Cursdar2 is a CUDA-native NEXRAD workstation designed to provide fast live Level 2 radar data ingest, responsive 2D/3D rendering, archive review, and warning-overlay analysis. Built as a standalone successor to cursdar, it retains the same GPU-first approach while offering a cleaner operator workflow and independent evolution.
Key Features:
Real-time Level 2 data ingestion from AWS NEXRAD and select IEM experimental feeds.
Responsive single-site 2D rendering with tilt stepping functionality.
National mosaic rendering for comprehensive radar network visualization.
Live 3D volume rendering and draggable vertical cross-sections.
Archive playback with frame scrubbing and support for multi-site case studies.
Live and historic warning overlays, including polygons, watches, and alerts.
Customizable basemap options with relief, satellite, and hybrid imagery modes.
Experimental storm interrogation tools, including TDS, hail, and mesocyclone markers.
GPU-accelerated performance profiles for optimized VRAM and RAM usage.
Audience & Benefit:
Ideal for meteorologists, researchers, and weather enthusiasts who require high-performance radar analysis. Cursdar2 enables efficient data visualization, real-time monitoring, and detailed storm investigation, enhancing situational awareness during severe weather events. It is particularly suited for professionals seeking a GPU-optimized tool to work with live or archived NEXRAD Level 2 data.
Cursdar2 can be installed via winget, ensuring seamless integration into your workflow.
README
cursdar2
cursdar2 is a CUDA-native NEXRAD workstation focused on fast live Level 2 ingest, responsive 2D/3D radar rendering, archive review, and warning-overlay analysis.
It is being built as a standalone successor to cursdar: same GPU-first mentality, much cleaner operator workflow, and a repo that can evolve independently.
Highlights
Live Level 2 ingest from AWS NEXRAD plus selected IEM-hosted experimental/testbed feeds
Fast single-site 2D rendering with working tilt stepping
National mosaic rendering across the radar network
Real-time 3D volume mode
Draggable vertical cross-sections
Historic case playback with frame scrubbing
Archive snapshot loading, including the March 30, 2025 multi-site case
Live and historic warning polygons, watches, and related alert overlays
Custom basemap system with relief, ops-dark, satellite, and hybrid imagery modes
Support for Norman ROC/KCRI experimental feeds including KCRI, DAN1, DOP1, FOP1, NOP3, NOP4, ROP3, and ROP4
Highly configurable warning styling:
per-category toggles
outline/fill controls
opacity and line scaling
custom colors
Experimental storm interrogation overlays:
TDS markers
hail markers
mesocyclone / TVS markers
Storm-relative velocity mode
GR / RadarScope-style color table import
Early GR-style polling link intake
Hardware-aware performance profiles with live VRAM / RAM telemetry
Docked workstation UI with station browser, inspector, warnings panel, and historic timeline
Live Data Model
cursdar2 now uses tiered live polling instead of a blunt full-network refresh loop.
Active station: fast polling for newest available scans
In-view stations: medium cadence polling
Background stations: slower maintenance polling
Warning overlays: separate live polling loop
Once a station has already loaded a scan, the app uses incremental S3 listing against the last known volume key instead of re-listing the entire day each time. That keeps the hot path much lighter while still picking up newly published volumes quickly.
Practical note: display latency is still bounded by upstream publication. The app can only render a scan once NOAA/AWS has published a complete object.
Experimental/testbed radar site support alongside the core WSR-88D network
3D volume rendering
Cross-sections
Tilt browsing in single-radar mode
Archive playback
Archive snapshot loading
Live warning overlays
Historic warning overlays matched to archive timestamps
Warning customization controls
Custom basemap renderer with vector overlays and cached raster imagery
Color table import and per-product reset
Polling-link fetch and inspection
CUDA-backed rendering pipeline
Performance profiles with lighter 3D quality and reduced archive caching
Not finished yet:
Full GR2 feature parity
Full placefile rendering
Measurement / interrogation tools
Broader polling-link product support
Long-duration operational validation across many live weather events
Build Requirements
Windows
NVIDIA GPU with CUDA support
CUDA Toolkit
Visual Studio 2022 Build Tools
Ninja
Linux
NVIDIA GPU with CUDA support
CUDA Toolkit
CMake
A recent C++17 compiler
Build
Windows
build.bat
Binary output:
build/cursdar2.exe
Linux
chmod +x build.sh
./build.sh
Controls
1-7: select radar product
Left / Right: cycle products
Up / Down: cycle tilts
A: toggle national mosaic
V: toggle 3D volume
X: toggle cross-section
S: toggle storm-relative velocity
R: refresh live data
Home: reset to CONUS
Escape: return to auto-track
Space: play / pause historic playback
UI Notes
The inspector shows the latest scan time for the active site.
The operator console exposes Auto, Quality, Balanced, and Performance profiles, plus current/peak VRAM and process RAM.
The warnings panel can show live or historic polygons depending on mode.
The basemap panel exposes relief, ops-dark, satellite, and hybrid map modes plus overlay controls.
Ops Dark is the default basemap for instant startup; imagery styles stream in on demand and fall back to cached parent tiles while sharper zoom levels load.
Color tables can be loaded from the operator console with a file browser.
Polling links are currently ingested and inspected, but not yet fully rendered as full GR-style placefile content.
Status
This is already a real GPU radar workstation, but it is still an active build-out rather than a finished operational replacement for mature commercial software. The fast path is there now; the remaining work is on feature depth, workflow polish, and repeated validation with real events.