Nvidia has announced the RTX Spark, a groundbreaking AI server technology that features a Grace CPU with 20 cores, a Blackwell GPU boasting 6,144 CUDA cores, and 128GB of LPDDR5X RAM. This sets the stage for future advancements in server technology, potentially extending to consumer devices.
The Vera CPU forms an integral part of the Vera Rubin platform, complemented by the Rubin GPU. Nvidia claims that Vera delivers an impressive 1.8x average speedup compared to unnamed leading x86 CPUs.
Vera is a robust processor featuring 88 Olympus cores based on the ARM architecture, equipped with Spatial Multithreading capability, allowing for 176 threads per socket. It can be paired with up to 1.5TB of LPDDR5X RAM, enabling an extraordinary 1.2TB/s bandwidth, essential for AI inference tasks.
This CPU can function independently for various applications, including agentic AI workloads, reinforcement learning, and data processing. Furthermore, Nvidia has developed the Vera CPU Rack, accommodating 256 CPUs to create a powerful system with 22,528 cores and 45,056 threads.
Alternatively, Vera can serve as a host CPU in conjunction with Rubin GPUs. For instance, the NVIDIA Vera Rubin NVL72 integrates 36 Vera CPUs and 72 Rubin GPUs, allowing communication at speeds up to 1.8TB/s via Nvidia's NVLink-C2C interconnect.
Nvidia has secured several key partnerships with organizations including Anthropic (Claude), OpenAI (ChatGPT), and SpaceXAI (Grok), who will deploy Vera CPUs. Additionally, major cloud service providers such as ByteDance, CoreWeave, and Oracle Cloud Infrastructure are on board.
OEMs such as Dell, HP, Lenovo, and Supermicro are also committed to manufacturing standalone Vera CPU systems. Other partners include Asus, Compal, Foxconn, Gigabyte, Pegatron, Quanta Cloud Technology, Wistron, and Wiwynn. Even the New York Stock Exchange, processing 1.1 trillion messages daily, is collaborating with Redpanda and HP to enhance its infrastructure.
Source
