Files
claudetools/.claude/machines/acg-guru-5070.md
Mike Swanson cef97725b8 Add Linux workstation machine spec (acg-guru-5070)
Lenovo Legion Pro 7 with Core Ultra 9 275HX, RTX 5070 Ti,
CachyOS. Documents known GPU firmware bug and custom kernel.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
2026-03-21 17:49:58 -07:00

2.5 KiB

Machine: acg-guru-5070

Hostname: acg-guru-5070 Last Updated: 2026-03-21


Hardware Specs

Spec Value
Model Lenovo Legion Pro 7 16IAX10H (DMI: 83F5)
CPU Intel Core Ultra 9 275HX (24 cores, up to 5.4 GHz)
Memory 32 GB DDR5
GPU NVIDIA GeForce RTX 5070 Ti Laptop GPU (12 GB VRAM)
Storage 1 954 GB NVMe (SK Hynix) - CachyOS root, btrfs
Storage 2 954 GB NVMe (SK Hynix) - /home, ext4

Software

Spec Value
OS CachyOS Linux (Arch-based)
Kernel 6.19.9-1-cachyos
DE KDE Plasma 6.6.3 (Wayland)
NVIDIA Driver 595.45.04 (open kernel module)
CUDA 13.2
Python 3.14

Claude Code Environment

  • Working Directory: /home/guru/ClaudeTools
  • User: guru
  • Shell: fish
  • Git: Configured for Gitea (git.azcomputerguru.com)

Network

Interface Address
WiFi (wlan0) 10.3.36.218
Tailscale 100.95.216.79

Capabilities

  • Git operations
  • SSH access to infrastructure
  • GrepAI semantic search (watcher running)
  • Ollama local AI (qwen3:14b, codestral:22b, nomic-embed-text)
  • MCP servers available
  • NVIDIA GPU (CUDA compute)
  • Claude Code CLI

Known Issues

GPU Firmware Bug (RTX 5070 Ti)

The RTX 5070 Ti enters an error state (NVRM rpcSendMessage 0x00000062) after ~3-5 minutes of sustained GPU compute. This is a known Blackwell/RTX 50-series GSP firmware bug on Linux (NVIDIA bug #5953411). Affects all tested drivers (580.x, 590.x, 595.x).

Impact: GPU-accelerated ML workloads (Whisper transcription, etc.) cannot complete. GPU enters full ERR! state requiring hard power-off (warm reboot hangs with spinning symbol).

Workarounds tried (none effective):

  • Disable Runtime D3 power management
  • Enable persistence mode
  • Lock GPU clocks
  • Power cap reduction

Status: Waiting for NVIDIA driver fix. Heavy GPU compute delegated to Mac (M4).

Custom Kernel for Audio

Running a custom-patched CachyOS kernel with the nadimkobeissi/16iax10h-linux-sound-saga patch for Awinic AW88399 smart amplifier support. Stock kernel has terrible speaker output. Patch is not upstreamed.


Notes

  • Primary development workstation
  • GPU works fine for display, light compute, Ollama inference — only fails under sustained heavy compute (Whisper, training)
  • Sudo: NOPASSWD configured for guru user
  • Old btrfs @home subvolume on nvme0n1 (from initial install before /home was moved to nvme1n1)