Made a Rust DB run spatial queries on gaming GPU RT cores, beating an H100
7 hours ago
- #GPU Acceleration
- #Spatial Database
- #Ray Tracing
- SedonaDB 0.4 introduces GPU-accelerated spatial joins using gaming GPU ray tracing cores, outperforming H100 in some cases.
- It features RayBooster, a ray tracing engine with GPU-friendly storage, a monolithic index via Z-stacking, a universal predicate engine, and memory-aware execution.
- Performance tests show speedups up to 5.93x, cost reductions of 59.02% on AWS, and consumer GPUs like RTX 3090 beating H100 on certain queries.
- Usage involves pulling a Docker image and enabling GPU with a command, with support detailed in a GPU Acceleration guide.
- The release includes 187 issues resolved, 26 new functions, and upcoming features like Python/R APIs, geography support, and GeoParquet writes.