Nvidia Unveils GB200 NVL4 with Two Grace CPUs and Four Blackwell GPUs for Modern Data Center Workloads
- Nvidia CPUs designed for scalable solutions with PCIe connectivity
- Delivers 144 cores and four GPUs for AI acceleration
- Thanks to its 1.3 TB memory, NVL4 is ideal for data-heavy applications
Nvidia has unveiled the GB200 NVL4, an advanced platform designed to meet the needs of modern data centers and computational workloads.
The company’s decision to introduce the GB200 NVL4 comes after Nvidia recently restructured its product lineup, pushing aside other NVL platforms in favor of newer options like the NVL4.
The GB200 NVL4 combines two Arm-based Grace CPUs with four Blackwell GPUs. It is part of Nvidia’s Grace Blackwell series, which aims to provide flexible and scalable solutions for various deployment scenarios.
High speed and scalable
The GB200 NVL4 is designed as a mid-range option in the Grace Blackwell family and sits below high-performance configurations such as the NVL72 and NVL36. Each unit features two Grace CPUs, each equipped with 72 Arm Neoverse V2 cores, for a total of 144 cores. The addition of four Blackwell GPUs provides robust acceleration for AI, HPC, and other compute-intensive tasks.
The six MCIO connectors under each CPU provide PCIe connectivity, allowing high-speed data transfers. It also supports the inclusion of NICs, SSDs and other essential components while maintaining efficiency.
Power consumption is a critical factor in modern data centers, but Nvidia estimates that the GB200 NVL4 will consume just over 6 kW per server when fully configured. While this represents a significant power consumption, it is a marked improvement over larger platforms, with previous systems such as the Nvidia DGX-1 or HGX-1 consuming around 3.5 kW.
Moreover, this device also supports up to 1.3TB of combined memory, ensuring efficient data handling and processing and making it a strong candidate for memory-intensive applications. The NVL4 is expected to fill a niche for organizations looking for powerful yet relatively power efficient solutions compared to the higher GPU count NVL platforms.
The GB200 NVL4 and the more expensive NVL72 may seem the same at first glance, but there are notable differences. The NVL72 features edge connectors for backbone configuration and targets larger-scale deployments with dual compute resources. The NVL4, on the other hand, is designed to be a more compact and power-conscious node, while still delivering significant performance for its size.