NVIDIA unveiled its next-generation Vera data center CPU, featuring in-house "Olympus" cores designed specifically for agentic AI and modern cloud workloads. The processor is slated for a release in the second half of 2026 and will serve as the host CPU for the NVIDIA NVL72 Vera Rubin system. Unlike the predecessor Grace CPU which utilized Arm Neoverse-V2 cores, the Vera chip employs an 88-core Olympus design that supports the Armv9.2 ISA. Key architectural specifications include support for FP8 precision, 176 total threads via spatial multi-threading, and integration with LPDDR5X memory to provide up to 1.2TB/s of memory bandwidth. The processor also features a 2MB L2 cache per core, a 164MB unified L3 cache, and support for both PCIe Gen 6 and CXL 3.1 connectivity.

Early benchmarks conducted on pre-production systems at NVIDIA's Santa Clara headquarters demonstrate performance levels competitive with modern x86_64 server CPUs from Intel and AMD. According to the data, the chip claims to deliver double the performance of the Grace predecessor while maintaining high energy efficiency. The unit tested operated at a peak 450 Watt socket TDP. While power management and frequency monitoring were restricted during this initial assessment due to ongoing driver tuning, the chip showed robust compatibility with the upstream Linux ecosystem.

The Vera CPU supports mainstream ARM64 Linux distributions such as Ubuntu and Fedora. Upstream Linux kernel support is already active in the 7.1+ series, and NVIDIA has provided necessary patches within its modified Ubuntu Base OS. Developers can build optimized binaries using GCC 16.1+ or LLVM Clang 21+, as NVIDIA upstreamed its Olympus core support well in advance of the hardware launch. The company confirmed that future production models will be optimized for enclosed server chassis designs, with more comprehensive benchmarking and power efficiency data expected as the release date approaches later in 2026.

NVIDIA unveiled its next-generation Vera data center CPU, featuring in-house "Olympus" cores designed specifically for agentic AI and modern cloud workloads. The processor is slated for a release in the second half of 2026 and will serve as the host CPU for the NVIDIA NVL72 Vera Rubin system. Unlike the predecessor Grace CPU which utilized Arm Neoverse-V2 cores, the Vera chip employs an 88-core Olympus design that supports the Armv9.2 ISA. Key architectural specifications include support for FP8 precision, 176 total threads via spatial multi-threading, and integration with LPDDR5X memory to provide up to 1.2TB/s of memory bandwidth. The processor also features a 2MB L2 cache per core, a 164MB unified L3 cache, and support for both PCIe Gen 6 and CXL 3.1 connectivity.

Early benchmarks conducted on pre-production systems at NVIDIA's Santa Clara headquarters demonstrate performance levels competitive with modern x86_64 server CPUs from Intel and AMD. According to the data, the chip claims to deliver double the performance of the Grace predecessor while maintaining high energy efficiency. The unit tested operated at a peak 450 Watt socket TDP. While power management and frequency monitoring were restricted during this initial assessment due to ongoing driver tuning, the chip showed robust compatibility with the upstream Linux ecosystem.

The Vera CPU supports mainstream ARM64 Linux distributions such as Ubuntu and Fedora. Upstream Linux kernel support is already active in the 7.1+ series, and NVIDIA has provided necessary patches within its modified Ubuntu Base OS. Developers can build optimized binaries using GCC 16.1+ or LLVM Clang 21+, as NVIDIA upstreamed its Olympus core support well in advance of the hardware launch. The company confirmed that future production models will be optimized for enclosed server chassis designs, with more comprehensive benchmarking and power efficiency data expected as the release date approaches later in 2026.