WUMBOLABS / PROJECT RECORD READ-ONLY / INSPECTION

← Back to Projects

WUMBOLABS / PROJECT RECORD

WumboJetsII

Fedora 44 workstation for daily development, local AI evaluation, wumbOS work, and WumboLabs evidence.

WumboJetsII is the primary WumboLabs desktop workstation: a real daily-use Fedora 44 system, not a dedicated headless benchmark appliance.

It runs Fedora 44 with Hyprland and supports development, wumbOS Shell work, LLMGauge testing, and deeper local-model evaluation on consumer hardware.

Active daily workstation / evidence source

At a Glance

OS Fedora 44
Desktop Hyprland
GPU NVIDIA GeForce RTX 5070 12GB
CPU / Memory Ryzen 7 9800X3D / ~32GB DDR5
Storage Samsung 9100 Pro 2TB NVMe
AI Stack llama.cpp CUDA
Local AI RTX 5070 evaluation
Role Daily desktop workstation

Role in WumboLabs

WumboJetsII is where WumboLabs development, wumbOS Shell development/testing, and local AI testing happen.

It is a constrained consumer-workstation environment for LLMGauge runs and deeper model-evaluation campaigns. The RTX 5070 also drives the Fedora/Hyprland session during testing, so practical free VRAM is lower than its physical 12GB capacity. That constraint is intentional evidence context, not controlled-lab isolation.

The current Qwen3.8 campaign used the RTX 5070 with llama.cpp/CUDA; read the Lab Record for its runtime-specific findings and limitations.

Focus areas

Linux Desktop

Fedora 44, Hyprland, Wayland workflows, keyboard-driven interaction, theming, and practical daily use.

Local LLM Evaluation

LLMGauge workflows and deeper GGUF evaluation across llama.cpp builds, context limits, generation speed, VRAM behavior, and bounded reliability.

Development Workstation

Primary system for wumbOS Shell, LLMGauge, website work, scripting, documentation, and local experiments.

CUDA Work

Local llama.cpp/CUDA testing on SM120. Current Qwen campaign runtime: driver 610.57.04, CUDA 13.3, and llama.cpp b10449.

Operating goal

The goal is to keep the system clean, stable, fast, keyboard-friendly, and useful for real work.

Experiments are welcome, but the baseline should remain reliable.