Intel Roadmap Update: Alder Lake In H2’21, Ice Lake-SP Late This Yearby Ryan Smith on July 23, 2020 11:15 PM EST
- Posted in
- Ice Lake
- Sapphire Rapids
- Alder Lake
Among several different updates tucked into Intel’s Q2’2020 earnings report, the company included a brief update on some of their future products. While the bulk of the company’s focus is currently on their next-generation Tiger Lake CPUs, which are launching this quarter, the company is also looking at what comes after Tiger Lake, as well as the future of their highly profitable server business.
First off, Alder Lake has finally been formally outed. The successor to Tiger Lake now has an official launch window of the second-half of 2021. The 10nm chip will be for both mobile and desktops, making it the first 10nm chip that Intel has confirmed will come to desktops. Very little is otherwise officially known about the chip, but Intel’s ISA documents have previously revealed that there will be some new instructions found in that chip.
Otherwise the six-month window for kicking off production shipments is a fairly wide one for a chip that doesn’t rely on a new process node. Intel product cycles are rarely under a year long, so at first blush we’d be surprised if this was anything earlier than a late 2021 product. But with Intel’s recent 7nm delay and planned ramp-up of their 10nm process, it may be that Intel will be trying to pull it in and launch it in Q3, similar to this year’s Tiger Lake launch.
Meanwhile on the server side of matters, Intel is preparing for both Ice Lake-SP as well as its successor, Sapphire Rapids. One of the many victims of Intel’s 10nm woes, Ice Lake-SP is Intel’s first 10nm server chip. As of late the company has been riding a wave of profitability based on its server parts, so a newer part that improves on core counts and energy efficiency will be a welcomed addition to Intel’s product lineup, not to mention better able to fend off AMD’s powerful EPYC “Rome” processors.
Initial production shipments for Ice Lake-SP are set to start by the end of this year. Though Intel’s language is loose enough that this may mean that larger volumes of the chip may not ship until 2021.
Following Ice Lake-SP will be Sapphire Rapids, Intel’s second-generation 10nm server part. Along with getting Intel’s product release cadence closer to being back on track, Sapphire Rapids will play an important role in unifying Intel’s split Xeon families. Intel’s oddball 14nm Cooper Lake Xeons, which are currently shipping, support bfloat16, but Ice Lake-SP will not. For 10nm chips that support is finally being rolled into Sapphire Rapids, making the new chip the successor to both Cooper Lake and Ice Lake-SP in every way.
Sapphire Rapids will follow Ice Lake-SP by roughly a year. According to Intel’s presentation deck, chips will begin sampling in H2’2020, while CEO Bob Swan’s prepared remarks state that initial production shipments will begin at that time.
What follows these chips, in turn, will be the big question that Intel is currently wangling with in light of their 7nm delay. The company has made it clear that they intend to maintain an annual release cadence, divorced from their process roadmap if necessary. Depending on the state of their 7nm process, that may mean 7nm chips, 10nm chips, chips using dies from both processes, or even using dies from third-party fabs. Intel has opened the door to all possibilities, and their 2022 chips will likely be their first chance to embrace their new pragmatic approach.
Post Your CommentPlease log in or sign up to comment.
View All Comments
azfacea - Friday, July 24, 2020 - linkGive an engineering company to a marketing department and what did you expect the outcome to be ? surprise surprise Bob Swan is not Elon Musk. but its not just Bob Swan, intel has been run by the same corporate sales force mentality that has destroyed orcale. but thats just my opinion. I am sure someone thinks Oracle is never been better
azfacea - Friday, July 24, 2020 - linksteve jobs on why companies fail:
SystemsBuilder - Friday, July 24, 2020 - linkNo you are right about your Oracle comparison and analysis! The fundamental problem is that Intel, a tech company, is currently run by MBAs who do not have even the most basic understanding of what matters in this bleeding edge, high capital investment industry: Process excellence is what it is all about! Without that it does not matter how well you can design a CPU etc.
For a MBA it is simply too attractive to make short term profit off of the 14nm process and leverage the market lock in position (that Intel has or is looking slowly slip away from) as long as possible. To make things worse executive compensation equity plans reinforce this behavior, driving them to maximize net profit short term. In this business, short term is a few years and we are now finally coming to the end of this "milking 14nm+++++" strategy.
In short: you need an engineer at the top of a high tech company who understands deeply what Intel needs and what strategically drives intel’s long term success, who does the right thing even if it hurts a quarter or 2 and invest in the long term.
Andy Grove famously said: “Only the paranoid survives” and for the last few Intel has just not been paranoid enough but has been dominated by executives who are blinded by next few quarters of profit and corresponding executive pay.
In the end it is of course the board’s fault or rather the long term strong investors fault (who vote for who should be on the board). They should have fixed the executive level before it got to this situation.
As a parallel Microsoft went through roughly a decade or so with a sales guy at the top (Steve Balmer) and flat share price as a consequence. So Bill G. was convinced that the “next CEO of Microsoft shall be an engineer” so when the time comes, sure enough, in comes Satya N. (a distinguished engineer at MS). He does what needs to be done, stock price takes off and the rest is history.
Intel shareholders take note and act.
