In a landmark move, tech titan Nvidia is taking a daring leap into the realm of artificial intelligence for the world of gaming. The firm is launching an avant-garde AI-powered tool known as the Avatar Cloud Engine or Ace, initially designed for service industries but now repurposed to revolutionize how gamers interact with non-player characters (NPCs) in the virtual world.
Ace represents a trifecta of sophisticated AI tools: Nemo, Riva, and Omneeverse Audio-2-Face. With Nemo, game designers are equipped with a potent language model, enabling them to craft unique dialogues and narratives for individual characters. Simultaneously, Riva empowers real-time, interactive conversations with Nemo by combining advanced speech recognition and text-to-speech conversion. In a grand finale, Audio-2-Face transforms Riva’s audio outputs into realistic facial animations for in-game characters, notably through the conduit of Unreal Engine 5’s Meta-Human characters via Omneeverse connections.
An exciting partnership with AI start-up Convai offers a tantalizing preview of a gaming character energized by Ace. In an awe-inspiring demo, NPC Jin communicates with players in everyday language, exhibiting an incredible level of real-time responsiveness and adherence to a pre-defined backstory.
The true magic behind this capability lies in Ace’s fine-tuned AI models. They offer a range of performance capabilities, optimizing for low latency to ensure immersive, lifelike interactions. Nvidia’s promise is that their Ace for Games foundry service will help developers calibrate these models, ready to be launched via the Nvidia DGX Cloud, Jee-Force RTX PCs, or locally for real-time inferencing.
Among the pioneers ready to harness Ace’s capabilities is GSC Game World, planning to implement Audio-2-Face in their upcoming release, Stalker 2 Heart of Chernobyl. However, eager gamers may need to exercise patience for fully AI-rendered narratives around virtual bonfires.
But Nvidia’s AI breakthroughs are not confined to the gaming world. The tech behemoth has also unveiled the DGX GH 200 AI supercomputer. This formidable device, equipped with 256 Grace Hopper chips, holds an astonishing 144 terabytes of memory and delivers exaflop AI performance. This monstrous machine can train GPT3 over twice as fast as a DGX H100 cluster, as per Nvidia’s claims.
Big tech firms Google, Meta, and Microsoft are among the first to access the DGX GH 200, keen to assess its potential. Through these developments, Nvidia is leading the charge in an AI revolution within and beyond the gaming industry.
The DGX GH 200’s supercomputer architecture leverages NV-Link interconnect technology and the Switch System to unify the power of 256 GH200 Superchips, acting as a single gargantuan GPU. The result is an astonishing one exaflop of performance and an unparalleled shared memory capacity of 144 terabytes. In comparison to its predecessor, the NVIDIA DGX A100, the DGX GH200 shines with almost 500 times more memory, significantly boosting its AI capabilities.
Moreover, the supercomputer boasts 48 times more NV-Link bandwidth than its predecessor, blending the sheer power of a colossal AI supercomputer with the ease of programming a solitary GPU.