Nvidia Vera ARM CPU to be 80% faster than leading x86 CPUs
Nvidia has just unveiled the RTX Spark – AI server technology that, with a Grace CPU (20 cores), a Blackwell GPU (6,144 CUDA cores), and 128 GB of LPDDR5X memory, will now also be available to consumers. And here’s a glimpse into the future of servers – and perhaps also of end-user devices one day: the new Nvidia Vera CPU.
The new Nvidia Vera CPU is one of two CPU halves of the Vera Rubin platform – the other half being the Rubin GPU. Vera promises an average speed increase of 1.8 times compared to leading x86 CPUs (Nvidia did not explicitly specify which ones). This is achieved through 88 Olympus cores (based on the ARM instruction set) with spatial multithreading for 176 threads per socket. The processor can be paired with up to 1.5 TB of LPDDR5X RAM, enabling an impressive bandwidth of 1.2 TB/s – crucial for AI inference.
Nvidia Vera with up to 22,528 cores and 45,056 threads
Nvidia Vera can be used as a standalone CPU for agent-based AI workloads, reinforcement learning, data processing, and analytics. Nvidia even developed the Vera CPU Rack, which houses 256 CPUs with 22,528 cores and 45,056 threads. Alternatively, Vera can be used as a host CPU in conjunction with Rubin GPUs. For example, the NVIDIA Vera Rubin NVL72 features 36 Vera CPUs and 72 Rubin GPUs. The CPUs and GPUs communicate via the NVIDIA NVLink C2C interface with a data transfer rate of 1.8 TB/s.
Nvidia has already won important customers – Anthropic (Claude), OpenAI (ChatGPT) and SpaceXAI (Grok) will use Vera CPUs, as will hyperscalers such as ByteDance, CoreWeave and Oracle Cloud Infrastructure.
Furthermore, Dell, HP, Lenovo, and Supermicro will develop standalone Vera CPU systems. Asus, Compal, Foxconn, Gigabyte, Pegatron, Quanta Cloud Technology, Wistron, and Wiwynn are also interested. Even the New York Stock Exchange (NYSE) is showing interest – it processes 1.1 trillion messages daily and is therefore working with Redpanda and HP to build a new infrastructure.
