# NVIDIA'S HUGE AI Announcements Will Change Everything (Here's Why) Source: https://www.youtube.com/watch?v=y8oHx6dPgqE Recap page: https://rapidrecap.app/video/y8oHx6dPgqE Generated: 2026-02-25T21:49:19.579+00:00 --- ## Quick Overview NVIDIA announced significant advancements across its AI ecosystem, including the Blackwell GPU generation, the Vera Rubin compute tray, the new ConnectX-9 SuperNIC, and the Quantum InfiniBand Switch, all designed to deliver up to 10x performance improvements over previous generations, particularly for large AI models and data-intensive tasks. **Key Points:** - NVIDIA Blackwell GPUs offer up to 10x performance gains over the previous generation, specifically in factory throughput for AI reasoning tasks like Kimi K2-Thinking (32K/8K). - The Blackwell-based Vera Rubin NVL72 Compute Tray features a modular, liquid-cooled design with no fans or external hoses, designed for easy serviceability and high density. - The Vera Rubin system contains 72 GPUs and 2 Grace CPUs per rack, achieving performance gains across both compute and networking. - NVIDIA ConnectX-9 SuperNIC, featuring the BlueField-4 DPU, provides 800 Gb/s Ethernet, 2x networking performance vs. BF3, and enhanced security features. - The Blackwell generation uses 6 new chips co-designed to work together, including the NVLink Switch chips that deliver 1.8 TB/s per port and 130 TB/s of aggregate bandwidth for all-to-all GPU communication. - The Quantum-X InfiniBand Switch and Spectrum-X Ethernet Switch (Co-Packaged Optics) offer significant bandwidth increases (800 Gb/s per port for Quantum-X) while reducing power consumption and complexity compared to previous architectures. - The host invites viewers to attend GTC online from March 16-19 for further details and offers a chance to win an RTX 4090 graphics card. ![Screenshot at 00:24: The video agenda slide outlines the key topics to be covered: AI Agents, Blackwell, Vera Rubin, Future of AI, and a GTC Sneak Peek, setting the stage for the subsequent hardware deep dive.](https://ss.rapidrecap.app/screens/y8oHx6dPgqE/00-00-24.jpg) **Context:** The video features an interview with Joe DeLaere, Product Lead of AI Infrastructure at NVIDIA, discussing several key hardware and infrastructure announcements, likely stemming from a recent major event like GTC. The discussion centers on the generational leap in performance, density, and efficiency provided by the new Blackwell architecture, contrasting it with the previous generation (implied to be Hopper/Grace) across GPUs, networking components, and entire rack designs like the DGX SuperPOD and the new Vera Rubin architecture. ## Detailed Analysis The video details NVIDIA's latest AI infrastructure announcements, starting with the massive performance leap provided by the new Blackwell GPUs, which deliver up to 10x greater factory throughput compared to the previous generation for AI reasoning tasks. Joe DeLaere explains that the entire ecosystem is co-designed for this performance, highlighting the Blackwell-based Vera Rubin NVL72 compute tray. This new tray features a highly modular, liquid-cooled design, eliminating fans and hoses for increased reliability and density, allowing all 72 GPUs and 2 Grace CPUs within the tray to operate efficiently. DeLaere contrasts the Rubin architecture with the previous generation's architecture, noting that the new design allows for servicing individual compute modules more easily. For networking, the ConnectX-9 SuperNIC, powered by the BlueField-4 DPU (which itself offers 2x networking and 6x compute improvements over BF3), provides 800 Gb/s Ethernet speeds. Furthermore, the massive internal bandwidth is enabled by 18 NVLink Switch chips, supporting 72 GPUs with 1.8 TB/s per port, totaling 130 TB/s of aggregate bandwidth for all-to-all communication. The discussion also covers the new networking hardware: the Quantum-X InfiniBand Switch and the Spectrum-X Ethernet Switch, which utilize co-packaged optics to maintain high performance while reducing complexity and power consumption. Finally, the host promotes attending the upcoming GTC event live from March 16-19 for more details and offers viewers a chance to win an RTX 4090 graphics card by registering via the provided QR code. ### Blackwell & Vera Rubin Architecture - Vera Rubin NVL72 Compute Tray is modular, liquid-cooled, and cable-free; it houses 72 GPUs and 2 Grace CPUs, achieving 10x performance gains in factory throughput over the previous generation. ### Blackwell Compute Performance - Generational performance gains are seen at both the chip level and rack level, with the new architecture designed to handle increased compute demands for models exceeding a trillion parameters. ### Networking & Connectivity - The system utilizes the new 6th generation NVLink Switch chips for GPU-to-GPU communication, offering 1.8 TB/s per port. The ConnectX-9 SuperNIC provides 800 Gb/s Ethernet and includes the BlueField-4 DPU for offloading management and security tasks. ### Spectrum-X Photonics - NVIDIA's Spectrum-X Ethernet switch utilizes co-packaged optics, offering 102.4 Tb/s scale-out infrastructure with 352 billion transistors, contrasting with Quantum-X InfiniBand solutions. ### Serviceability & Density - The modular design of the Vera Rubin tray, featuring hot-swappable bases and simplified liquid cooling connections, improves serviceability and density compared to previous generations which often relied on fans and complex cabling. ### GTC Promotion - The video concludes by encouraging viewers to register for the NVIDIA GTC conference (March 16-19) for free online access and an opportunity to win an RTX 4090 graphics card. ![Screenshot at 00:10: The presentation shows the GPU/System management dashboard interface used to launch AI jobs, indicating a complex software environment.](https://ss.rapidrecap.app/screens/y8oHx6dPgqE/00-00-10.jpg) ![Screenshot at 00:21: A detailed look at the internal PCB featuring the NVIDIA BlueField-3 DPU, highlighting the dedicated infrastructure processor.](https://ss.rapidrecap.app/screens/y8oHx6dPgqE/00-00-21.jpg) ![Screenshot at 01:11: Jensen Huang presents multiple server modules on stage, illustrating the evolution or range of NVIDIA's compute offerings.](https://ss.rapidrecap.app/screens/y8oHx6dPgqE/00-01-11.jpg) ![Screenshot at 03:39: An exploded view animation shows the components of the GB200 NVL72 Superchip, detailing the two Blackwell GPUs and one Grace CPU working together.](https://ss.rapidrecap.app/screens/y8oHx6dPgqE/00-03-39.jpg) ![Screenshot at 09:06: Two types of large server racks are shown—one containing compute trays and the other a specialized switch rack—highlighting the physical separation of compute and networking elements in the architecture.](https://ss.rapidrecap.app/screens/y8oHx6dPgqE/00-09-06.jpg)