How to fix the "GPU drivers not detected" error in HiveOS?

Troubleshooting GPU Driver Detection in HiveOS

1. HiveOS relies on properly loaded kernel modules to recognize and manage GPUs. If the system fails to detect GPU drivers, it usually indicates a mismatch between the installed driver version and the running kernel or an incomplete initialization sequence during boot.

2. The error commonly appears after a kernel update without corresponding driver recompilation. HiveOS uses DKMS (Dynamic Kernel Module Support) to rebuild proprietary drivers like AMDGPU-PRO or NVIDIA’s nvidia-uvm against new kernels. A failure in this process leaves the GPU modules unloaded.

3. Users may observe missing entries under /proc/driver/nvidia for NVIDIA cards or absent /sys/class/drm devices for AMD GPUs. These paths serve as direct indicators of whether the kernel has successfully bound GPU hardware to its driver stack.

4. HiveOS dashboard displays “No GPUs” or “0 devices” alongside the “GPU drivers not detected” message. This is not always a hardware fault—most cases trace back to software-level misconfiguration rather than physical GPU failure.

5. Manual inspection via SSH reveals empty output from commands like nvidia-smi or rocm-smi. Running lsmod | grep -i nvidia or lsmod | grep -i amdgpu returns no results, confirming the absence of active driver modules.

Verifying Kernel and Driver Compatibility

1. HiveOS versions ship with specific kernel-driver pairings. For example, HiveOS 0.6-225 targets Linux kernel 5.15.x and expects NVIDIA driver version 525.85.07 or AMDGPU firmware compatible with ROCm 5.4.1.

2. Executing uname -r shows the current kernel version. Cross-referencing this with HiveOS release notes confirms whether the installed driver package supports that kernel revision.

3. The command dkms status lists all registered DKMS modules and their build status. A state showing “built” but “not installed” means the module compiled successfully but failed to load into the kernel.

4. Checking /var/lib/dkms/nvidia/525.85.07/build/make.log (path varies by driver version) exposes compilation errors such as missing headers, incompatible GCC versions, or unresolved symbol references.

5. HiveOS auto-updates may silently install newer kernels without triggering driver rebuilds if DKMS hooks are disabled or corrupted. This leads to persistent driver detection failures until manual intervention occurs.

Reinstalling GPU Drivers Manually

1. Access the HiveOS web interface, navigate to Settings → Advanced → Console, and launch the terminal. Alternatively, connect via SSH using the default credentials.

2. Run sudo hive-replace-driver --nvidia or sudo hive-replace-driver --amd depending on the GPU vendor. This triggers a full reinstall using HiveOS’s curated driver binaries.

3. After execution, monitor the output for messages like “DKMS module added”, “initramfs updated”, and “reboot required”. Skipping reboot prevents the new modules from loading at next startup.

4. Some users report success only after clearing stale DKMS entries first: sudo dkms remove nvidia/525.85.07 --all, followed by sudo dkms autoinstall.

5. If the issue persists, forcing a clean reinstall with sudo hive-update --force pulls fresh OS components including firmware blobs and kernel modules, often resolving deep-seated dependency conflicts.

Hardware and Firmware Considerations

1. PCIe link training failures can prevent GPU enumeration entirely. Check dmesg | grep -i 'pcie|error' for messages indicating ASPM (Active State Power Management) incompatibility or insufficient slot power delivery.

2. Older AMD Polaris or NVIDIA Pascal cards require legacy firmware packages. HiveOS may omit these from default installations unless explicitly enabled via hive-config flags during setup.

3. UEFI Secure Boot interferes with unsigned DKMS modules. Disabling Secure Boot in BIOS resolves detection issues on certain ASUS and ASRock mining motherboards.

4. BIOS settings such as CSM (Compatibility Support Module), Above 4G Decoding, andResizable BAR must be correctly configured. Misalignment here causes GPU memory mapping failures visible in dmesg logs.

5. Physical inspection reveals bent PCIe pins, oxidized contacts, or insufficient 12V rail capacity on PSUs. Even if the GPU powers on, marginal voltage stability prevents proper driver handshake sequences.

Frequently Asked Questions

Q: Can I use custom NVIDIA drivers instead of HiveOS-provided ones?Yes, but only if compiled against the exact kernel version. HiveOS disables third-party driver installation by default to maintain stability across updates.

Q: Why does nvidia-smi show “NVIDIA-SMI has failed because it couldn’t communicate with the NVIDIA driver”?This occurs when the nvidia-drm and nvidia-uvm modules fail to load, typically due to version skew or conflicting nouveau drivers still resident in memory.

Q: Does HiveOS support NVIDIA datacenter GPUs like A100 or H100?HiveOS officially supports A100 with ROCm-compatible drivers starting from version 0.6-231. H100 requires specific firmware patches not yet integrated into stable releases.

Q: What does “Failed to initialize NVML” mean in the miner logs?NVML (NVIDIA Management Library) depends on successful nvidia.ko module loading. This error confirms the driver stack did not reach initialization stage, pointing to kernel module or permission issues.