Intel aims for supply chain edge with new data center chip

(Reuters) – Intel Corp on Tuesday introduced its newest flagship data center microprocessor, hoping its in-house manufacturing operations will help it navigate a chip shortage to better compete against rival Advanced Micro Devices Inc’s (AMD) faster chips.

The new “Ice Lake” chip is designed to be used by cloud computing providers and others who run massive data centers, and Intel said it has already shipped about 200,000 test units.

Intel’s new chip will be slower than AMD’s new flagship data center server chip released last month. Intel years ago lost its speed lead because of fumbles with the in-house 10-nanometer manufacturing technology it uses to make the chip, while AMD outsources its server chip manufacturing to Taiwan Semiconductor Manufacturing Co Ltd and has gained market share against Intel.

But Intel is hoping that greater control over its supply chain – in addition to a portfolio of data storage, networking and other chips designed to work well with the “Ice Lake” processor – will help make the chips competitive during the midst of a global chip supply crunch, in part by mitigating customer concerns about shortages.

“Nobody else in the industry has the intersection of (intellectual property), architecture, design and manufacturing. We think that’s a particularly important differentiator for the company at a time when demand is exploding and supply is short,” Navin Shenoy, executive vice president and general manager of the data platforms group at Intel, told Reuters.

To fight the loss of its speed title, Intel has focused on tailoring its chips and systems to optimize them for specific kinds of computing. One area of focus has been 5G networks, which are shifting from being built on proprietary systems such as those from Huawei Technologies Co Ltd toward using commodity parts like Intel chips. Shenoy said Verizon Communications Inc plans to use the new “Ice Lake” chips.

Verizon will appear at an event showcasing the chips, alongside Microsoft Corp, Oracle Corp and others.

Intel’s 5G push “gives economies that are worried about security issues or geopolitical tensions a lot more flexibility. They can lean on Western supply chains to build out their next-generation 5G infrastructure,” Shenoy said.

Intel to spend $20 billion on U.S. chip plants as CEO challenges Asia dominance

(Reuters) – The move by CEO Pat Gelsinger on Tuesday aims to restore Intel’s reputation after manufacturing stumbles sent shares plunging last year. The strategy will directly challenge the two other companies in the world that can make the most advanced chips, Taiwan’s Semiconductor Manufacturing Co Ltd (TSMC) and Korea’s Samsung Electronics Co Ltd.

And it will aim to tilt a technological balance of power back to the United States and Europe as government leaders on both continents have become concerned about the risks of a concentration of chipmaking in Taiwan given tensions with China.

Intel shares rose 7.5% after the company disclosed its new strategy and full-year financial guidance for 2021. Some investors such as Third Point LLC had previously urged Intel to consider spinning off its costly chip manufacturing operations.

Intel said it expects $72 billion in revenue and adjusted earnings per share of $4.55, compared with analyst estimates of $72.9 billion and $4.77 per share, according to Refinitiv data. The company said it expects to spend $19 billion to $20 billion on capital expenditures.

Gelsinger said that 2021 forecast “reflects the industry-wide shortage” of some components such as substrates.

Intel is one of the few remaining semiconductor companies that both designs and manufactures its own chips. Rival chip designers such as Qualcomm Inc and Apple Inc rely on contract manufacturers.

In an interview with Reuters, Gelsinger said Intel has “fully resolved” its problems with its most recent manufacturing technology and is “all systems go” on chips for 2023. It now plans a massive manufacturing expansion.

That will include spending $20 billion on two new factories at an existing campus in Chandler, Arizona, that will create 3,000 permanent jobs. Intel will then work on future sites in the United States and in Europe, Gelsinger said.

Intel will use those factories to make its own chips but also open them to outside customers in what is called a “foundry” business model in the chip industry. Gelsinger said the new factories will focus on cutting-edge computing chip manufacturing, rather than the older or specialty technologies that some manufacturers such as GlobalFoundries specialize in.

