Altera and xilinx are in the same FPGA business, why do these two companies have very different strategies?
Time:2024-03-15
Views:370
On March 1, 2024, during the FPGA Vision online seminar, Intel officially announced the establishment of a new independently operated FPGA company - Altera, the appointment of Sandra Rivera as the CEO of the new company, and Altera was also officially announced as the new company brand.
9 years is not a long time, but a lot has happened in the FPGA industry: AMD‘s acquisition of Xilinx, Microchip‘s acquisition of Microsemi, Lattice‘s deep plowing of the low-power market, and the market share of local FPGA vendors has rapidly risen from less than 5% five years ago to the current 20-30%. But from the acquisition of Altera in 2015 to the announcement of the establishment of Altera in 2024, and it seems that nothing has happened.
The whole process is not complicated and the timeline is well organized:
In 2015, Intel acquired Altera, a then-aspiring independent FPGA vendor, for $16.7 billion and operated it as Intel‘s Programmable Solutions Group (PSG);
On October 4, 2023, Intel officially announced the divestiture of the PSG division, with plans to begin operating it as a standalone business as of January 1, 2024, and plans to conduct an IPO for PSG within the next 2-3 years to accelerate the growth of their business.
At present, the new company‘s core business, such as chip engineering, software engineering, architecture, product management, product marketing and other businesses have always been relatively independent, so it is relatively easy to separate from Intel, this year will continue to separate the supporting departments, including human resources, IT, legal, financial, corporate services and so on. It is expected that on January 1, 2025, Altera will be able to operate completely independently.
At the same time, Intel maintains a controlling position in Altera, but also plans to bring in investment and sell a portion of its shares. 2-3 years, when the time is ripe, the plan is to carry out an IPO.
Sandra Rivera, an engineer by trade, has led several business units at Intel over the past two decades, and prior to this new position, she was responsible for the data center and artificial intelligence businesses. When discussing why the new company is once again under the Altera brand name, Sandra Rivera responded, "Altera has an excellent reputation for FPGA innovation and leadership, and we are very excited to bring it back. I am also personally excited about the opportunity to lead Altera and continue its proud history."
The FPGA market has experienced rapid growth over the past few years. Data shows that it is expected to reach about $55 billion in the next few years, with a market size of around $9-10 billion between 2022-2023, and growing at a CAGR of 7-8%. Consider that AI is creating a huge opportunity for the FPGA market. FPGA-based solutions are expected to have an additional market growth of $3 billion by 2028.
"I think this forecast is slightly conservative given the rate of growth in innovation and accelerated computing. "Thanks to versatility and programmability, FPGAs are able to meet the needs of different markets and verticals in terms of performance, power, size and cost, making FPGAs suitable for diverse solutions," Sandra said. altera is also meeting the demanding needs of all types of markets with an ever-expanding portfolio of products that include power, performance, features, power, performance, features, price, space and energy consumption.
Founded in 1983, Altera has been committed to FPGA innovation for more than 40 years. The company broke new ground in the early 1980s with the introduction of programmable logic devices, pioneered FPGAs supporting 116Gbps transceivers, and demonstrated devices with optical interface interconnects. Over the past few years, Intel has introduced FPGA-based SmartNICs and IPUs, and combined FPGAs with IoT devices to enable end-to-end smart factories. And it has also been very productive in small chip-based design, with more than 10 million small chip products shipped.
In eight years, many aspects have changed dramatically, whether it is the competitors, the market environment, or the speed of technological innovation and progress, as well as the innovation required in the age of AI. Sandra said that the industry‘s need for more acceleration, more programmable capabilities, and more high-performance computing will drive the new company‘s development and business growth. With the split, Altera will not only be able to move forward faster, but will also be able to make more flexible and varied decisions to focus more on the FPGA portfolio rather than the broad portfolio of FPGAs offering CPU, GPU, and AI acceleration as it has in the past.
In the future, as the market will increasingly focus on die and module development, Altera wants to continue to lead the industry when it comes to UCIe (Universal Computing Engine for Intelligence) and provide customers with the modules they need when designing, building and deploying their products. In addition, advanced packaging capabilities are one of Altera‘s strengths. On top of having the flexibility of an independent company, Altera is also able to leverage the multiple capabilities in the Intel ecosystem to accelerate the launch of innovative products, and also facilitates the development of multi-chip modules in collaboration with customers who have their own IP and capabilities.
