CASwell Releases Whole Range Of Data Processing Units For AI Acceleration And More

New DPU Series Based On Marvell OCTEON With ARM Neoverse N2 For Data Center And Edge Computing

CASwell, a leading manufacturer of innovative and high-performance SD-WAN (uCPE/vCPE) and network security systems, launches its first family of Data Processing Units. These DPUs are based on Marvell's latest OCTEON 10 chips with 64-bit ARM Neoverse N2 cores to provide significantly greater computing performance while doubling energy efficiency compared to the previous generation. The new HSC-5340, HSC-5360, HSC-5361 and HSC-5370 DPUs are the perfect addition to CASwell's data center and edge computing appliances in need of AI acceleration.

As Artificial Intelligence (AI) and Machine Learning is being used in an increasing number of applications and industries, CASwell provides the suitable systems and acceleration modules like these new DPUs. When used in traditional server or edge appliances, the DPUs can improve performance, efficiency and cost-effectiveness of data-processing tasks like deep learning and AI.

Utilizing DPUs effectively not only offloads workloads from other parts of the data computing system, but also leads to higher performance. This is because DPUs are vital components in specialized hardware units that accelerate specific data processing tasks and provide higher performance and energy efficiency than traditional general-purpose processors like CPUs or GPUs. Combined with a DPU, these systems are ready for the next generation of AI applications.

CASwell's DPU series uses different versions of Marvell's OCTEON 10 chips with different numbers of cores, but all of them are based on the latest 64-bit ARM Neoverse N2 technology. This Armv9 infrastructure offers an enormous 40 percent performance uplift (at instructions per clock) over Neoverse N1, while retaining industry-leading performance-per-watt efficiency, supported by TSMC's 5-nanometer-process that Marvell is using for its OCTEON 10.

For the HSC family CASwell adds a MCU (Microcontroller Unit) by NXP to the DPUs for automation and control of specific devices and appliances. The DPUs also have their own memory and can be expanded by M.2 SSDs for storage purposes. Each HSC module provides a specific number of Ethernet ports for network connectivity. These DPUs fit into common PCI Express slots and are powered through the slot with additional power provided by the power supply of the host system.

Summary and availability

CASwell's DPU-based acceleration modules are ready to take your network server or edge appliance to the next level of AI integration and performance while maintaining superior energy efficiency to keep operation costs down. The best solutions when using these modules are combinations with CASwell's systems, but the DPU series can also be deployed with compatible setups.

The HSC series of DPU cards by CASwell is available now for customer evaluation.
Please contact your local CASwell representative or sales for any question!

More details about CASwell's DPU-based acceleration modules can be found here:
https:///www.cas-well.com/product-category/network-adapters-modules/dpu-based-acceleration-modules/

CASwell Inc.
No. 308, Jianguo 1st Road,
Xinzhuang District,
New Taipei City 242047,
Taiwan
https://www.cas-well.com/

CASwell, Inc. was founded in 2007 by a group of engineers with a desire to create dynamic system solutions for embedded applications. While CASwell is a young company, it has proved itself to its customers with advanced technology, professional service and superior design and manufacturing capability by delivering a great portfolio of solutions based on Linux. In 2014, the well-known IPC manufacturer Ennoconn Corporation (subsidiary of Foxconn Technology Group) decided to invest in CASwell, becoming the largest shareholder. CASwell has since become the subsidiary of Ennoconn, a member of Foxconn Technology Group. CASwell is dedicated to providing its customers with an unparalleled one-stop shopping experience for their network security and management needs.

This release was published on openPR.