“We are absolutely committed to leading process technology capabilities at scale for the industry, and for our customers,” Gelsinger said, adding that Intel has lined up customers for the new factories but could not disclose their names.

He did say on a webcast Tuesday that Amazon.com Inc, Cisco Systems Inc, Qualcomm Inc and Microsoft Corp support its efforts to offer chip manufacturing services. On a conference call, Gelsinger said that Intel “will pursue customers like Apple.”

The move is a direct challenge to TSMC and Samsung. The two have come to dominate semiconductor manufacturing business, moving its center of gravity from the United States, where much of the technology was once invented, to Asia, where more than two-thirds of advanced chips are now manufactured.

“Intel’s investment will help to preserve U.S. technology innovation and leadership, strengthen U.S. economic and national security, and protect and grow thousands of high-tech, high-wage American jobs,” U.S. Secretary of Commerce Gina Raimondo said in a statement.

Gelsinger said Intel will aim to change the global chip manufacturing balance by embracing the foundry business where it historically has been a minor player. Intel will offer chip customers the ability to license out its own technological crown jewels – known as x86 computing cores – as well as offer to build chips based on technology from Arm Ltd and RISC-V technology from startup SiFive.

“We will be picking our next sites within the next year for U.S. and Europe,” he said.

The American sites could benefit from a $30 billion subsidy package that lawmakers hope to bring to the floor of the U.S. Senate next month. The bill remains largely unwritten, and Gelsinger said on a conference call that Intel’s plan “does not depend on a penny of government support. It is the right strategy for us going forward.”

Intel also announced plans for new research collaboration with IBM focused on computing chip and packaging technology.

But even as Intel jumps into competition with TSMC and Samsung, it also plans to become a larger customer of theirs by turning to them to make subcomponents of its chips called “tiles” to make some chips more cost-effectively.

“I’ll pick the best process technologies wherever they exist,” Gelsinger said. “I leverage internal and external supply chains. I’ll have the best cost structure. That combination of supply, products and costs, we think is a killer combination.”

Intel has given few details of exactly how it will use outside factories, but analyst Patrick Moorhead of Moor Insights and Strategy said he expects Intel to use them as “gap fillers for some of the highest performance” chip parts until Intel can regain a manufacturing lead over its rivals.

Intel buys time with ‘retrofit’ Rocket Lake desktop PC chips

Intel has launched its latest desktop PC chips having had to retrofit some of its recent semiconductor designs to work on older transistor tech to deliver the processing power required.

The stopgap effort has consequences for both the speeds and heat they produce.

The US firm claims the Rocket Lake chips deliver a 19% headline-rate gain over the last generation, and introduce features that will help PCs keep pace with the latest gaming consoles.

Intel continues to dominate the sector.

But rivals have benefited from outsourcing production, leading to suggestions that Intel’s position as the leading CPU (central processing unit) provider is not as secure as it might appear.

Intel still leads the pack. CPU chips (Oct-to-Dec 2020).  .

Tiny switches

Transistors are basically tiny on-off switches, and billions of them are arranged in different patterns to carry out calculations on a chip.

The benefit of making them smaller is that more can be packed into the same space – allowing computers to run more quickly while potentially using less power.

Intel’s new Rocket Lake chips rely on 14 nanometre transistors, and are made within its own fabrication plants.

By contrast, its chief competitor AMD uses a contract manufacturer – Taiwan’s TSMC – to build its latest Ryzen desktop PC chips, which benefit from smaller 7nm transistors.

Intel manufacturing
image captionIntel is still refining its manufacturing processes to reliably build powerful desktop chips that use 10nm and 7nm transistors

And Apple is in the process of weaning itself off Intel to use its own designs, also produced by TSMC but using its even more advanced 5nm tech.

Fab-less pressure

There is no set way to measure transistor sizes, and Intel claims its 14nm tech equates to TSMC’s 10nm. Even so, the US firm acknowledges it is running behind. 

