JonathanChiu

信息截止到 11 月底：没有，但可以非常确定出问题的是谁。Cos 软件工程师做了许多测试来自证清白。我把我的 RTX 5070 放 Windows 上跑了半天，没有任何硬件、软件问题；又放 Ubuntu 上测试，也是完全可以正常工作的。 在发完贴之后没几天，跟 GPT 评估了我的需求后，我就换普通 Windows 11 当服务器了。 服务器的配置，误打误撞成了家里最强，现在就当全天开机的无头 PC 用了。其他电脑基本成了空壳，或者拥有文件的本地副本，随时有需要就用 RDP 等协议远程。买了个视频采集卡用来调试，无用时两头都插服务器上，这样就能串流原生运行的游戏。资源管理也更优秀，爆 12 GB 显存、爆 20 GB 内存、12 核 CPU 100% 再也没有了。 高强度折腾 Unraid 2 个月，发现了很多刁钻问题。macOS SMB 每层文件夹都要转圈或卡死（把 SMB 所有选项排列组合都尝试了一次，发现关闭“增强的 macOS 互操作性”解决，真有意思），魔改 Linux 系统的生态和社区资源、支持问题，包括这显卡……虽然在 7 * 24 小时的研究下解决了部分问题，但是各种配置排列组合的实验、为了配置生效无数遍重启、操作莫名复原反复返工，已经把我搞疯了，同时也是比较失望的。 因为存储设备杂乱，看中了 Unraid 阵列功能。但是 Windows 上 Stablebit 系列就能实现类似的效果，甚至功能更丰富。与其为了一个功能牺牲整个系统，不如在一个全能的系统上安装几个小软件，想要什么就加什么。对于我这种不需要虚拟机功能的用户来说，是这样的。

Thank you for your help, I’m so sorry. If it’s a problem with your NVIDIA driver for Unraid, please fix it. Thanks!

I just tested the RTX 5070 on Ubuntu 24.04 with open-kernel driver 580.95.05, and everything works fine. I have done everything you recommended, but the problem still persists on Unraid. The following is the diagnostic file. jonathan-server-diagnostics-20251128-1353.zip

