Processor Architectures/Configuration TRIZ Cases
- Micromap Tree Traversal for Faster Ray Tracing on Primitives
Micromap tree traversal speeds ray-primitive queries by storing sub-region properties compactly, cutting bandwidth and shader calls.
- Hybrid Input Feedback Architecture for Low-Latency User Response
A dual-path input architecture uses local low-latency processing plus synchronized high-latency state handling to make feedback feel immediate.
- GPU Tensor Core Tensor-Train Blocking for High-Dimensional Decomposition
Blockwise tensor-train primitives map matrix operations onto GPU tensor cores to cut decomposition time and memory for larger tensors.
- Radix Sort Offset Calculation for Early Pass Termination
Upfront global offset checks detect uniform radix digits so GPUs can skip local offset work, cutting sort time, power use, and data movement.
- Shared GPU Rasterizer for Real-Time 3D Gaussian Splatting
A shared GPU rasterizer switches between triangle and Gaussian logic to speed 3D Gaussian splatting with low area and power overhead.
- Ray Tracing Processor Node Culling Using Distance Records
Distance records let ray tracing hardware cull nodes beyond a ray's maximum range, cutting unnecessary tests and storage use.
- Sub-Image Streaming via DPU-GPU Pipeline for Low-Latency Processing
A DPU routes only needed image sections from FPGA streams to the GPU, cutting CPU latency and overhead while preserving packet integrity.