[Editor's Note] FPGAs are non-memory semiconductors capable of circuit redesign, featuring software-based parallel processing, programmability, and low power consumption. Since 2016, the AI acceleration processor market has grown and changed rapidly alongside the growth of AI. Currently, their utility is recognized in various industries requiring flexible chip processing, such as AI, data centers, and semiconductor equipment. This magazine met with Na Hee-seok, head of Uniquest, the official domestic distributor of Intel FPGAs, to learn about large-scale data processing methods and the latest Agilex M-Series and Agilex I-Series devices in the Intel Agilex FPGA product line.

■ What are the considerations for large-scale data processing?

In server and storage systems that must process large amounts of data, high-speed data processing and transmission, as well as safe and fast storage processes, are very important.

Memory must be used appropriately to increase data storage capacity and connection speed.

Simply increasing external memory capacity or using only high-speed memory cannot create the memory hierarchy required by various systems.

Implementing various high-speed transmission connections and multiple memories in a single system is not suitable in terms of technical limitations of board manufacturing, development costs, and system energy efficiency.

Ultimately, what is needed is a flexible system design that can be used in a variety of applications.

■ Are there any limitations to CPU utilization when processing data at high speed?

Just increasing CPU performance like in existing systems has limitations in processing large amounts of data at high speeds.

Although data processing performance can be improved by increasing the number of multi-cores in the CPU, the CPU This is because there are many limitations in improving the bandwidth of internal/external memory and high-speed connection between devices due to the limited number of IOs per package.

In addition, various compression algorithms must be processed at high speed to reduce data transmission capacity.

You may need a device to accelerate parallel data processing.

In systems that process this explosive growth in data, fast-response connections between devices and high-bandwidth memory solutions are required, and the implementation of such systems is possible with the applied technology of Intel Agilex products.

■ Please give a brief introduction to the Intel Agilex M-Series and Agilex I-Series.

First, the M-Series features the I-series base, supports DDR5 interface, and has an upgraded 32GB HBM2e built-in, making it the optimal device for fields requiring high-bandwidth memory.

In addition, it supports 116Gbps PAM4, PCIe Gen5, and CXL (computing express link) supported by the I-series.

*CXL (Compute Express Link) is a newly proposed interface for more efficient use of accelerators, memory, and storage devices used together with CPUs in high-performance computing systems.

With the emergence of the metaverse, AI (artificial intelligence), and big data, which require processing of vast amounts of data, it is attracting attention as an alternative that can overcome the limitations of existing computing performance.

Both the M series and the newly added I-series device family are planned for production launch in the first half of next year, starting with the ES devices in the second half of this year.

■ Utilizing Intel FPGAs What about the case?

As of now, the Agilex M series and I series devices have only been released in limited quantities, so there are not many application cases. However, I understand that a demo system has been built to verify high-speed transmission devices exceeding 400G and HPC functions in some cloud fields.

The Agilex F Series product line, already released, is being developed for high-speed data processing and transmission in wired, wireless, and broadcast applications.

As low-capacity Agilex devices are released in the future, they are expected to be used more widely across various industries.

■ Views on Intel's future FPGA market strategy

With respect to the limitations discussed, we believe that there will continue to be a variety of fields that require the circuit changeability and flexibility features of FPGAs.

The goal is to continuously respond to and update the area by targeting niche markets for broad demand.

■ Lastly, please say a word to the readers.

Many industries are looking for solutions for rapid processing and transmission of exploding data and for storing large amounts of data.

However, technological limitations still exist and it is difficult to find solutions that are suitable for each application.

The Intel Agilex FPGA family offers developers unique, optimized solutions that address their specific needs. These powerful FPGAs can support a variety of interface standards and integrate multiple functions.

We hope you'll join us for a webinar on Tuesday, June 14th at e4ds.com to learn about the new technologies and features of the Intel Agilex family.
;