NVIDIA Hot Chips: Inference, AI, Networking & Innovation
NVIDIA to showcase AI Advancements at Hot chips Conference
Next week’s hot Chips conference (August 24-26 at Stanford University) will heavily feature advancements in AI reasoning, inference, and networking. The conference is a key forum for processor and system architects from both industry and academia, and will showcase innovations driving the trillion-dollar data center computing market and advancing AI factories.
NVIDIA will participate in a tutorial session on August 24,alongside Google and Microsoft,focusing on designing rack-scale architecture for data centers. NVIDIA experts will also present at four sessions and one tutorial, detailing:
Networking: How NVIDIA ConnectX-8 SuperNIC delivers AI reasoning at rack- and data-center scale (presented by idan Burstein, principal architect of network adapters and systems-on-a-chip at NVIDIA). NVIDIA NVLink, NVLink Switch and nvlink Fusion provide scale-up connectivity for low-latency, high-bandwidth data exchange.
Neural Rendering & Inference: Advancements powered by the NVIDIA Blackwell architecture, including the NVIDIA GeForce RTX 5090 GPU, offering next-level graphics and simulation capabilities (presented by Marc Blackstein, senior director of architecture at NVIDIA).
Co-packaged Optics (CPO): How CPO switches, utilizing silicon photonics for faster and more efficient data transmission, enable gigawatt-scale AI factories. This will include a discussion of NVIDIA Spectrum-XGS Ethernet, a technology for unifying distributed data centers into AI super-factories (presented by Gilad Shainer, senior vice president of networking at NVIDIA).
NVIDIA GB10 Superchip: Its role as the engine within the NVIDIA DGX Spark desktop supercomputer (presented by Andi skende,senior distinguished engineer at NVIDIA).
NVIDIA’s technologies are focused on accelerating inference and driving AI innovation across all scales. AI reasoning,requiring rack-scale performance,relies on networking as the central nervous system of data centers. NVIDIA Spectrum-X Ethernet connects clusters and orchestrates GPU-to-GPU interaction, while Spectrum-XGS Ethernet extends this performance to interconnect distributed data centers. The NVIDIA GB200 NVL72, featuring 36 GB200 superchips, delivers exascale computing within a single rack.