May 9, 2019
Joel Hruska

Intel Will Launch 7nm Chips in 2021, Ice Lake Ships in June

For over a year, discussions around Intel and its semiconductor performance have revolved around ongoing delays to the company’s 10nm process. Intel’s 10nm node has been repeatedly pushed back, from an original expected release date of 2015, all the way to the 2019 holiday season. Intel, having failed to deliver on repeated timelines for 10nm introduction, has committed itself repeatedly to having hardware in-market by this date. Now, Intel has gone further, committing itself to have 7nm hardware in-market by 2021.

As far as the 10nm commitments are concerned, I expect Intel will make this deadline. 10nm chips like the Core i3-8121 are already in-market and Intel has far too much riding on that “Holidays 2019” date to miss it at this point. We don’t know how wide the 10nm release will be (a recent leak implied 10nm chips would be confined to mobile, with no desktop chips released on that node through 2020), but I think Intel will have 10nm chips meaningfully for sale in at least some segments by this holiday season.

The company announced yesterday it would begin shipments of Ice Lake in June, promising “approximately 3 times faster wireless speeds, 2 times faster video transcode speeds, 2 times faster graphics performance, and 2.5 to 3 times faster artificial intelligence (AI) performance over previous generation products.” While Intel and AMD both sometimes inflate these comparisons by using older CPUs from previous product families, the footnotes to Intel’s claims reveal the company is comparing against chips like the Core i7-8565U. Despite the fact that it’s beginning shipments in June, Intel is still claiming that 10nm will be on store shelves for the holidays, implying a relatively long ramp.

7nm Coming in 2021, With GPUs Leading the Way

Intel didn’t just announce that it would ship 7nm hardware in 2021 — it plans to lead the way with a GPU design rather than a CPU. The company still intends to ship a commercial GPU in 2020, but it’ll have a 7nm follow-up for the Xe product family coming along the following year. Intel’s full quote:

Intel’s 7nm process technology [will] deliver 2 times scaling and is expected to provide approximately 20 percent increase in performance per watt with a 4 times reduction in design rule complexity. It will mark the company’s first commercial use of extreme ultraviolet (EUV) lithography, a technology that will help drive scaling for multiple node generations.

The lead 7nm product is expected to be an Intel Xe architecture-based, general-purpose GPU for data center AI and high-performance computing. It will embody a heterogeneous approach to product construction using advanced packaging technology. On the heels of Intel’s first discrete GPU coming in 2020, the 7nm general purpose GPU is expected to launch in 2021.

That’s another interesting claim from the company. The implication is that Intel will use EMIB or Foveros to connect functional blocks and that this 7nm product will be focused on data center work rather than a consumer GPUSEEAMAZON_ET_135 See Amazon ET commerce launch. Presumably, that market will be adequately addressed by what Intel launches in 2020, or the company simply may not have revealed its consumer timetable yet.

While Intel didn’t explicitly refer to the PAO (Process-Architecture-Optimization) schedule that it established in the wake of its first delays to 10nm, it did reiterate a promise to deliver an initial improvement based on node scaling and density, followed by another mid-node gain due to ongoing optimizations.

Intel-Presentation-10nm

This slide, shown by Dr. Murthy Renduchintala, Intel’s chief engineering officer, implies that at least one aspect to the leaks we reported last week was correct. Intel will continue to use 14nm chips through 2021. This would make 14nm the longest-lived node still used for “big core” chip production as opposed to motherboards or other ancillary products in Intel history.

Intel’s decision to accelerate its 7nm and EUV rollout is unsurprising, though the company is setting an aggressive goal for itself after years of foundry problems. Intel’s 10nm is expected to match the performance of 7nm from TSMC and Samsung, but both of those firms will begin introducing EUV products late this year or early next. Even delaying to 2021 represents a technology leadership gap that Intel is undoubtedly institutionally loathe to permit, having led the foundry industry for decades. At the same time, industry analysts and readers could be excused some skepticism about this roadmap. Intel didn’t forecast the 14nm problems that delayed that node, and it certainly didn’t expect its 10nm ramp to be delayed by four years.

At the same time, it’s not impossible for a foundry company to move this quickly. Because each company defines its own process nodes according to its own internal metrics, Intel can define “7nm” to be anything it likes. 7nm could mean “10nm, plus we pour chocolate sauce on it,” as far as Intel is concerned. It doesn’t mean that, but it could. [We would hope it doesn’t mean that. What a mess! -Ed] Presumably, Intel will incorporate other advances to the 10nm node other than just the addition of EUV, but the company hasn’t told us what kind of density gains or performance improvements to expect beyond the aforementioned 20 percent estimated improvement per watt in the context of its GPU architectures.

Launching 7nm in 2021 would probably constitute “catching up,” at least as far as AMD or Nvidia is concerned. While TSMC and Samsung should have 5nm silicon in-market by then, the PC foundry customers have not been transitioning every node the way the mobile companies have. Over the past few years, Apple and Qualcomm deployed chips on every node — 20nm, 16/14nm, 10nm, and 7nm — while AMD and Nvidia have used just 16/14nm and 7nm. (AMD and Nvidia both made use of a refined 16/14nm that was labeled as 12nm by GF and TSMC, but this was not a true new node and represented refinements to the existing 16/14nm processes.) It’s entirely possible that both companies will repeat this trend, especially given that TSMC’s 5nm is only expected to offer relatively small improvements to power and performance compared with 7nm.

It makes perfect sense for Intel to chase a fast 7nm transition. Will the company actually succeed? That’s harder to know.

Now Read: