LM Studio silently broke my model load after a runtime auto-update — the NTSTATUS exit code decode
If you run local LLMs on Windows with an NVIDIA GPU, you've probably seen LM Studio's "runtime extension packs." They're the bundled CUDA/ROCm/Vulkan builds of llama.cpp that actually run your models. And by default, LM Studio auto-updates them in the background. That bit me on an RTX 5090. One day a model that had loaded fine for days suddenly refused to load. The dialog just said: Failed to…
LM Studio users on Windows with NVIDIA GPUs may have encountered an issue where a model that previously loaded fine suddenly refused to load after a runtime auto-update. The error dialog displayed "Failed to load the model / Error loading model / (Exit code: 18446744072635812000)." The exit code is a Windows NTSTATUS and not a random number.
Upon converting the exit code to hex, it revealed that the low 32 bits, 0xC00008A0, indicated a specific CUDA error related to shared object initialization failure in the flash-attention kernel. Further investigation revealed that the issue stemmed from the CUDA 12 runtime v2.27.1, which contained a broken fallback path on RTX 5090 (sm_120) architecture.
The fault did not originate from the model file or configuration, but rather from the runtime build. To resolve the issue, users were advised to roll back to an older CUDA build (v2.25.2) and disable automatic runtime updates until a fixed version was released. This incident highlighted the potential for runtime regressions in bundling a runtime with local LLM tools, and the importance of monitoring for specific failure signatures such as cryptic exit codes and cryptic error messages in logs.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
