Skip to content

Gamepad results by protocol

These are virtual source devices. HID ReadFile and XInput are different observation APIs; their numbers must not be pooled into a global ranking. Every filtered row additionally requires PnP attachment, the complete generated capture count and independently observed modified content.

Profile Receiver Backend Filter Operation Runs States Median run P50 (µs) Median run P99 (µs)
ds4 HID ReadFile vdevice Gamepad-Controller modify 2 2000 82.28 743.40
ds4 HID ReadFile vdevice baseline forward 3 3000 17.50 104.23
ds4 HID ReadFile vigem baseline forward 3 3000 5.70 44.60
xbox-one-gip XInput vdevice baseline forward 3 3000 101.30 64272.09
xbox360 XInput vdevice Combined modify 3 3000 51.10 267.25
xbox360 XInput vdevice Gamepad-Controller modify 3 3000 38.10 170.39
xbox360 XInput vdevice baseline forward 7 7000 51.70 123.96
xbox360 XInput vigem baseline forward 6 6000 53.25 616.05

DS4 reports record the observed thumb axes, twelve-button/D-pad field, triggers and report size in CSV. GIP uses a native virtual USB profile observed through XInput, with trigger conversion explicitly handled by the producer. Neither is a physical Bluetooth/USB measurement.

Initial filter association may require a guest restart. The fixture uses a stable USB serial so Windows can retain per-instance filter configuration. This is a descriptor-level fixture adaptation, not a driver or license modification. Failed pre-restart attempts remain in the dataset.

CPU figures in per-run JSON are producer/receiver process-lifetime milliseconds, including initialization and warm-up. Peak working set includes process pages and SDKs. Neither is isolated kernel CPU or driver pool usage.

Per-run controller results · Raw data · Method