Hi, I’m running into a weird and very inconsistent issue with my RTX 5070 FE on Unraid. On fresh boot, GPU Statistics can see the GPU normally, but as soon as anything tries to use the GPU (starting a Docker container, doing a CUDA task, or sometimes even when doing nothing), it quickly throws: Vendor command returned unparseable data. Sometimes the GPU works for a while and I can do transcoding or ML tasks, but it’s extremely unstable. Restarting or stopping GPU-dependent containers may temporarily fix it. I’m on Unraid 7.2.0 (same issue on 7.1.4), using the NVIDIA open-kernel driver (580.95.05). I also tried switching to the proprietary driver (580.105.08), but then the GPU is not detected at all — the plugin shows “No GPU found”, and nvidia-smi says “No devices were found”. HDMI/DP video output works normally. Unraid “System Devices” correctly lists the card as GeForce RTX 5070. The GPU is an RTX 5070 Founders Edition 12GB. i5-11600KF with ASUS TUF B560M (PCIe 4.0). No iGPU on the CPU. PSU is Seasonic Focus ATX3 GX-850. To rule out hardware issues, I tested on Windows Server 2025 with the latest Game Ready 581.80 driver, running OCCT VRAM and 3D Adaptive tests for one hour each. Both tests were completely stable — temps and clocks rose to normal levels and stayed flat with no drops, errors, or instability. This makes me think the hardware itself is probably fine. Here are examples of the repeating Unraid syslog errors when “Vendor command returned unparseable data.” appears (open-kernel driver (580.95.05); only pasting a few; the logs spam continuously): A: Nov 26 19:02:16 Jonathan-Server kernel: NVRM: nvCheckOkFailedNoLog: Check failed: Out of memory [NV_ERR_NO_MEMORY] (0x00000051) returned from memdescAllocInternal(pMemDesc) @ memdesc.c:1359 Nov 26 19:02:16 Jonathan-Server kernel: NVRM: nvAssertOkFailedNoLog: Assertion failed: Out of memory [NV_ERR_NO_MEMORY] (0x00000051) returned from nvStatus @ message_queue_cpu.c:242 Nov 26 19:02:16 Jonathan-Server kernel: NVRM: _kgspInitRpcInfrastructure: GspMsgQueueInit failed Nov 26 19:02:16 Jonathan-Server kernel: NVRM: kgspConstructEngine_IMPL: init RPC infrastructure failed Nov 26 19:02:16 Jonathan-Server kernel: NVRM: osInitNvMapping: *** Cannot attach gpu Nov 26 19:02:16 Jonathan-Server kernel: NVRM: RmInitAdapter: osInitNvMapping failed, bailing out of RmInitAdapter Nov 26 19:02:16 Jonathan-Server kernel: NVRM: GPU 0000:01:00.0: RmInitAdapter failed! (0x22:0x51:763) Nov 26 19:02:16 Jonathan-Server kernel: NVRM: GPU 0000:01:00.0: rm_init_adapter failed, device minor number 0B: Nov 17 02:36:53 Jonathan-Server kernel: NVRM: nvCheckOkFailedNoLog: Check failed: Out of memory [NV_ERR_NO_MEMORY] (0x00000051) returned from _memdescAllocInternal(pMemDesc) @ mem_desc.c:1353 Nov 17 02:36:53 Jonathan-Server kernel: NVRM: nvAssertOkFailedNoLog: Assertion failed: Out of memory [NV_ERR_NO_MEMORY] (0x00000051) returned from status @ kernel_gsp.c:4615 Nov 17 02:36:53 Jonathan-Server kernel: NVRM: nvCheckOkFailedNoLog: Check failed: Out of memory [NV_ERR_NO_MEMORY] (0x00000051) returned from kgspCreateRadix3(pGpu, pKernelGsp, &pKernelGsp->pGspUCodeRadix3Descriptor, NULL, pGspFw->pImageData, pGspFw->imageSize) @ kernel_gsp.c:4502 Nov 17 02:36:53 Jonathan-Server kernel: NVRM: kgspInitRm_IMPL: Error preparing GSP-RM image Nov 17 02:36:53 Jonathan-Server kernel: NVRM: RmInitAdapter: Cannot initialize GSP firmware RM Nov 17 02:36:53 Jonathan-Server kernel: NVRM: GPU 0000:01:00.0: RmInitAdapter failed! (0x62:0x51:1941) Nov 17 02:36:53 Jonathan-Server kernel: NVRM: GPU 0000:01:00.0: rm_init_adapter failed, device minor number 0 Here is the syslog when using the proprietary driver (580.105.08) — the Blackwell 5070 needs open kernel modules: Nov 27 10:38:24 Jonathan-Server kernel: NVRM: The NVIDIA GPU 0000:01:00.0 (PCI ID: 10de:2f04) Nov 27 10:38:24 Jonathan-Server kernel: NVRM: installed in this system requires use of the NVIDIA open kernel modules. Nov 27 10:38:24 Jonathan-Server kernel: NVRM: GPU 0000:01:00.0: RmInitAdapter failed! (0x22:0x56:884) Nov 27 10:38:24 Jonathan-Server kernel: NVRM: GPU 0000:01:00.0: rm_init_adapter failed, device minor number 0Given the GPU has 12GB VRAM and nothing is even using it at the time, the “Out of memory [NV_ERR_NO_MEMORY]” part seems unlikely to be real VRAM exhaustion. Is this likely a problem with the Linux open-kernel driver for the RTX 50-series? Or something with Unraid’s kernel? Or could this still be some rare hardware or compatibility issue? Appreciate any insights — thanks!