In order to achieve the above goals, Sandra Rivera said it will start from three aspects: first, to provide a first-class customer experience. As a user needs-oriented company, Altera is committed to providing customers with powerful IP, as well as service support for developers, system and solution developers. At the same time, through internal teams and manufacturing partners, it has built an agile and resilient supply chain with product distribution, community, and product delivery capabilities to better meet customer needs.
Second, a comprehensive product portfolio covering cloud, network and edge. At this stage, as the demand for computing power continues to grow across a wide range of industries, Altera‘s comprehensive product portfolio creates solutions that meet the diverse needs of customers across a range of price points, features, power and size.
Briefly, these include the cost-optimized Agilex 3, Agilex 5 for the mid-market, and Agilex 7/Agilex 9 for the high-end market.When broken down into industry segments, the Agilex 7/Agilex 9 are primarily targeted at communications service providers building network infrastructures, as well as cloud service providers building robust data centers; the Agilex 3/Agilex 5 series are more commonly used in areas such as video, visual processing, infrastructure and security equipment, and embedded IoT. video, visual processing, infrastructure and security devices and embedded IoT, the Agilex 3/Agilex 5 series is more commonly used.
Third, retaining the best talent. It was revealed that Altera employees have more than a dozen years of valuable experience, some of them from the previous Altera and Intel, but also from the outside of the "fresh blood".
"Currently, some customers are facing increasingly complex technical challenges, and we are committed to building differentiation, accelerating product implementation practices, and driving the FPGA market. We are committed to pioneering programmable solutions through a bold, flexible and customer-centric approach and delivering solutions that are AI-enabled and easy to apply across a wide range of application areas such as communications, cloud, data center, embedded, industrial, automotive, etc.", said Sandra.
It is important to emphasize that providing functions such as control, boards and system management is not all of Altera‘s goals, but also AI. It is well known that FPGAs are often used for data preprocessing to improve the efficiency of data ingestion before training models, which is a natural advantage for Altera to capture the AI market. By adding Tensor modules to the massively distributed DSP modules in the logic fabric, Altera has enhanced its support for AI functions at the hardware level; when model training is complete, it can efficiently support the deployment and execution of large-scale inference tasks.
At the software level, to allow developers to more easily integrate AI into the platform without having to learn different tool flows or different AI integration capabilities, Altera introduced a standard software environment. After introducing the AI suite to the market, developers will still be able to achieve easy deployment with the toolchain and environment they are used to, i.e. Quartus Prime.
According to Sandra Rivera, it is still in the early stages of the AI era and the AI revolution, and customers are looking for faster acceleration and stronger performance to cope with the demands created by AI models and the large amount of innovation. But the fact is that much of the business growth, especially the accelerated growth that Altera expects, will come primarily from edge-side reasoning.
At this stage, the industry is already very comfortable with applications such as computer vision, self-driving vehicles, robotics, drones, etc. Sandra Rivera points out that all of these devices are looking to add more AI capabilities, which can be accomplished quickly by simply adding AI functionality to FPGAs when there are already FPGAs in the system, without having to remove or introduce new devices and new software toolchains to enable more acceleration and data processing. This is a potential growth opportunity for FPGAs to fully meet customer demand for programmability and flexibility.
There are also short video and social media platforms, where AI is used to a very deep degree, and most Chinese companies in particular have a deep understanding of this. altera also has a number of related businesses and is committed to building intelligent, FPGA solutions with data acceleration processing and data mobility, aiming to provide these types of edge-vision solutions for the telecom and networking sectors.
After talking about Intel, the industry line of sight naturally turned to AMD. 2020 October, AMD announced its intention to acquire FPGA maker Xilinx (Xilinx) in an all-stock transaction for $35 billion, which was approved by shareholders of both parties; on Valentine‘s Day, February 14, 2022, AMD successfully completed this merger and acquisition of the largest mergers and acquisitions in the history of semiconductors. The latest financial data show that thanks to the acquisition, AMD embedded business 2023 annual revenue of $5.3 billion, an increase of 17%.
