Troubleshooting

Diagnose and fix common issues. Start with Quick Diagnosis, then jump to the relevant section.

Your SystemUnknown OS • GPU not detected

Quick Diagnosis

App Won't Start

SymptomLikely CauseFix
Stuck at "Starting..."Health check URL mismatchVerify runtime.port and health_check.path in manifest
Crashes immediatelyMissing Python packageCheck Console for ImportError; reinstall app
Permission deniedAppImage not executablechmod +x thumper-run.AppImage
Port already in useOrphan process from previous sessionss -tlnp | grep PORT, then kill the PID
Python not foundMissing system PythonInstall python3.10+ from your package manager
CUDA not availableWrong PyTorch buildCheck patch applies correct CUDA/ROCm torch index

Generation Fails

SymptomLikely CauseFix
Black/NaN imagesFP16 overflow on AMD APUUse BF16 dtype + FP16-fix VAE
Out of memory (OOM)Text encoder + diffusion both on GPUSet text_encoder_device: cpu in pack manifest
Model not foundWrong install_dir or incomplete downloadRe-download from Models page
Workflow error (ComfyUI)Missing custom nodeInstall missing node via ComfyUI Manager
Infinite loadingModel too large for VRAMUse smaller quantization or split to CPU

Performance Issues

SymptomLikely CauseFix
Slow first generation (AMD)MIOpen find-db buildingOne-time ~5 min warmup, then set MIOPEN_FIND_MODE=FAST
Slow every generation--lowvram on APU (shuffles same RAM)Remove --lowvram flag from patches
High memory, GPU at 5%Tiled VAE on short contentUse non-tiled VAE for content under 60s
CPU inference very slowExpected behaviorUse GPU, or accept 10-50x slowdown vs GPU
Model reloads every timeSmart memory disabledEnable smart memory (default); remove --disable-smart-memory
NEVER use --lowvram on AMD APU. It does per-layer offload between the SAME physical RAM — 5% GPU utilization, 50× slower.

Downloads & Models

SymptomLikely CauseFix
Download stuck at 99%Corrupted .part fileDelete .part file, retry (auto-resumes from scratch)
GPU not detectedMissing or outdated driversNVIDIA: nvidia-smi; AMD: rocminfo; Intel: sycl-ls
Model shows 0 B sizeHF cache ghost repo (aborted download)Rescan models from Settings; ghost repos are filtered
Wrong model loadedFilename collision in install_dirCheck model path; rename or re-download

Benchmarks

Reference benchmarks for common tasks. Your results may vary based on system configuration, cooling, and background processes.

Image Generation (SDXL 1024×1024, 20 steps)

GPUTimeNotes
RTX 4090~3.2sFull CUDA
RTX 3060 12GB~8.5sFull CUDA
RX 7900 XT~5.1sROCm 6.2
AMD 780M APU~152sBF16, first run ~188s
Apple M2 Max~12sMPS
CPU (Ryzen 7)~300sVery slow

LLM Inference (Qwen3-4B Q4_K_M)

GPUSpeed
RTX 4090~120 tok/s
RTX 3060~45 tok/s
AMD 780M APU~25 tok/s
CPU (8-core)~8 tok/s

Music Generation (ACE-Step, 10s track)

GPUTime
RTX 4090~2s
AMD 780M APU43s (warm) / 55s (cold)

Cross-App Performance Reference

Real-world benchmarks across common workloads. Times are for a single generation with warm model cache.

AppWorkloadRTX 4090RTX 3060RX 7900 XTAMD 780MM2 ProCPU
ComfyUISDXL 1024×1024~3s~15s~6s~45s~12s~180s
ComfyUILTX Video 5s~15s~60s~25sN/A~45sN/A
Open WebUIQwen3 4B (tok/s)~120~45~90~15~50~5
ACE-Step30s music~9s~35s~15s~45s~25sN/A
JanLlama 3.2 8B (tok/s)~110~40~85~12~45~4

Agent Tools Reference

The AI assistant has access to 28 tools across 6 categories. Each tool follows a common trait interface.

CategoryCountExamples
Launcher6app_install, app_launch, app_stop, app_uninstall, app_health, process_list
GPU Diagnostic7gpu_detect, vram_report, rocm_check, cuda_check, pytorch_index, npu_list, benchmark
Model Intelligence5model_search, model_download, model_verify, model_link, model_cache_size
Manifest v23manifest_validate, manifest_generate, cache_diagnose
Diagnostic2system_info, port_scan
Filesystem5file_read, file_write, file_list, file_search, file_diff

Tool Trait Interface

rust
#[async_trait]
pub trait AgentTool: Send + Sync {
fn name(&self) -> &str;
fn description(&self) -> &str;
fn parameters_schema(&self) -> serde_json::Value;
fn is_safe(&self) -> bool;
async fn execute(&self, params: Value) -> Result<Value>;
}

Invoking Tools

Tools are invoked through the agent chat interface. Type a natural language request and the agent selects appropriate tools. Safe tools run immediately; dangerous tools show a confirmation dialog with the planned action.