During Intel’s quarterly conference call last week, CEO Bob Swan confirmed that the company is, at long last, moving into volume production on 10nm. If you thought Intel had basically given up on scaling its process technology into the new node, that’s not the case.
Swan made a number of comments related to 10nm during the call. Ice Lake servers have been sampled to enterprise customers, with early production expected in 1H 2020 and volume production in the back half of the year. Cooper Lake (14nm) will share a platform with Ice Lake when those server parts launch in 2020. Regarding 10nm client launches, Swan said:
We began shipping Ice Lake clients in the second quarter supporting systems on the shelf for the holiday selling season and expect to ship Agilex, our first 10-nanometer FPGA later this year.
We now have two factories in full production on 10-nanometer. We are also on track to launch 7-nanometer in 2021. With a roughly 2x improvement in density over 10-nanometer, our 7-nanometer process, which will be comparable to competitors’ 5-nanometer nodes, and will put us on pace with historical Moore’s Law scaling.
Mobile in Q4 2019, Server in 2H 2020, Desktop …
Intel’s current plans for Ice Lake/Sunny Cove in its desktop CPU product families are unclear. If a Dell roadmap that leaked earlier this year is accurate, Comet Lake will refresh Intel’s product line through Q4 2020 with up to 10 cores, but still built on 14nm. We’ve seen various predictions about the state of Hyper-Threading; the most recent ones claimed Intel will reactivate it after removing it for the 9th Generation family. Restoring Hyper-Threading support would definitely improve performance compared with not-having it on various parts, but whether Intel will actually take this step is still uncertain.
Swan was actually rather open about expecting competitive pressure from AMD. While Intel has been talking a great deal about the possibilities of a $300B expanded TAM (based on the full valuation of the spaces Intel competes in), he also took care to say that Intel expects to be facing a reinvigorated AMD.
“Stepping back and just looking at the macro environment over the next several years and particularly in the second half of the year on the data center side, what we’ve indicated is it will be a much more competitive environment,” Swan said. Later in the call, he spoke to the topic again:
And our expectations over time are to protect our market share position, while continuing to invest in new prospects for growth… I’d say the competitive intensity on the PC side started probably in the first part of 2017. And during that time frame, we’ve either protect our position, while moving end customers up to higher performance products that generate higher ASPs and with that have the capacity also to fight back and meet comps in targeted areas, where we need to.
This is pretty frank talk, by Wall Street standards. The one thing Swan doesn’t do is speak to when we might see Ice Lake/Sunny Cove CPUs on desktops. Right now, it looks as though we’re still looking at a 2021 time frame for desktop 10nm, and 7nm chips are supposed to debut that year as well, though Intel has committed to leading the 7nm charge with GPUs, not CPUs.
The Ice Lake mobile CPUs that Intel has unveiled to date are reputed to be up to 1.18x more efficient than Intel’s old Sky Lake CPUs in terms of IPC, but Intel has given back a great deal of its clock speed gains over the past four years to deliver that improvement. The Skylake Core i7-6660U was a 2.4GHz CPU with a 3.4GHz maximum clock speed. Ice Lake is 1.18x faster in terms of IPC and runs at up to 4.1GHz. The real-world gains should, therefore, be significantly larger, once clock and IPC are both factored in — except, Ice Lake is the follow-up to Whiskey Lake, and the improvements relative to that chip are less certain. With a 4.8GHz single-core maximum, Whiskey Lake was clocked up to 1.41x faster than Skylake in the first place.
In short, it’s possible Ice Lake will be much faster than Skylake but roughly on par with Whiskey Lake. Given that we have no idea what the performance or power characteristics of Intel’s next-generation mobile GPU are, we’d also need to know how its power consumption and capabilities factor into Intel’s maximum defined clock speeds. The GPU configuration is much wider on these new chips, and that could definitely be eating into the total headroom Intel gives these processors. 15W, after all, is not a terribly large envelope.
With Intel’s 10nm desktop chips nowhere in sight and AMD’s latest Ryzen 3000 APUs still based on its 12nm second-generation Ryzen refresh, we have an amusing situation to consider. Even once Intel has shipped 10nm chips, its 10nm chips will not compete against AMD’s 7nm chips. That won’t happen until either AMD ships 7nm mobile parts or Intel ships 10nm desktop and server parts. We haven’t heard anything about a 7nm APU refresh in 2019. Assuming AMD doesn’t pull one of its hat, we may not see AMD 7nm face-off with Intel 10nm until sometime between April and June 2020.
Granted, I don’t think AMD is going to complain about having room to stretch its metaphorical legs. But normally when two companies start talking about their cutting-edge process node deployments, we expect to actually see CPUs facing off against each other shortly thereafter.