In the last 12 months I have entirely rebuilt the WSV3 Tactical application as an extremely low-level, bare-metal application written in raw Pascal using direct Win32 OS calls, complicating and slowing development due to perfectionism and obsession over performance, which is excellent.
This is necessary to truly differentiate next-gen and make it truly tactical-grade, ultraperformant, and eternal-execution-stable.
It launches within hundreds of milliseconds as a tiny self-contained executable itself only a few hundred kilobytes in size. There is no installer - it launches instantly and downloads-on-demand any resources it needs.
This is unheard-of for modern Windows software. It uses zero dynamic memory - static only - similar to NASA space rover coding standards. It will never crash. Unlike WSV3 Professional 2015, WSV3 Tactical Mesoanalyst is meant to run 24/7/365.
Further multi-month perfectionism delays were from making a new GPU rendering engine based on world-class modern D3D12 for maximum GPU performance vs the preexisting D3D11 renderer, and doing an ultra performant server ingest system for the realtime data. The latter, still under development, is using bare metal Pascal code on modern ARM64 Linux instances using Amazon EC2's most optimized compute instance class in 2025 (c8g instances).
In other words, June 2024 to June 2025 has seen abundant non-visual low-level engineering to ensure WSV3 Tactical will offer the fastest and most optimized display of realtime geospatial data available on planet earth.
I do plan to release a demo/early version sometime before December, 2D single map only, to test new infrastructure. Visible CONUS satellite (full res, not 1/4th resolution like first gen) and basic functionality of the new RadarHalo Polyradial Mosaic would be a great goal for that.
There cannot be a free demo/beta test - we (TempoQuest) have to charge for the data as now WSV3 Tactical is taking responsibility to do absolutely everything on private servers with zero client downloading from government servers.
On the bandwidth side, the value we're adding comes from how inefficient the raw government formats are, requiring to pay for extensive cloud processing as a business expense. This is a good thing. I remember telling my boss Gene in 2023 that our AWS bills will be going up markedly and that's actually a good thing, literally a consequence of the goal of this project, because the increase in added value to client will be even larger proportionally.
For instance, GOES CONUS Band 2 visible satellite imagery frame is around 70mb per frame, 5 minutely. Imagine looping that over just 3 hours. That is 70123mb - over 2GB of data you'd have to download for a 3hr loop if using direct raw government data.
The 2021 raster engine in first-gen does do private server processing for this, but downscales it to 1/4th resolution. WSV3 Professional 2015 does not have a multiresolution tiled raster engine. They are single frame images with no view-based tiles.
WSV3 Tactical will offer a narrower subset of the most useful products and deliver them at far higher spatiotemporal resolution, made possible by its new highly efficient, fast-downloading multiresolution tiled engine.
Another example is NEXRAD L2. Gibson Ridge and WSV3 first-gen download the original raw volumes. This is extreme waste, up to 20mb per volume even if you're only viewing base tilt REFL, which would be 0.5mb-1mb max.
So we're undertaking the large responsibility for the first time of realtime 24/7/365 processing the entire country NEXRAD L2 feed on custom servers and streaming bandwidth optimized proprietary render formats to client immediately over WebSockets with ultra low processing and transmission latency.
I'll finally announce my strong desire and cost optimization goal to keep the primary subscription level for Tactical Mesoanayst the same $25/month price as first-gen and not increasing it, only scaling back the diverse amounts of distinct datasets as necessary to achieve this, offering a streamlined, simplified curation of the fewer most useful products given the far higher server side processing requirements. Then offer higher tiers for industrial, e.g. a 2x cost workstation license with heavier extended datasets mainly for industrial users, but keeping the core competencies in the primary level. Foreseeably this would include the entire country NEXRAD Level 2 feed even though we're now undertaking processing that privately which is a large cost. If I attain this goal it should be a 100x value product over first gen for the same price and impress people and make everyone happy.
For the demo, as soon as I have the new raster engine working, I can release something. Right now there's just the world's most slick and efficient global elevation dataset viewer (static mapping data) at 120fps and using hierarchical delta tiles for ultra fast loading, of which I posted screenshots in other thread. But there needs to be some realtime data before releasing the demo as that demonstrates the core compentency.
The new WSV3 Tactical raster engine renders multiple dataset-native projections in a single composition pass using deferred rendering and compute shaders. This is world class GPU rendering commonly found in AAA game titles and will eventually enable sophisticated blending techniques in Mesoanalyst with all the customization.
Everything is tiled for ultra fast previewing and progressive download. There are 5 dimensions: X,Y tiles at multiple downscaled LOD/zoom levels (3 dimensions), then the time dimension for animations (4), and then multiple products (5).
Arguably there is a sixth dimension to this system because the current deferred renderer is per-projection with planned support for multiple parameters per projection, so as to avoid wateful redundant projection gbuffer shading. The multi-param single-proj support benefits e.g. GOES multispectral RGB products or model precip typing, where multiple rasters share the same projection.
I'm not aware of any similar weather graphics software in existence. Probably the industrial TV systems run on such fat hardware they just forward rasterize shaded meshes per-layer, not worrying about efficiency. That's reasonable for that case, but WSV3 Tactical must optimize power consumption and rendering efficiency.
To clarify, the new WSV3 Tactical raster engine is rendering the actual mathematically computed projection coordinates in realtime to a gbuffer using compute shaders and streaming multiresolution progressive loading tiles in original non-square projections, and shading multiple raster layers with a single-pass compute shader.
By comparison, WSV3 Professional uses the obvious traditional rendering method of pre-baked triangle meshes for the projections and forward-rendering each layer/projection one-by-one in separate draw calls.
So the new unified deferred compositor system using compute shaders allows for a meaningful performance optimization to avoid wastefully shading where there are full-opacity occluding/overlapping layers. In WSV3 Professional, the layer-by-layer forward rendering causes, e.g., the background map pixels to be fully shaded, then, e.g., the satellite layer could be almost completely overlaying/covering those pixels, resulting in wasted rendering. Then the radar layer could be on top of that, resulting in even more wasted rendering of pixels that never contributed to the final image.
This is the classic raytracing vs rasterization distinction in GPU graphics. WSV3 Tactical's raster engine is essentially doing 2D raytracing to enable a single composition pass for all raster layers.
The ambition of the new method is to avoid those wasteful passes in forward-rendered shaded triangle meshes, so that, once a pixel reaches full opacity in a top-down compute-based shading loop, it early-outs and does not waste GPU cycles to fetch textures/blend pixels that would be invisible, because they are completely covered by a higher layer.
If you look at 10 random screenshots of WSV3 Professional you see this is a common case. Oftentimes there are large areas of occlusion of one layer over another. Just turning on the 2M temperature fill and zooming into CONUS with the map background on causes the entire map background rendering to be an expensive waste of GPU on every frame, contributing zero difference to the final image, because it's underneath the solid model fill.
WSV3 Tactical earns its name by offering the most performance/power optimized realtime GPU rendering of scientifically precise, original projection datasets for efficient usage in mobile contexts (such as storm chasing or any field use operational context), in addition to the bandwidth use optimization, largely a product of the multiresolution tile engine (only downloading tiles for current view) but also of further cutting-edge techniques to be discussed later.
