"Cerny explained that current technologies like FSR and PSSR are extremely demanding on GPUs and system memory. The first new concept revealed was Neural Arrays, which Hyung explained was designed to overcome inefficiencies encountered by subdividing neural network problems across many compute units. Neural Arrays allow compute units to "team up" and share data, and the end result here is that this can lead to better FSR and ray tracing effects, more efficiency, and better scalability as a workload grows."
"Cerny and Hyung also expect ray tracing to improve through the use of Neural Radiance Caching--which was unveiled at Computex earlier this year--and Radiance Cores. Together, ray tracing and path tracing can be handled together in real time, allowing developers to access a new level of realistic lighting while also allowing the CPU to focus on geometry and simulation tasks."
AMD and Sony are collaborating through Project Amethyst to advance next-generation gaming hardware and software. Three recent developments aim to improve upscaling, ray tracing, and memory efficiency. Neural Arrays enable compute units to cooperate and share data, reducing inefficiencies from subdividing neural-network tasks and improving FSR, denoising, and scalability. Neural Radiance Caching and Radiance Cores allow ray tracing and path tracing to run in real time while offloading geometry and simulation work from the CPU, yielding cleaner and more efficient lighting pipelines. Universal Compression targets future GPUs and system-on-chips to dramatically reduce memory bandwidth demands for gaming workloads.
Read at GameSpot
Unable to calculate read time
Collection
[
|
...
]