NVIDIA’s 72-Core Grace “ARM” CPU Is Almost As Fast As 96-Core AMD Threadripper 7995WX Chip In Geekbench

NVIDIA's 72-Core Grace "ARM" CPU Is Almost As Fast As 96-Core AMD Threadripper 7995WX Chip In Geekbench 1

NVIDIA's Grace CPU has appeared within the Geekbench benchmark, showcasing some strong multi-threaded performance against Intel & AMD offerings.

NVIDIA's Grace CPU was designed to meet the high-performance computing requirements for the server and data center ecosystems. The chip has been deployed in NVIDIA's GH200 Superchip that offers the fastest AI performance at the moment and will also be used to power the next-gen B200 solutions such as the GB200 which is going to feature two Blackwell chips alongside a single Grace CPU.

Talking about specifications, the NVIDIA Grace CPU is based on the ARM architecture and features a total of 72 Arm Neoverse N2 cores. The CPU features LPDDR5x memory on the motherboard and there are two distinct platforms, the Grace Superchip with two of these Grace CPUs for a total of 144 cores, and the Grace Hopper Superchip with a 72-core Grace CPU and the Hopper H200 GPU. Each chip features 117 MB of L3 cache, 68 PCIe Gen5 lanes, and is built on the TSMC 4N process node.

Some of the main highlights of Grace include:

High-performance CPU for HPC and cloud computing

Super chip design with up to 144 Arm v9 CPU cores

World’s first LPDDR5x with ECC Memory, 1TB/s total bandwidth

SPECrate2017_int_base over 740 (estimated)

900 GB/s coherent interface, 7X faster than PCIe Gen 5

2X the packaging density of DIMM-based solutions

2X the performance per watt of today’s leading CPU

Runs all NVIDIA software stacks and platforms, including RTX, HPC, AI, and Omniverse

NVIDIA has previously shared some impressive performance and efficiency figures for its Grace CPU against Intel's Sapphire Rapids and AMD's Genoa offerings but this time, we are looking at the Geekbench performance of this server-oriented chip against some high-end workstation parts from Intel and AMD.

The performance was evaluated on the Supermicro ARS-111GL-NHR system which is a 1U design that supports the NVIDIA GH200 platform with a single Grace CPU. The CPU was running at clock speeds of 3.35 GHz across all cores at peak. The Grace CPU also has a rated TDP of around 200W since the Grace Superchip is rated at 500W which includes the power for the LPDDR5x modules.

The NVIDIA Grace CPU scored 1636 points in the single-core and a massive 74,440 points in the multi-core test. This puts the chip close to AMD's flagship Threadripper 7995WX CPU which is a 96-core behemoth running at much higher clock speeds of 5.1 GHz and sipping in up to 350W power. The chip also crushes Intel's top Xeon Workstation offering even with LN2 cooling and over 1000W TDP.

These scores are very impressive and considering that server offerings are lower-clocked, they will be slower in the same benchmark even with the same amount of cores so this is a really good showcase for NVIDIA's Grace CPU but this is just one benchmark and performance varies workload from workload. So while the performance is great in Geekbench, it might be a whole different story within server applications.

NVIDIA is expected to launch its next-gen Vera CPUs in 2026 which will be the next chapter in NVIDIA's Arm endeavors plus the company is also expected to unveil Arm-based consumer-oriented options for the "AI PC" segment in the coming years so lot of development is expected within the Arm markets.

News Source: HXL (@9550pro)