Nvidia announced Ada Lovelace GPU architecture, new cloud services, and began shipping H100 datacenter GPUs.

At its Fall GTC, NVIDIA unveiled the Lovelace GPU, designed primarily for gamers and crypto miners, which promises to transform gaming from pre-calculated imagery into a fully simulated virtual world. The Lovelace GPU features 70% more cores than Ampere and 90 TFLOPS with shader execution reordering. DLSS-3, a breakthrough in real-time graphics and ray tracing, delivers 4X better performance for real-time ray tracing using AI to generate entire intermediate frames at 100 Frames Per Second, rather than relying solely on pixel generation. Beyond gaming, NVIDIA announced numerous enterprise products including an H100 update, NeMo LLM services, IGX for Medical Devices, Jetson Orin Nano, Isaac SIM, a new Drive platform, and Omniverse Cloud.

The company advanced its data center roadmap with Hopper, the new GPU offering a five-fold boost in performance for training Transformer networks. The PCIe version is now shipping to partners including Dell, HPE, Supermicro, Cisco, and Lenovo for availability the following month, while SXM and NVLink versions using NVIDIA DGX servers and HGX modules will ship in Q1 2023, with wide-scale public cloud support expected in the same timeframe. Hopper significantly advances inference capabilities, supporting 300 concurrent users of LLM models—a 30-fold increase over the A100—critical as generative AI adoption accelerates, exemplified by Midjourney's over 1M active users creating art from text prompts and DALL-E 2's recent opening to another million users.

NVIDIA also detailed Grace, an Arm-based server chip under development targeting twice the performance of today's servers. The Grace-Hopper superchip will deliver 7X the fast-memory capacity at 4.6TB and 8000 TFLOPS versus current CPU-GPU configurations, essential for the recommender models used by eCommerce and media super-websites. Grace will support 72 fast cores at twice the single-threaded performance of today's Arm and X86 servers. To address adoption barriers, NVIDIA announced NeMo LLM Cloud services, enabling customers to add domain-specific prompts to fully-trained models without the costs exceeding $10M required to train multi-billion parameter models from scratch.

NVIDIA expanded its cloud offerings with new services for Omniverse, its metaverse platform targeting professional and creative communities, with existing customers including home improvement provider Lowe's, Charter Communications for 5G, and German railroad operator Deutsche Bahn. The company unveiled the second-generation OVX server platform for enterprises building and operating Omniverse applications at scale, featuring 8 next-generation GPUs, 3 SmartNICs, 2 Intel CPUs, and 16 TB of NVMe memory. These announcements, including partnerships with Deloitte and Booz Allen Hamilton, underscore NVIDIA's positioning across gaming, data center AI training and inference, cloud services, and professional metaverse infrastructure.