Intel's woes as it struggles to move to full-fledged production on the 10nm node have done almost incalculable damage to the company's roadmaps, and that continues today with a report from ServeTheHome that verifies SemiAccurate's reporting this week that Intel has significantly pared down its plans for its next-gen Cooper Lake Xeon lineup.
However, Intel's latest alteration to its server roadmap actually bodes well for the company's confidence in its 10nm process finally coming to the data center in the Ice Lake Xeon lineup slated for the end of 2020. Those processors will face stiff competition from AMD's EPYC Milan processors, which are scheduled for release with the Zen 3 microarchitecture in the same time frame.
In short, Intel planned to slot its Cooper Lake processors into the single- and dual-socket Whitley platform, but has now decided to axe those plans. Instead of landing in single- and dual-socket servers, now Cooper Lake will be relegated to quad- and octo-socket systems built on the Cedar Island platform.
|Old Roadmap||Year - Platform||New Roadmap||Year - Platform|
|Cascade Lake Refresh||Now - Purley Refresh||Cascade Lake Refresh||Now - Purley Refresh|
|Cooper Lake||H1 2020 - Whitley (1S, 2S) / Cedar Island (4S, 8S)||Cooper Lake||H1 2020 - Cedar Island (4S, 8S)|
|Ice Lake||H2 2020 - Whitley / Cedar Island||Ice Lake||H2 2020 - Whitley / Cedar Island|
The move makes a lot of sense. Intel had planned for the Whitley platform to first launch with support for 14nm Cooper Lake processors, but the platform is also compatible with Ice Lake processors. That drop-in compatibility was important because the company planned to release Ice Lake a few months later, but many questioned if server vendors would adopt the Cooper Lake lineup given it's incredibly short lifespan, especially in light of the relatively long deployment cycles in the server market.
But then Cascade Lake refresh happened, presaging the change we see today. Intel announced last month that it would slash pricing on its existing lineup of Cascade Lake Xeons by releasing new models that come with up to a 60% price-per-core discount, muddying the waters. Intel's server roadmap was obviously too crowded to make sense with new Cascade Lake processors just coming to market, Cooper Lake debuting a mere few months later, then Ice Lake landing in the second half of the year.
Intel's decision to kill off the single- and dual-socket Cooper Lake platform, yet retain the quad- and octo-core platforms, is also easy to rationalize. The company announced last year that Cooper Lake would come with up to 56 cores in a single package, and those hefty core counts are well-suited for that class of server. Those beefy chips will also help fend off the core-packed AMD EPYC Rome chips in the interim, but Intel's Cascade Lake refresh will obviously serve as the company's weapon to fight Rome for the rest of 2020.
So how does this bode well for the 10nm Ice Lake roadmap? Intel has consistently iterated its Xeon lineup on Skylake/14nm derivatives, buying itself some time as it continually pushed the 10nm Ice Lake chips further into the future. As you can see from Intel's record-breaking revenue each and every quarter, that strategy has paid off, but industry speculation has run rampant that Ice Lake will also face yet another delay.
Discarding the single- and dual-socket Cooper Lake platforms essentially erases any buffer that the company could have between its recent refresh and the 10nm Ice Lake era, meaning its level of confidence in the Ice Lake schedule has to be pretty strong. Erasing that safety net is probably a gamble Intel wouldn't take if the odds were long.
Interestingly, a leaked slide popped up earlier this week detailing Intel's second-gen Optane DIMM roadmap, and this caused some confusion because it didn't include Cooper Lake listed with the Whitley platform. Instead, the slide has Ice Lake on Whitely with Cooper Lake on the Cedar Island platform. Now we know why.
Intel's 10nm Ice Lake chips will purportedly land at the tail end of this year. They'll face intense competition from AMD's EPYC Milan processors that will land with the 7nm process (probably an enhanced "+" variant) and should have architectural modifications, like a unified L3 cache, that will make them even more performant in a broader spate of workloads. That means Intel will still face intense competitive pressure, especially given that the company recently stated that it doesn't expect to reach process node parity with the TSMC 7nm node found in AMD's chips until late 2021.