NVIDIA has taken advantage of CES to present NVIDIA Rubin, its new computing platform for artificial intelligence, which promises to set a new standard in the construction, deployment, and protection of advanced AI systems, optimizing costs and accelerating the global adoption of this technology. During the event in Las Vegas, the company also announced updates such as DLSS 4.5 Super Resolution, multiple frame generation, and native GeForce Now support for Linux, reinforcing its leadership in hardware and services for video games and data centers. However, the most notable launch is Rubin, a platform that integrates six key chips: the NVIDIA Vera CPU, the NVIDIA Rubin GPU, the NVLink 6 switch, the SuperNIC ConnectX-9, the BlueField-4 DPU, and the Spectrum 6 Ethernet switch. This extreme design significantly reduces training times and inference costs, according to Jensen Huang, CEO of NVIDIA. Rubin, named in honor of astronomer Vera Rubin, includes solutions such as the Vera Rubin NVL72 rack and the HGX Rubin NVL8 system. Its innovations include the sixth generation of NVLink, the Transformer Engine, confidential computing, and the RAS engine, along with the new Vera CPU. These technologies drive agency AI, advanced reasoning, and large-scale MoE (Mixture of Experts) model inference, achieving a cost per token up to 10 times lower than the Blackwell platform and allowing MoE models to be trained with four times fewer GPUs. The Rubin platform addresses the growing demands of complex AI workloads, such as video generation and multi-step reasoning, which require processing and analyzing long sequences of data. Its five main technologies are: - NVLink 6: Provides 3.6 TB/s of bandwidth per GPU and 260 TB/s per rack, surpassing the bandwidth of the entire Internet, and facilitates collective operations and improved resilience. - NVIDIA Vera CPU: With 88 custom Olympus cores and ARMv9.2 support, it offers energy efficiency and high performance for AI data centers. - Rubin GPU: Incorporates a third-generation Transformer engine and 50 petaflops of NVFP4 processing for inference. - Confidential computing: Ensures the security of data and models on CPU, GPU, and NVLink, protecting critical workloads. - RAS engine: Provides real-time monitoring, fault tolerance, and proactive maintenance, with a modular design that streamlines assembly and service. NVIDIA Rubin will be backed by major industry players such as AWS, Google, Meta, Microsoft, OpenAI, Dell, Lenovo, Oracle, among others. The platform is already in production, and the first products will be available through NVIDIA's partner ecosystem in the second half of 2026. Source: muycomputerpro.com