This article is more than 1 year old
Too little, too late: Intel's legacy is eroding
Can Pat stop the rot – or is the chip giant facing a Kodak moment?
Comment Intel CEO Pat Gelsinger's carefully assembled house of cards is collapsing around him. And it's not really that surprising when you look at the hand he's been dealt.
For those that haven't been following Intel product roadmap that closely, here's a quick recap of where things stand today. The company is stuck on an aging 10nm process; its desktop CPUs are ludicrously power hungry; its dedicated GPUs aren't particularly competitive, even when their drivers do work; its upcoming processor families are hopelessly behind schedule; and it just killed off Optane, which was arguably its most promising development in recent memory.
Of course, none of that is Gelsinger's fault. All of these projects were well underway long before he agreed to return to Intel early last year. Even the decision to shelve Optane had more to do with Micron pulling out of the arrangement and a lack of demand than anything.
But the fact remains that these are his problems now, and they appear to be catching up with him. Making matters worse, investors are undoubtedly losing patience with the x86 giant after the company's disastrous second quarter saw Intel record a $454 million net loss for the three-month period.
In response to the dismal earnings and even more pessimistic outlook, Gelsinger offered this statement: "This quarter's results were below the standards we have set for the company and our shareholders," he said. "We must and will do better."
But how does a company like Intel do better when it faces declining demand across nearly every business unit and the strongest competition from rival chip giants in more than a decade?
Perhaps the more important question for Gelsinger is how do you instill confidence in a corporation that can't ship a product on time to save its life?
Intel is falling behind
Intel is a massive organization filled with brilliant scientists, electrical engineers, and software developers. And yet, despite all of this talent, the company is falling behind rival AMD and faces a mounting threat from Arm processor competitors, which have gained a foothold in the lucrative cloud segment in recent years.
Intel's Sapphire Rapids Xeon Scalable processors are without a doubt its most ambitious chips ever. In a single generation, Intel committed to not only delivering DDR5, PCIe 5.0, and CXL to the datacenter market, but that it would employ a multi-die chiplet architecture that would close the core-count gap with AMD.
Based on the performance we've seen from Ice Lake, the chip would have been a formidable challenger to AMD's Milan-X had it shipped on time.
However, it appears that trying to perfect what took AMD years to get right wasn't as easy as Intel thought. And beginning in mid 2021, the company began pushing the launch back first to Q1 2022, then to 1H 2022, then to 2H. And this week, it was revealed in NDAed documents obtained by Igor's Lab that the chip may not be ready until Q1 2023.
What's more, the documents indicate that Intel has struggled to make the thing work, with the yet-to-ship chip said to be on its 12th stepping – twelfth – something I'm assured is rather atypical for a business at Intel's level. That means the server processor has gone through about a dozen revisions already, from A0 to E5, and still hasn't launched.
Given what's at stake, it's understandable why Intel would go to such lengths to make it work right.
However, it's hard to ignore Sapphire Rapid's absence when companies like AMD have been pushing ahead unabated. The House of Zen has enjoyed bumper Epyc server processor sales amid the absence of Sapphire Rapids, and is slated to release its Ryzen 7000-series desktop, fourth-gen Epyc datacenter CPUs, and RDNA-3 GPUs this year.
If the reports of Sapphire Rapids latest delays are true, that's going to create an even more unfavorable comparison, especially among customers that have put off their server refreshes in anticipation of DDR5 and PCIe 5.0-compatible systems.
But it's not just Sapphire Rapids that Intel appears to be struggling with. A recent TrendForce report indicates the GPU tile destined for its upcoming Meteor Lake CPUs may suffer more delays.
- Intel's first discrete GPUs won't be a home run
- AMD boasts of record sales, says 5nm Zen 4 Ryzen 7000 coming this quarter
- Ampere: Cloud biz buy-ins prove our Arm server CPUs are the real deal
- Why Intel killed its Optane memory business
The firm reports that the GPU tile has already been pushed back due to "process and design verification issues," but the reasons for the latest delay remain unknown.
Speaking of long delayed products that have only recently begun shipping, Intel's Arc dedicated GPUs recently became a point of controversy over the chipmaker's insistence on shipping them with unfinished and borderline non-functional software and drivers.
It doesn't help that the cards face much the same fate as Sapphire Rapids from a competitive standpoint, with gaming stalwarts AMD and Nvidia expected to release their latest generation graphics cards this fall, and Intel is entering with a first-gen product.
Again, Intel is faced with a rather unfavorable comparison. Benchmarks indicate middling performance, putting them on par with now two-year-old low-to-mid-tier cards from Nvidia and AMD. This isn't surprising — or even all that damning — for Intel's first meaningful attempt to break into the GPU market. And we know that plenty of players don't need the latest and greatest silicon to enjoy their games; low and mid tier is within their budget and will keep them happy. But it arguably underscores Intel's failure to execute.
Had its Arc GPUs shipped even a year ago, amid the worst of the chip shortage and booming crypto-mining craze, the comparison would have been much different. All Intel would have needed to do then was have a reasonably priced, lower-end GPU from its factories that people could actually buy. But Intel, as usual, missed its moment.
Intel 4 can't afford another delay
Intel's GPUs have at least one advantage over the rest of its lineup. Intel only has to worry about designing the chips and writing the software; the actual manufacturing is in TSMC's hands.
That's not the case for the rest of its product lineup, which is stuck on a 10nm process that's looking more and more dated by the quarter, and Intel's cheeky move to rebrand the process as Intel 7 isn't fooling anyone.
Gelsinger's ascension to CEO can be attributed in large part to Intel's failure to bring a working 7nm process — now dubbed Intel 4 — to market on time. And in the 1.5 year since taking over, Gelsinger has managed to shift the conversation away from its long-delay process.
Intel can claim up and down that Intel 4 is on track to ship in 2023 with Meteor Lake, but the company's track record suggests that's wishful thinking.
Gelsinger hasn't made things any easier for himself in this regard. Opening Intel's fabs to contract manufacturing was a bold move, but who is going to buy your chips if your process tech never arrives on time? If Intel wants to steal share from TSMC for leading edge chip manufacturing, it needs a competitive process. As it stands, it doesn't look like it has one.
Intel must break the pattern
To put it bluntly, Intel has managed to squander its lead over rivals and its massive reserves of talent on over-ambitious projects that have time and again arrived late to the party.
And yet, this is a company that thinks it's worthy of the lion's share of the $52 billion in factory subsidies passed by Congress last month.
If Intel is to succeed, it must break this pattern of delivering too little too late, and that means developing competitive silicon on time for a change, even if it means forgoing features of debatable value or relevance. ®