PC freezes ~0.5s while firing, buffered input replays after

One more form of the same thing: sometimes the freeze does not recover. Yesterday in Team Deathmatch the picture held for about a second and then the game closed straight to the desktop, with no error dialog and no error code. Once I saw a window flash for a single frame just before it vanished. So the crash looks like the same stall, only long enough to kill the process.

Still happening after that repair, so I measured it instead of guessing.

GPU utilisation goes 81% -> 26% -> 11% and board power 222W -> 122W inside a single 100ms tick, while the process still holds 1.1-1.5 cores. My sampler interval stays at 92-122ms across that window, so the freeze is real and not a logging artefact.

The process runs 97 threads. Six carry the frame. At the freeze five of those six drop to zero for the same tick: no GPU work (power falls as above), no input reads (my character keeps doing whatever I told it last), no simulation. One thread keeps a full core. That thread also holds a full core while I sit in the main menu with GPU at 3-19%, so it is not a render thread; it looks like a resident worker that does not depend on the frame at all.

Five threads from unrelated subsystems stopping in the same tick is not something the game does to itself. They must be waiting on something they share. I could not identify which module they belong to: the module list on r5apex_dx12 reads back empty and thread start addresses cannot be queried, which I assume is the anti-cheat handle protection.

Same range, single player, firing at dummies: within a few minutes, every session. Every other game on this PC is fine. No TDR events, no WER dump, no watchdog, and the current driver was installed after the issue started.

Is this a known interaction where the anti-cheat worker holds a lock that the render and input threads need?