Presentation

How GPUs Must Adapt in the Age of Machine Learning and AI
DescriptionMachine learning can now synthesize nearly any image you can imagine. We can't do that in real time (yet!), but today, we can use neural technology in real-time rendering to reconstruct, denoise, and upscale the majority of pixels. Neural technology is already changing the way the computer graphics industry renders images, enabling a huge leap forward in rendering efficiency. The challenge ahead is enabling these technologies on the strict power budgets associated with mobile devices.

In this talk we will explore:
How neural graphics techniques are advancing so rapidly that standards are still catching up.
Viable ways to support heterogeneous workloads, a mix of hardware accelerators, and fragmented APIs.

Ways to harness and deploy experimental techniques in your titles and go beyond using networks trained by others to create networks to suit your needs.
The emergence of hardware, APIs, and real-time neural networks to solve performance and fidelity bottlenecks.

Each year brings new innovations that shape how we approach graphics. We’ll share how you can prepare for future GPU advances that will transform graphics performance on next generation mobile devices.
Presenter