All rumors should be taken with a grain of salt. This one should be taken with a lot of salt. If I didn’t think it was at least possible I wouldn’t be discussing it, but I think it’s important to make that point upfront.
According to the German site HardwareLuxx, Intel has canceled all of its 10nm desktop CPUs. 10nm, 10nm+, and 10nm++ will all be mobile and server parts, while desktop chips will supposedly remain on 14nm until 7nm chips are ready in 2022, with the advent of a new chip, Meteor Lake. The website states, “From insider circles we have received some exclusive information about future desktop processors from Intel. The source has proven itself several times in the past on CPU issues.” (Translation via Google Translate).
The problem is supposedly Intel’s 10nm node and the difficulty of bringing clock speeds up to where they need to be for desktop parts. There is a reason to think this could be true: Ice Lake’s CPU clocks dropped substantially with 10nm, though its IPC came up to compensate. Intel has previously stated that it expected its 10nm+ and 10nm++ nodes to compensate for this, and the company’s original 10nm presentation from back in 2017 didn’t show 10nm transistor performance crossing over with 14nm++ until the 10nm+ node.
Up until this rumor dropped, my assumption was that Intel probably wasn’t going to launch 10nm desktop chips, or at least not many. To date, they’ve been noticeably absent from any discussion of its 10nm product families. Tiger Lake, an upcoming 10nm+ CPU, was supposed to have a desktop launch, but that’s the part HardwareLuxx is claiming has been canceled. Rumor suggests that Rocket Lake might use a 10nm GPU and 14nm CPU, but if Intel waits for 7nm to launch new desktop CPUs, that’s going to leave them using the same fundamental desktop architecture for seven years.
This outcome could be downright problematic for Intel, but it’s also why I’m uncertain to sign on to this rumor in the first place. The 10nm delay Intel has already suffered is the company’s worst production miss in the past two decades. Pushing Skylake until 2022 would mean one architecture had dominated Intel’s desktop product family for seven years. Intel, of course, is in vastly better shape than AMD was when it was forced to keep Piledriver around for five years (2012-2017) to anchor its desktop product stack. But no company wants to be in this position. Having 10nm, 10nm+, and 10nm++ parts to ship in mobile and server would help the company remain competitive in those markets, but desktop is still an important segment as well.
Giving up on 10nm+ and 10nm++ for desktop would mean ceding the market to AMD for the entirety of TSMC’s 7nm node. We would expect AMD to move to 7nm+ in 2020, followed by 5nm in 2021. Even if AMD delays moving to 5nm, it would still have the option to beat Intel to 5nm in 2022, assuming Meteor Lake doesn’t launch in January of that year.
Let’s block ads! (Why?)
Read more here: ExtremeTechComputing – ExtremeTech