CONTEXT
Why this comes up
Puget Systems updated its dual Radeon AI PRO R9700 study after community feedback and testing on a bare-metal host.
THE USEFUL LESSON
The authors attributed their earlier multi-GPU failure to initialization in a virtualized passthrough environment. They subsequently reported a working bare-metal configuration. The environment belongs in the result, not in a footnote.
ON YOUR MACHINE
What to record and test
- 01
State whether the run is bare metal, containerized or inside a virtual machine.
- 02
Publish the accelerator topology, passthrough configuration and runtime version.
- 03
Keep failed runs available with their original conditions.
- 04
When a reproduction changes the finding, add a dated correction and link both configurations.
This is an attributed summary of Puget’s work, not a UT benchmark. See the original study for its hardware, methods and revised measurements.