It had originally intended to make the transition to 10nm desktop chips between 2017 and 2019. As it is, this will not happen to a future generation launched in late 2021-2022.

And Intel is also experiencing delays in making the next leap forward to 7nm.

However, there are advantages to keeping production in-house.

It helps keep costs down.

And it also means it avoids feeding into wider concerns that the US has become over-reliant on overseas chip-producers.

Intel recently appointed a new chief executive – Pat Gelsinger – who has made it clear he intends to resist pressure from some investors to become a “fab-less” firm, a term used to refer to chip designers who do not operate fabrication plants of their own.

Pat Gelsinger
image captionChief executive Pat Gelsinger aims to reinvigorate Intel following the ousting of his predecessor Bob Swan

“The factory is the power and soul of an enterprise, and we must become even better in the future,” said Mr Gelsinger in January.

Faster frame rates

The new “11th generation” desktop chips take the microarchitecture for their CPU cores from one set of 10nm laptop chips – 2019’s Ice Lake series – and their graphics architecture from another – 2020’s Tiger Lake family.

Intel has described the process of reworking these designs for 14nm transistors as “backporting”.

Computer chip diagram
image captionThe new chips feature up to eight cores to divide up computational tasks

“We’ve been making 14nm CPUs for a long time, and part of the benefit of that is it is a very established manufacturing process to the point where we know it inside and out,” spokesman Mark Walton explained.

“So we really know how to ramp up the clock speeds – and for a gaming product, that’s really, really important.”

The firm’s own benchmarks indicate its new i9-11900K chip will deliver a boost of 14% more frames-per-second when playing Microsoft Flight Simulator over the last-generation i9-10900K, when set at high quality graphics, for example.

And Intel is also playing up other benefits, including support for PCie 4.0, which increases the bandwidth available to third-party components such as add-on graphics cards and solid state drives, effectively allowing them to shunt data about more quickly.

“You will have much faster loading times, textures will load more quickly in games, and you get a much more seamless experience,” said Mr Walton, suggesting this would counter one of the key advantages the Xbox Series X and PlayStation 5 enjoyed.

Testing hot

Intel will be marketing the new chips as offering a 19% improvement in “instructions per cycle” over their predecessors.

But the site Anandtech said it had only noticedmodest gains when testing some of the chips with games of its own choice, and in some cases said the differences were imperceptible.

“It trails behind its competitor AMD at times by a significant margin,” the site’s Dr Ian Cutress told the BBC.

AMD Chip
image captionAMD is also targeting its latest range of desktop chips at gamers

Moreover, the review also highlighted that the chips could run very hot.

Intel has removed a very thin layer of material between the block of semiconducting material on which the CPU’s circuits are fabricated and the “heat spreader” on top to help keep the components cool.

But Anandtech recorded peak temperatures of 104C (219F) in one of its tests, which is only just in the safety zone.

“Intel’s laptop chips were designed specifically for the characteristics of 10nm, and that includes thermal performance and frequency, and voltage,” explained Dr Cutress.

“Because they’ve done a retrofit, compromises have had to be made – they still get the performance gains, but it’s at the expense of needing lots of power and potential heat issues.”

Intel chip
image captionIntel has used a process called “dye thinning” to remove a layer of material from the chip to help keep it cool

In real-world terms, that means if users give the chip taxing tasks and do not have proper cooling, the chip may throttle its own performance to avoid damage.

One added consequence is that Intel is only offering Rocket Lake chips with a maximum of eight cores – the more cores you have the faster a program can run if optimised to share the load between them.

By contrast, AMD’s Ryzen chips go up to 16.

However, AMD’s newest processors remain in short supply. 

And one advantage Intel has as its own manufacturer, is that it can relatively easily adjust production to match demand.

By contrast, AMD must vie with Apple, Nvidia and others for TSMC’s capacity.

By Leo Kelion
Technology desk editor