Author: Zhou Yuan / Wall Street Watch

The FPGA, which lies between general-purpose GPUs and dedicated ASICs, has finally gained independence as Altera, formerly part of Intel's Programmable Solutions Group (PSG), specializing in this field.

On January 10, Altera announced on the X platform, "Today, we proudly mark a milestone moment, officially raising the Altera flag as an independent FPGA company. We are passionate and will drive future innovation with agility and focus, shaping the next era of FPGA technology at altera.com."

This "new company" will maintain a strategic partnership with Intel, but in the field of FPGA, it will have greater decision-making freedom than when it was part of Intel's PSG.

The relationship between general-purpose GPUs, FPGAs, and dedicated ASIC chips can be simply summarized: FPGAs are positioned between the other two, and compared to standard GPUs, FPGAs are a semi-custom form that is particularly suitable for adapting to rapidly changing technological demands without incurring the high costs of ASIC development.

Standardized GPUs have relatively fixed architectures, which may limit efficiency for certain specific non-graphical and non-highly-parallel computing tasks.

FPGAs are programmable hardware devices, with internal logic blocks and wiring resources that can be reconfigured through programming. This flexibility allows FPGAs to customize hardware circuits based on specific application requirements, thus achieving high performance for certain specific computing tasks.

The advantages of ASIC are clear: it is highly optimized for specific algorithms or tasks, achieving extremely high performance, efficiency, and power consumption ratios; however, the disadvantages are also significant: once the design and manufacturing are completed, it is difficult to modify, thus lacking flexibility, which leads to higher design and manufacturing costs, and is typically suited for specific applications that require large-scale mass production.

FPGA is mainly used in application scenarios that require high customization and flexibility, such as high-speed signal processing (communication), real-time control and signal processing of equipment like Electric Machines and Siasun Robot&Automation in industrial automation, and in Datacenters to accelerate certain specific workloads; GPU is widely used in graphics processing as well as deep learning and inference; fields with extremely high performance and power requirements and relatively fixed functions belong to the domain of ASIC.

Altera and Xilinx were once the two largest companies in the FPGA field. In 2015, Intel acquired Altera for 16.7 billion dollars. This was Intel's largest acquisition in history.

In the field of GenAI technology, FPGA is highly flexible, and the cost of adding new instructions and data formats is relatively low. In addition to custom AI applications, FPGA can also be used for simulation testing on chips before the launch of AI processors.

This simulation is built on integrated frameworks such as Altera's FPGA AI Suite, TensorFlow, and PyTorch, allowing developers to construct optimized solutions compatible with industry standards.

Since it is independent from Intel's original system, Altera must bundle technological supply with Intel. For example, Altera will use Intel's foundry services (IF). However, this technological cooperation is not exclusive. Altera can also use other foundries, such as Taiwan Semiconductor or Samsung Electronics for chip manufacturing.

The CEO of Altera is Sandra Rivera, and the COO is Shannon Poulin.

Given that the standard GPU manufacturer NVIDIA is so successful, and the ASIC-focused Broadcom also shines brightly, after Altera deepens its focus on FPGA and gains the decision-making power that an independent company should have after separating from the Intel system, will it also demonstrate impressive performance?

Intel is expected to be optimistic about this, so at the right time, Altera is likely to promote an IPO.

Intel, currently mired in financial difficulties, has swiftly replaced its former CEO (Pat Gelsinger) and simultaneously opened the door for independent operation of its various potential businesses.

In addition to Altera, it also includes Intel's perception computing business RealSense.

Recently, there have been reports that Intel intends to establish RealSense as an independent operation under Intel Capital, expected to be completed in the first half of this year. In response, Intel stated, "The decision to divest RealSense is not related to its recent financial difficulties."

RealSense is both a business of Intel and a technological framework used for robots (as shown in the figure).

Intel's RealSense technology aims to change the way robots perceive their environment and interact with it.

RealSense technology includes hardware and software integration, with hardware mainly consisting of depth cameras with built-in visual processors; software includes the open-source RealSense SDK 2.0, which supports all Intel RealSense camera products and is compatible with various operating systems, ROS (Robot Operating System), and programming languages, making it easy to integrate with solutions from third-party software providers.

Intel is trying to set a new industry standard for robotic intelligence and vision by leveraging its RealSense technology, which features high-resolution imaging and depth perception.

It is difficult to clarify Intel's intentions regarding RealSense technology when looking at the company's overall attitude.

In 2012, Intel began investing in perceptual computing R&D, establishing a Fund with an investment of $0.1 billion; in early 2014, Intel officially renamed the perceptual computing technology to RealSense and launched it into the market.

At CES 2016, LENOVO GROUP decided to integrate Intel's RealSense technology into multiple system plugin options, and introduced a display supported by Windows Hello, equipped with a RealSense camera.

In 2020, despite facing many challenges, Intel's RealSense applications in fields like robotics, digital signage, and 3D scanning still saw some development. The bionic robotic dog released by Xiaomi uses RealSense depth sensing technology.

In August 2021, Intel announced it would shut down the RealSense computer vision department; by February 2024, Intel 'restarted' the application of RealSense technology in the field of humanoid robots through investments in companies like Figure AI.

Currently, the latest advancement regarding RealSense technology is the release of the 'RealSense' depth camera module D421 by Intel on September 24, 2024.

Now, does Intel want to spin off the RealSense business into an independent company? Is it focusing on the future potential of this technology in the field of intelligent robotics?

Intel denied rumors that the spin-off of the RealSense business into an independent company was due to financial difficulties, stating a very nice line - this aligns with Intel's commitment to 'incubating cutting-edge, disruptive technologies and businesses to validate customer needs and market acceptance.'