NVIDIA AI: Graphics Enhancements in Games and Talking AI Jarvis

On May 14, Nvidia held an online conference GTC 2020. In an opening presentation, CEO Jen-Hsun Huang talked about the latest advances in artificial intelligence and where this technology will be applied in the future.

At the presentation, the first computational accelerators were presented based on the Ampere architecture, on the capacities of which the AI ​​of the future will work.

AI in Computer Graphics

The presentation part devoted to games and graphics mainly focused on the already known things — real-time ray tracing and DLSS 2.0 upscaling technology. Meanwhile, if you miss old-school games and are always ready to play Mario or The Pokémon, check the Roms Planet emulators with numerous games available.

Especially for the presentation, Nvidia prepared a demo with a glass ball. Without exception, all objects and coatings are rendered, taking into account the physical properties of their textures, and light is reflected and propagated according to real optical laws — Minecraft RTX works similarly, but in the demo with realistic graphics, the effect of full path-tracing is much stronger.

The Marble demo was launched on a professional Quadro RTX 8000 (such a card costs more than $6500) — on modern RTX 20-series gaming video cards, the frame rate will be lower.

At the same time, DLSS 2.0 is used to improve performance. A specially trained neural network, which is processed by individual tensor cores in RTX video cards, builds an image from a lower resolution to a higher one based on the “experience” obtained during training on images at maximum resolution.

The effect of smart upscaling was demonstrated on the example of a frame from the Infiltrator demo on Unreal Engine 4. The image in 1080p is less clear and detailed than the one restored from 720p. This happens because the neural network is guided by how the image looks at 16K resolution.

Currently, smart upscaling is used in only a few games: Control, Wolfenstein: Youngblood, Deliver Us the Moon, but over time DLSS 2.0 and similar technologies (FidelityFX from AMD) will become more widespread. This will affect the way developers approach system requirements and make high-tech AAA games more accessible.

Talking AI Assistants with Realistic Facial Expressions

Nvidia believes that an interface capable of maintaining a dialogue with a person is the highest point in artificial intelligence development. To create realistic AI assistants, the company has developed the Jarvis platform, on which you can create various custom-builds of “talking” AI.

Nvidia’s AI platform’s main feature is the integration of speech and visualization — the system not only perceives speech, understands the essence of the question, and formulates a grammatically correct answer but also selects a suitable visualization with a mimic model and, if necessary, special effects.

The platform allows you to perform all the necessary operations for a full response in a few hundred milliseconds — this helps a person perceive communication with a bot as a real dialogue.

Modern AI can perceive not only the human voice but also “understand” animals. To demonstrate the speed of processing requests, the company showed a demo in which a neural network analyzes birds’ singing and determines which species of bird the sound belongs to and in which part of the world it can be heard.

Among the promising options for using talking AI, the company names assistants for cars, answering machines for call centers, and moderation systems for video conferencing that can help human participants and summarize business conference calls.

The era of artificial intelligence is gradually coming, turning from distant plans and fantasies of science fiction writers into a familiar reality. Nvidia is far from the only tech company researching and developing artificial intelligence. Google, Amazon, Facebook, Xiaomi, and many others create algorithms, devices, services, and ecosystems for smart gadgets, the Internet of Things, and autonomous cars.

Neural networks are used in smartphone cameras, gaming video cards and to create personalized feeds on social networks — they have already become another familiar tool.

%d bloggers like this: