at end of tiled inference on 4-chip PCIe with multiple streams is gone.
• axmonitor reports correct Avg/Min/Max power on 4-chip PCIe cards.
• Multiple semantic segmentation fixes (multi-class thresholding, single-class transpose, default sigmoid, missing prediction field in binary seg, missing preprocessing resize).
• render_to_wx.py is no longer painfully slow when adding a stream. stream_select.py and multiple_pipeline.py have been removed in its favour.
• OpenGL renderer now handles NV16, YUY2 and grayscale sources correctly.
• Firmware upgrade on Raspberry Pi 5 with Metis M.2 now works.
• axcompile --transform is restored.

It’s All Yours

Full notes on GitHub. Questions on Windows, power tuning or anything else, drop a post in here.