Intel Shows Off New Xeon Max Processors Ahead of Supercomputing ‘22

November 14, 2022 by Aaron Carman

Intel's newest CPUs and GPUs are purpose-built for high-performance, advanced computing applications.

Ahead of its Supercomputing '22 event, Intel announced the Max Series product family consisting of two processors built for high-performance computing (HPC) and artificial intelligence (AI) applications. The announcement represents Intel’s efforts to commercialize its own high-performance processors and compete with big names in the HPC community like NVIDIA and AMD. The Intel Max Series offers some competitive specs, giving HPC system architects a new option to improve their computational capacity.


Intel Max Series product information

Intel Max Series product information highlighting key features of the newly announced CPU and GPU. Image used courtesy of Intel

To gain a better understanding of Intel’s contributions to the HPC space, we'll compare the Max Series processor both to previous generations of Intel products and to analogous CPUs and GPUs available on the HPC market. 


Taking Performance to the Max

Intel's Max Series includes two new processors for HPC applications: the Xeon CPU Max Series (previously code-named Sapphire Rapids HBM) and the Intel Data Center GPU Max Series (previously code-named Ponte Vecchio). Both products will be included in the Aurora supercomputer at Argonne National Laboratory.

The Xeon CPU Max Series builds upon Intel's existing Xeon product family, improving the computational performance by sporting up to 56 performance cores with a maximum thermal design power (TDP) of 350 W. In addition, the Xeon Max CPU is the first x86-based processor to contain high-bandwidth memory (HBM), giving it an appreciable boost in performance and circumventing the limitations of the von Neumann architecture.


Max Series CPU with onboard HBM

Max Series CPU with onboard HBM shown along the edge of the module. The included HBM allows the CPU to effectively accelerate complex calculations. Image used courtesy of Intel

The Xeon Max processor includes up to 64 GB of in-package HBM, giving each core over 1 GB of memory capacity, making it sufficient for most HPC applications. Intel reports up to 4.8x better performance on real-world HPC workloads with up to 68% less power usage than AMD’s Milan-X cluster.


Scalable Processing for All HPC Users

Making its appearance alongside the Xeon Max CPU, Intel’s Data Center GPU Max Series also reports some major performance advantages compared to legacy hardware. Max Series GPUs will include up to 128 Xe-HPC cores, purpose-built to tackle tough HPC workloads. The flagship module will also include 128 GB of HBM, further speeding up calculations in an HPC environment.

If flagship performance is not required, Intel offers scaled solutions to ensure that designers can reliably choose the module to accomplish their own goals. In addition, using the Intel Xe Link bridge, HPC architects can easily connect multiple modules to enable inter-GPU communication. The Max Series GPUs have reportedly allowed for a 2.4x performance boost in finance-based computing and a 1.5x boost in physics simulations compared to NVIDIA’s A100 card.


Bolstered Computing for Next-Gen Applications

With some known HPC applications in mind, Intel hopes the Max Series will help designers make the transition to Intel hardware both simple and enticing. For AI use, the Xeon Max includes advanced matrix (AMX) extensions alongside other accelerators to improve performance and gives the designers the flexibility to include DDR memory in addition to the HBM included in the package.


Xe matrix extensions

Xe matrix extensions highlight the processor’s ability to accelerate computation with common AI data types. Screenshot used courtesy of Intel

Intel will demonstrate the Max Series’ performance at Supercomputing ’22 alongside system designs using the Max Series modules. Although the performance of Intel’s designs hasn’t been verified by outside parties, the reported numbers are a positive step toward a more diverse catalog of HPC modules for advanced computing applications.