AntonErtl - Friday, July 24, 2020 - linkOn the contrary, since roughly the 90nm days, the process has become less and less important, and that's why Intel has been able to compete despite their 10nm and 7nm woes (compare that to the late 1990s; what if Intel's 250nm (Katmai) and 180nm (Coppermine) process had been delayed for 3+ years, and they would have been forced to compete with Klamath (at up to 300MHz) against Thunderbird (at up to 1400MHz)?). However, the fact that Intel tightly couples their microarchitecture implementations to processes means that process delays also delay their microarchitecture innovations, and then, yes, designing better microarchitectures does not help them; OTOH, this tight coupling probably gives them a clock speed advantage (compare Zen 1 clock speed to Skylake++ clock speed in 2017, both on 14nm).
edzieba - Friday, July 24, 2020 - linkProcess became the primary driver of performance with 22nm. Past there, cost/transistor started rising, area scaling started slowing dramatically, power density (capping peak performance) started increasing dramatically, the gate oxide thickness limit meant voltage hit 1v and stopped dropping, and practical frequency limits capped at 5GHz (either overclocked, or now mainstream parts). With the exception of situations where perf/watt is paramount (mobile and laptop), the larger the process you can get away with the better.
Quantumz0d - Friday, July 24, 2020 - linkNadella couldn't do shit to the Windows Phone he fucked it up even worse by making the Chevron WinPhone hack that could make a lot like Android type freedom during the old days a big edge over the competition and the store for applications. He was yapping "mobile first", and MS nuked it by C&D and closing all of it beleving in their utopia like Nokia's management which was not beleiving in their engg of Maemo but putting all eggs into that sellout trojan horse Elop.
And later he copied that Adobe's As a Service and stamped onto Windows with firing the Chief of Windows and dissolving that OS division, they make money because he also copied Amazon by Azure. He is just a puppet for the BOD. They have that monopoly grip due to the x86 and Wintel dominance as no other OS could be put into hands of the people thus they have that high stock boost, Win10 is pure cancer, every Tuesday there's a patch and they experiment on Home and Pro for Enterprise quality updates, they are forcing heavy telemetry into the OS and 6 month feature releases with insane bugs.
Intel on the other hand ruined by BK and the whole corporate structure internally, they also started to pander woke political trash because, California, that degenerate place can never forget how awful it is, with radical extremist laws now like mandated Senate Bill 826 and other BS nonsense. Firing BK not because of his failure but MeToo was horrible card. And now made a bean counter as CEO to appease investors. This company also lost Jim Keller perhaps too soon, since he does leave the companies once he is done but with Intel it was too soon esp given how excl. AT's Ian interviews were glowingly positive.
Intel has to get their shit right by hiring proper people who are fit for the goddamned job not a quota and have a long term plan, they have money but money cannot put R&D and success it's a hardwork and patience path not virtue signalling trash like how this weak CEO was doing after what is happening in US.
jospoortvliet - Monday, July 27, 2020 - linkyour post is as badly written as it is wrong... I don't like Microsoft one bit but they are making the right moves for their shareholders, that's for sure.
SystemsBuilder - Monday, July 27, 2020 - linkThere are so many errors in what you say about MS that it's hard go through them all except to say you are wrong.
For example: Steve B. was responsible for the phone fiasco 100%. One of Satya's first action after becoming CEO was to write off the whole catastrophic Nokia acquisition (I think the write off was $10B). Steve had persisted with his phone strategy for years even when the whole phone division was clearly sinking (huge loss leader for years). AND indeed the Nokia acquisition is why Steve B. had to go in the end. Bill G. made sure it was a graceful exit but nonetheless Nokia was the last straw.
Azure is a much older that you may think. MS started to develop it more than 10 years ago. If anything there is significant migration of employees (also executives) from MS to Amazon that took lot of that thinking with them… so it’s wrong that MS somehow copied AWS. If anything MS understood cloud early but because of the misguided priorities they did not give it enough focus (distractions like Nokia, aQuantive acquisition other wacko initiatives that have the organization flip flopping chasing the latest shiny thing - I’ve seen it first hand) until Satya took over and refocused the company on cloud. I could go on for hours. But the best proof you are wrong is just to look at MS share price comparing the 14 years with Steve B. as CEO (share price 2000: $40 -> 48 when he retired) with the years with Satya ($48 - > $200). It’s pretty stunning. Don’t’ know enough about the internals of Intel to comment on that. But I do know MS internals!
PaulHoule - Friday, July 31, 2020 - linkI used to drive my car on Nokia snow tires. Then around the time they split out the unprofitable parts of Nokia (cell phones) from the profitable parts (everything else) they started making the tires in Russia and I switched to Blizzaks.
peevee - Friday, July 24, 2020 - linkCost/transistor was the primary driver before multiple patterning had to be engaged.
But for a few years now power efficiency (in executing operations in a loop) is EVERYTHING, including even desktops. It is the BY FAR the most limiting factor both on supply side, and on cooling side. And shorter signal lines mean not just shorter local latencies, but less power dissipation everything else being equal.
That is why the physical memory architecture with DIMMS being in a separate place on a motherboard should have long been abandoned. At the very least, switch to SODIMM slots on the CPU package itself (4 sides, 4 slots), radically simplifying MB design and reducing latencies while allowing lower signal voltages.