kitsuneislife/nmx-intel: one-colour composites despite a finished GPU wait. Their NullMoth 2.5.4 port, where the
RTX composites the desktop and NVAccel copies frames to an Intel panel, saw the panel blink black: some composites
reached the copy still a single colour although the GPU wait had reported them ready. Their patch
(accel-0004-hold-blank-frames.patch, 2910e8a) keeps
up to three one-colour frames in a row off the panel when it flips RAM buffers, and counts them
(debug.nvaccel_iop_blank_held). The follow-up note
says the remaining blinks are not one-colour frames. This bears on our I-108 note that frames copied from the second
swap source in N-278 and N-280 hashed to 0. Theirs is NVIDIA compositing; ours is Intel through XeGfxAccel, so the
cause may differ. Medium relevance, Unverified for our stack. The repo has no licence, so we take the idea, not the code.
WhateverXe 0.5.0 and WhateverXe-Graphics 0.2.0 (0.5.0, v0.2.0): Alder Lake-S support, Tiger Lake GT1 resources, Display 12 register tracing, USB-C reconnection hangs documented as not OS-specific. Not Arrow Lake, so nothing changes for us. High confidence on what changed.
JulioNectar/Hackintosh-XPS13Plus-IrisXe-Tahoe: "GPUREC v2" recovers render-engine hangs on Apple's own Intel driver without a freeze (1c792c4), an 80 EU topology boot-arg, H.264 encode notes. A different stack from ours (Apple's driver on Tiger Lake), so ideas only.
kumibrr/oc-tahoe-pavilionplus14-ew0000 (install and Recovery Wi-Fi notes). Discovery: uchmk/yagura (a system monitor), CGR-MIX/AN515-macOS-Tahoe (Acer Nitro 5 case notes with NullMoth NVIDIA, nothing Intel GPU).