How 1catai/1catqa works
1Cat-V100-QA is a V100-focused GPU qualification suite. Users start a run through the CLI, terminal UI, or HTTP API; the orchestrator discovers GPUs, captures baseline health, runs NVLink/P2P and gpu-burn checks, compares post-run counters and temperature outcomes, then produces per-run reports. The README also documents remote result queries and webhook notifications. The graph reflects the sampled orchestration and interface code; some detailed test internals and API routes are described by the README rather than inspected source excerpts.
Main entry points: core.py Test orchestration · commands.py Command runner · httpd.py HTTP API · reporting.py Report generation
Components 20
User interfaces
- CLI commands
src/gpuqa/cli.py - Desktop entry
src/gpuqa/gui_entry.py - Terminal UI
src/gpuqa/tui.py - HTTP API
src/gpuqa/httpd.py
Qualification workflow
- Test orchestration
src/gpuqa/core.py - GPU discovery
src/gpuqa/core.py - Baseline health
src/gpuqa/core.py - NVLink P2P test
src/gpuqa/core.py - GPU burn test
src/gpuqa/core.py - Result assessment
src/gpuqa/core.py - Run data models
src/gpuqa/models.py
Results and reporting
- Report generation
src/gpuqa/reporting.py - Run report files
System integrations
- Command runner
src/gpuqa/commands.py - NVIDIA driver tools
- gpu-burn
- P2P benchmark
cuda/p2p_bandwidth_matrix.cu
Other components
- Operator
- Remote agent
- Webhook endpoint
Connections 25
- Operator → CLI commands: runs commands
- Operator → Desktop entry: opens UI
- Desktop entry → Terminal UI: starts
- Remote agent → HTTP API: requests tests
- CLI commands → Test orchestration: starts run
- Terminal UI → Test orchestration: starts run
- HTTP API → Test orchestration: starts run
- Test orchestration → GPU discovery: discovers GPUs
- Test orchestration → Baseline health: captures baseline
- Test orchestration → NVLink P2P test: runs link test
- Test orchestration → GPU burn test: runs stress test
- Test orchestration → Result assessment: evaluates outcomes
- Test orchestration → Run data models: builds run data
- GPU discovery → Command runner: queries devices
- Baseline health → Command runner: samples counters
- NVLink P2P test → Command runner: runs benchmark
- GPU burn test → Command runner: runs burn process
- Command runner → NVIDIA driver tools: invokes queries
- Command runner → gpu-burn: invokes stress test
- Command runner → P2P benchmark: invokes benchmark
- CLI commands → Report generation: writes reports
- Terminal UI → Report generation: writes reports
- HTTP API → Report generation: writes reports
- Report generation → Run report files: writes files
- HTTP API → Webhook endpoint: posts completion