[[["易于理解","easyToUnderstand","thumb-up"],["解决了我的问题","solvedMyProblem","thumb-up"],["其他","otherUp","thumb-up"]],[["很难理解","hardToUnderstand","thumb-down"],["信息或示例代码不正确","incorrectInformationOrSampleCode","thumb-down"],["没有我需要的信息/示例","missingTheInformationSamplesINeed","thumb-down"],["翻译问题","translationIssue","thumb-down"],["其他","otherDown","thumb-down"]],["最后更新时间 (UTC)：2025-09-03。"],[],[],null,["# Generate NVIDIA bug report for Blackwell GPUs\n\n*** ** * ** ***\n\nThis document explains how to create an NVIDIA bug report for your machine types\nthat use [NVIDIA Blackwell GPUs](https://www.nvidia.com/en-us/data-center/technologies/blackwell-architecture/).\nIf you want to identify which of your machine types are using the NVIDIA Blackwell GPU, see\n[GPU models](/compute/docs/gpus#gpu-models).\n\nIf your machine type uses the NVIDIA Blackwell GPU architecture, the generated\nNVIDIA bug report doesn't include critical low-level hardware data when you run\nthe [nvidia-bug-report.sh](https://docs.nvidia.com/deploy/rma-process/index.html#topic_3_1)\nscript. This data contains information such as the physical layer status of NVLink\nconnections, internal GPU register values, and raw diagnostic segments from the\nfirmware. This information is essential for diagnosing issues, especially those\nrelated to NVLink, which could lead to\n[GPU Xid errors](https://docs.nvidia.com/deploy/xid-errors/index.html#xid-error-listing)\nor unresolved performance degradation.\n\nGenerate an NVIDIA bug report\n-----------------------------\n\nTo generate a bug report, complete the following steps:\n\n1. Connect to your GPU instance. Choose one of the following options:\n\n - [Connect to Linux instances](/compute/docs/connect/standard-ssh)\n - [Connect to Windows instances](/compute/docs/instances/connecting-to-windows)\n2. Download and install the MFT package by selecting one of the following options:\n\n### Container-Optimized OS\n\n\nIf your instance uses a Container-Optimized OS (COS) as the\nguest operating system, use the open source [GCE COS NVIDIA Bug Report Collector](https://github.com/GoogleCloudPlatform/cluster-toolkit/tree/main/community/gce-cos-nvidia-bug-report)\ntool to generate the bug report with MFT. This tool automatically\ninjects supported MST kernel modules that match the COS kernel,\ninstalls the userspace tool, generates the bug report, and\noptionally uploads the result to a Cloud Storage bucket.\n\n### Other OS\n\nFor other Linux OSes, complete the following steps:\n\n1. Download and install NVIDIA Firmware Tools (MFT) software version 4.32.0 or higher from the [NVIDIA website](https://network.nvidia.com/products/adapter-software/firmware-tools/).\n2. Install the tool. For more information, see [Compilation and installation](https://docs.nvidia.com/networking/display/mftv4320/compilation+and+installation) in the NVIDIA Firmware Tools (MFT) Documentation. After you install MFT, the [nvidia-bug-report.sh](https://docs.nvidia.com/deploy/rma-process/index.html#topic_3_1) script automatically uses the MFT tools to generate the report. You don't need to interact with MFT tools directly.\n3. Run the `nvidia-bug-report.sh` script to generate a bug report. This process takes about two minutes.\n4. Extract the report.\n5. Verify that the report includes MFT data by running the following command on your extracted bug report file: \n\n ```\n grep -m 1 -A 30 \"Starting GPU MST dump..\" PATH_TO_UNZIPPED_BUG_REPORT\n ```\n\n The output is similar to the following example: \n\n ```text\n Starting GPU MST dump..\n ... (additional MFT data) ...\n \n ```"]]