AMD‘s acquisition of Xilinx and Intel‘s acquisition of Altera seem to have similar initial intentions, mainly because "with the acceleration of global digitization, the demand for gas pedals in data centers is exploding, and a large number of FPGA accelerators have been deployed in the data centers of the global cloud service giants." This is also in line with what Victor Peng, then CEO of Xilinx, said on the 2018
First, "Intel‘s acquisition of Altera" and "AMD‘s acquisition of Xilinx" these two mergers and acquisitions are not the same at many levels. First of all, in terms of market size, sales revenue, R & D strength, compared with Altera is in a leading position, and Altera was acquired when the size and revenue compared with Intel is clearly at a disadvantage, so the relationship between Sallis and AMD is different.
Second, "Intel‘s acquisition of Altera was not particularly successful either." Because even after the acquisition, Xilin Si‘s lead over Altera is still expanding. although Intel is involved in the three areas of CPU, GPU and FPGA, but at present it seems that its market share in the field of CPU is being constantly eroded by AMD; the specific progress in the field of GPU is still to be seen, while AMD‘s market share is constantly expanding. Victor believes that when AMD and Xilinx complete the merger, will have the ability to provide a full range of solutions, the future will make more significant progress. dSqesmc
Third, future computing challenges cannot be solved by just a single architecture like CPU/GPU/FPGA/Arm. He believes that over time, people will find that adaptive computing, FPGAs, with the effect of scale and investment from the AMD combination, will innovate more and faster to better serve the market and customers.
"In my previous discussions with Lisa Su, I reiterated that Xilinx holds a very strong commitment to its existing customer markets and applications, and Lisa Su also values Xilinx‘s businesses, so the post-acquisition development is by no means limited to data centers, but also takes into account Xilinx‘s other businesses. So please be confident that despite the acquisition, customers need not worry about any changes and Xilinx will provide the same support as always." Victor said. dSqesmc
3 years, these three responses and a commitment to the implementation of whether in place, may be the industry each have their own views and comments. But overall, Victor in 2018 to formulate the three new development strategy: data center first, accelerate the development of the core market, drive flexible and adaptable computing, is carried out and implemented. Xilinx has also completed its transformation from device to platform, not only pushing adaptive computing to more users, but also playing an important role in the evolution of the disruptive theme of AI.
For the data center, AMD has previously launched many complete product portfolios including chips, SmartNIC boards, storage and Alveo compute accelerator cards, which will not be repeated here. Including automotive, test and measurement and simulation, industrial and vision, medical and scientific, aerospace, broadcast and professional audio and video and consumer electronics, including the "core market", AMD in the past year or two is also a lot of action.
Kria system level module (SOM) dSqesmc SOM is accurately defined as "located in the core of the system small size embedded PCB (including processors, DDR, peripherals)", belonging to the board level rather than chip-level design, directly into the final product can be deployed for production. Kria SOM allows developers to start at a more mature point in the design cycle than chip-down design, reducing deployment time by up to nine months.
The first products in the Kria SOM portfolio are the Kria K26, built specifically for vision AI applications in smart cities and smart factories, and the accompanying Kria KV260 vision AI starter kit in April 2021; at the Embedded Vision Summit 2022, AMD introduced the Kria™ KR260 Robotics Starter Kit; and in 2023, the Kria™ K24, aimed at motor control, was launched. motor control, the Kria™ K24 System-on-Module (SOM) and KD240 driver starter kit were introduced.
Introduced the innovative Embedded+ architecture dSqesmc This new architectural solution focuses on industrial, healthcare, smart city, automotive and other sectors, accelerating time-to-market for original design manufacturer (ODM) partners by combining AMD Ryzen™ (Riptide) embedded processors and AMD Versal™ Adaptive SoCs on a single, integrated board, offering The combination of the AMD Versal™ embedded processor and AMD Versal™ adaptive SoC on a single integrated board provides ODM partners with a scalable and energy-efficient solution to accelerate time-to-market, and is "a great example of the synergies between the two companies." dSqesmc
Going Down to the Cost-Sensitive Edge
Spartan™ UltraSc
131-2899-0370
2881438908
xilinx@casi-ic.com
Elan Tsao