Intel: Attacked and countered in many ways

0

At first glance, Intel doesn’t look like a beleaguered company. Last year, sales were $ 77.8 billion and profits were $ 20 billion. The market capitalization is $ 220 billion as of mid-September 2021.

Still. If you are a leader, all of your competition is on you. Intel is working to lose the lead in manufacturing and process nodes, to lose market share to the resurrected AMD, Nvidia relentlessly challenges Intel for AI advantage, Atom processor arm in the mobile market three years later CEO who failed brilliantly and the third became.

But Intel wins the competition. Lisa Spellman, Corporate Vice President and General Manager of the Xeon and Memory Group, said: “So it’s no wonder that there are competitors who want something from it.”

“I’m not saying Intel is ‘under siege’, but it is certainly being attacked from various angles, including internal unrest. Glenn O’Donnell, vice president and director of research, Forrester Research said:

Other factors are difficult to fix.

As AMD gains momentum, Nvidia GPUs collapse, ARM architecture is now a great alternative, and countless other chips like communication interfaces, microcontrollers, and DSPs are all on the battlefield, the competition is unprecedented. It is not an exaggeration to say that it is going to be violent.

Intel got the much-needed shots with Gelsinger’s return to CEO in February. Gelsinger has long been considered one of the most important engineers in the company’s history and was considered a top candidate for the top spot when he was in exile in 2009.

Upon returning to Japan, Gelsinger started running hard and took a number of steps to mend Intel’s lost luster.

“I think he’s doing the right thing. And you know he’s a new guy. He is licensed to take brave steps, and they have been brave. IDC Computing and Shane Lau, Research Vice President of Semiconductors, said:

Some of Gelsinger’s advocacy projects actually began under his predecessor, Bob Swan. Bob Swan couldn’t get the recognition he deserved as he was the company’s former CFO and was more wealthy than an engineer. “Bob survived many difficult times,” said Spellman.

Gelsinger has gained extensive experience since leaving Intel in 2009 and now supports him in changing company guidelines. He was COO of EMC, took over as CEO of VMware, tripled sales and expanded well beyond core hypervisor technology. “It not only gave him the opportunity to run a large technology company, it gave him a non-Intel perspective,” said O’Donnell. This will help the gelsinger figure out what to do. “I think it was very valuable.”

IDM 2.0 shakes production

In 2008, AMD split into two companies, one developing chips and the other called GlobalFoundries, which made them. Fought to compete with Intel and could not afford to have manufacturing facilities in New York and Dresden, Germany, the company sold the fabs to two Abu Dhabi government mutual funds. The concept of a chip company called “Fabless” that outsources manufacturing to third parties just by developing the chip became popular, and Nvidia proved that the concept worked.

Wall Street and other analysts have started whispering that Intel should probably do the same thing since Intel was behind manufacturing. Instead, Intel introduced a strategy called Integrated Device Management 2.0 or IDM 2.0. First, IDM 2.0 includes a $ 20 billion investment in two new chip fabs at its Chandler, Arizona facility. In addition, IDM consists of three components:

  • Intel’s global internal factory network. Intel has fabless in the US, Europe, Central America and Asia, and IDM 2.0 promises to develop its own chips to avoid fabless.
  • Extended use of third-party foundry capacities. Intel plans to manufacture some chips in collaboration with other foundries, including Taiwan’s giant TSMC.
  • Build a world-class foundry company called Intel Foundry Services. Intel will make chips for other companies, including competing companies. This is new for Intel and already has two customers side by side. Both are competitors. Amazon Web Services and Qualcomm. AWS makes its own arm-based server processor called Graviton, but Qualcomm is competing for 5G chip space.

Spellman said Intel had previously manufactured for other parties. IDM 2.0 makes it more strategic to discuss products, solutions and ways to develop solutions. “

IDM 2.0 also includes additional technology elements such as management software and foundry business, said Rau. In his opinion, these are likely to be the most difficult components of the IDM 2.0 strategy.

“There is a need for leadership and manufacturing across the industry,” he said. “The connection has to be made between Intel’s foundry company and its customers, not Intel itself, which means that it has different capabilities, different IP portfolios, different designs and thus foundries. Can enable customer success. “

Spelman explains that IDM is more than just a business alliance between chip manufacturers, it benefits end users as well. “Customers who offer Infrastructure as a Service may have their own IP, their own customizations, or simply control other data flows that are best for their environment,” she says. This potentially means lower latency, more computational density that can be provided, and greater sustainability of computational performance. All of this is beneficial to the end user.

