Ever hit VK_ERROR_DEVICE_LOST and had no idea what actually caused it? This sample shows how to use VK_EXT_device_fault (plus VK_EXT_device_address_binding_report) to pull real diagnostic data out of a GPU crash: vendor fault descriptions, the exact faulting memory address, and a live log of every buffer’s bind/unbind range so you can correlate the crash back to the specific resource that triggered it. A big step up from guessing based on validation layer output alone.
This looks really useful for tracking down issues. Being able to see the faulting address and match it against buffer binding history should make GPU crashes much easier to diagnose than relying only on validation output.