平时通电待机 GPU Statistics 可以正常显示 GPU 状态，但是一旦调用 GPU，或者一启动可能需要 GPU 的 Docker 容器，甚至或者是任意时候，就容易报错 Vendor command returned unparseable data.。 偶尔能正常工作，进行编解码和机器学习，但极其不稳定。重启或关闭需要 GPU 的容器可能能恢复。 Unraid 7.2.0（包括 7.1.4）使用 Nvidia 开源驱动。当前 7.2.0 对应的驱动版本为 580.95.05。尝试过闭源驱动，但是直接识别不了 GPU。Nvidia Driver 显示无 GPU，nvidia-smi 输出 No devices were found。 GPU 是 RTX 5070 Founders Edition 12GB，没有核显，视频输出正常。电源为海韵 Focus ATX3 GX-850。“工具”>“系统设备”能正常显示 GeForce RTX 5070。Windows 下测试和游戏正常。暂未测试其他 Linux 系统下 GPU 的状态。 请问大佬们这是 Linux 版驱动的问题吗？是需要时间来解决的吗？还是说有硬件问题，或是 Unraid 的内核或者其他软件问题？ 以下为 GPU Statistics 报错 Vendor command returned unparseable data. 时，Unraid 的系统日志。问题期间持续输出，短时间内会有多份。 Nov 26 19:02:16 Jonathan-Server kernel: NVRM: nvCheckOkFailedNoLog: Check failed: Out of memory [NV_ERR_NO_MEMORY] (0x00000051) returned from memdescAllocInternal(pMemDesc) @ memdesc.c:1359 Nov 26 19:02:16 Jonathan-Server kernel: NVRM: nvAssertOkFailedNoLog: Assertion failed: Out of memory [NV_ERR_NO_MEMORY] (0x00000051) returned from nvStatus @ message_queue_cpu.c:242 Nov 26 19:02:16 Jonathan-Server kernel: NVRM: _kgspInitRpcInfrastructure: GspMsgQueueInit failed Nov 26 19:02:16 Jonathan-Server kernel: NVRM: kgspConstructEngine_IMPL: init RPC infrastructure failed Nov 26 19:02:16 Jonathan-Server kernel: NVRM: osInitNvMapping: *** Cannot attach gpu Nov 26 19:02:16 Jonathan-Server kernel: NVRM: RmInitAdapter: osInitNvMapping failed, bailing out of RmInitAdapter Nov 26 19:02:16 Jonathan-Server kernel: NVRM: GPU 0000:01:00.0: RmInitAdapter failed! (0x22:0x51:763) Nov 26 19:02:16 Jonathan-Server kernel: NVRM: GPU 0000:01:00.0: rm_init_adapter failed, device minor number 0还有另一种版本： Nov 17 02:36:53 Jonathan-Server kernel: NVRM: nvCheckOkFailedNoLog: Check failed: Out of memory [NV_ERR_NO_MEMORY] (0x00000051) returned from _memdescAllocInternal(pMemDesc) @ mem_desc.c:1353 Nov 17 02:36:53 Jonathan-Server kernel: NVRM: nvAssertOkFailedNoLog: Assertion failed: Out of memory [NV_ERR_NO_MEMORY] (0x00000051) returned from status @ kernel_gsp.c:4615 Nov 17 02:36:53 Jonathan-Server kernel: NVRM: nvCheckOkFailedNoLog: Check failed: Out of memory [NV_ERR_NO_MEMORY] (0x00000051) returned from kgspCreateRadix3(pGpu, pKernelGsp, &pKernelGsp->pGspUCodeRadix3Descriptor, NULL, pGspFw->pImageData, pGspFw->imageSize) @ kernel_gsp.c:4502 Nov 17 02:36:53 Jonathan-Server kernel: NVRM: kgspInitRm_IMPL: Error preparing GSP-RM image Nov 17 02:36:53 Jonathan-Server kernel: NVRM: RmInitAdapter: Cannot initialize GSP firmware RM Nov 17 02:36:53 Jonathan-Server kernel: NVRM: GPU 0000:01:00.0: RmInitAdapter failed! (0x62:0x51:1941) Nov 17 02:36:53 Jonathan-Server kernel: NVRM: GPU 0000:01:00.0: rm_init_adapter failed, device minor number 012GB 显存，且根本就没调用上 GPU，应该不会显存不足。