More architectures

Until the last decade, Intel was the only x86 show when it comes to computers. I haven’t used x86 for communications, networking chips, etc, but I’ve used it elsewhere (Itanium Processor of Destiny Intolerable). The Atom was an attempt by a mobile processor that essentially removed the x86 core. The first two GPU attempts, Larrabee and Xeon Phi, were x86-based. They all failed.

Then the current changed. In 2014, Intel acquired the FPGA vendor Altera, the AI ​​chip manufacturers Nervana and Habana, and the network experts Barefoot Networks, and invited the best AMD GPU experts to finally build the right GPU.

The multi-architecture approach is not just attributable to Intel. AMD takes over the FPGA manufacturer Xilinx. Nvidia has taken over network specialist Mellanox and is about to take over Arm Holdings. Marvell has also moved from making hard drive controllers to arm server processors and intelligent network adapters. Nobody offers a single architecture anymore.

“It’s a religious debate for a company like Intel, but as I said, Gelsinger has to eat his youth so to speak in order to steer Intel in a different direction. It doesn’t mean getting away from x86, but you should consider x86plus, ”says O’Donnell.

Recently, Intel added a new RISC architecture to its family. In June, we plan to sign a contract with RISC-V manufacturer SiFive to develop a new development platform called Horse Creek using SiFive’s new Performance P550 core. Intel has also registered SiFive with the Foundry Service. That same month, Bloomberg reported that Intel was trying to buy SiFive for $ 2 billion. However, Intel remained silent on the matter and refused to comment except to confirm its partnership with SiFive.

According to O’Donnell, the mobile computing ship has set sail, but Intel still has great opportunities in edge computing with RISC-based chips. “One of the biggest criticisms of x86 is its power consumption,” he said. “But IoT devices and edge computing devices will be huge. It’s a huge growth market. And power consumption is important there. And now it is everyone’s market to be won. “

Introduction of IPU

In addition to the SmartNICs already on offer, Intel has also introduced a new line of intelligent network controllers called Infrastructure Processing Units (IPUs). The idea behind SmartNIC is to take the burden off the CPU from routing network traffic and free up the CPU to run jobs that process data. Unlike traditional network controllers, SmartNICs have some programming power to perform tasks that traditional Dam NICs cannot, such as packet processing...

IPU is a Smart NIC that takes you to the next step. In the past, SmartNICs were local, but IPUs were designed specifically for data movement in the cloud and for communication services. The IPU was developed to separate the cloud infrastructure from the tenant or guest software so that guests have full control over the CPU with the software and service providers to maintain control over the infrastructure and trust paths.

The two IPUs have Intel Xeon-D and Agilex FPGA cores for processing. The third code name, MountEvans, is the first type of Intel. Mount Evans was developed in cooperation with leading cloud service partners and is based on Intel’s packet processing engine and has up to 16 ARM Neoverse N1 cores with a dedicated compute cache and up to 3 memory channels. The ASIC can support up to four hosts, Xeon, with a full dual bandwidth of 200 Gb / s in between.

XPU and oneAPI connect them all

Each processor architecture has its strengths and weaknesses, all of which are suitable for a specific application. The Intel XPU project announced last year aims to provide a programming model that is integrated into all types of processor architectures and to adapt all applications to the optimal architecture. XPU lets you integrate x86, FPGAs, AI, machine language processors, and GPUs into your network, which means your app will be compiled with the best processor for your task.

This is done via a oneAPI project that works with XPU. XPU is the silicon part and oneAPI is the software that connects everything. oneAPI is a heterogeneous programming model that uses common languages ​​such as C, C ++, Fortran and Python and uses code written in standards such as MPI and OpenMP.

The oneAPI Base Toolkit includes compilers, performance libraries, analysis and debugging tools for general purpose computing, HPC and AI. It also provides compatibility tools to migrate code written in Nvidia’s CUDA to Intel’s GPU language Data Parallel C ++ (DPC ++).

“This is a branded version of the trend that is well established with semiconductor manufacturers,” says Rau. “That means you can no longer be CPU-centric or GPU-centric. You can’t have one big chip in your portfolio. And to be successful, you can connect multiple types of processors. Requires processing and connection. “

Spelman has confirmed that the jury is still in development, so oneAPI is still being examined.

And she says never say anything about the ability to add architecture. “We never say it ended there. The industry is moving too fast and too dynamic, so we don’t know where to invest next, ”she said.

Join the Network World community on Facebook when LinkedIn is commenting on topics that come to mind.

Copyright © 2021 IDG Communications, Inc.


Source link

Share.

Comments are closed.