Howdy, Stranger!
It looks like you're new here. If you want to get involved, click one of these buttons!
Quick Links
Cannon Lake arrives, sort of. Is terrible.
https://www.anandtech.com/show/12749/first-10nm-cannon-lake-laptop-spotted-online-lenovo-ideapad-330-for-449
Apparently Lenovo is releasing a laptop with a Cannon Lake CPU, at least in China if not elsewhere. That's progress, right?
Until you see the specs. It's a dual core CPU without an integrated GPU. The max turbo for the CPU is a meager 3.1 GHz.
For comparison, the max turbo for laptop versions of Coffee Lake ranges from 3.6 GHz to 4.8 GHz. The max turbo for Ryzen Mobile ranges from 3.4 GHz to 3.8 GHz. Most of those CPUs have at least four cores, too, and all of them have an integrated GPU.
So basically, Cannon Lake is slower than any current generation laptop CPUs in single-threaded performance, with the possible exception of sometimes being able to catch the slowest of the Ryzen Mobile quad cores when only one core is in use.
And without an integrated GPU, Cannon Lake is also unsuitable for laptops, as a discrete GPU is going to burn way too much power at idle. Yes, this is a laptop part unsuitable for laptops.
Cannon Lake does have an integrated GPU in the chip. GPUs also tend to be pretty severable, as if a chip has eight compute units and three of them don't work, you can disable the ones that don't work and sell it as a GPU with five compute units. The only time you can't do that is when the defective part is something that the chip only has one of, or when you have multiple copies of some piece and they're all defective. Most of the chip consists of some pieces that are copied a bunch of times.
The specs on the Core i3-8121U demonstrate that yes, Intel does have working Cannon Lake 10 nm chips, for some loose definition of "working". But they also demonstrate that yields are so awful as to make the chips nearly unusable. Most likely, Intel cut a deal with Lenovo to take some early chips off their hands very cheaply so that they'd at least get more money for them than throwing them in the garbage. But those specs don't imply yields that are close to having a commercially viable product.
This does not mean that Cannon Lake is forever doomed. Process node improvements or respins could fix yields and ultimately make it into a terrific product, or at least a viable one. Many chips that go on to be very good had terrible yields early on before problems got fixed. But for most of them, "early on" doesn't mean "two years after the chip was originally supposed to launch". Intel has surely had time to do plenty of respins of Cannon Lake by now, meaning that it's almost certainly a process node problem.
Of course, a product that would have been excellent for its day if it had launched in 2016 might be mediocre for its day if it doesn't arrive until 2019. That's the problem with lengthy delays: even if you can eventually fix it, by the time you do, it's often too late for it to still be a good product.
Apparently Lenovo is releasing a laptop with a Cannon Lake CPU, at least in China if not elsewhere. That's progress, right?
Until you see the specs. It's a dual core CPU without an integrated GPU. The max turbo for the CPU is a meager 3.1 GHz.
For comparison, the max turbo for laptop versions of Coffee Lake ranges from 3.6 GHz to 4.8 GHz. The max turbo for Ryzen Mobile ranges from 3.4 GHz to 3.8 GHz. Most of those CPUs have at least four cores, too, and all of them have an integrated GPU.
So basically, Cannon Lake is slower than any current generation laptop CPUs in single-threaded performance, with the possible exception of sometimes being able to catch the slowest of the Ryzen Mobile quad cores when only one core is in use.
And without an integrated GPU, Cannon Lake is also unsuitable for laptops, as a discrete GPU is going to burn way too much power at idle. Yes, this is a laptop part unsuitable for laptops.
Cannon Lake does have an integrated GPU in the chip. GPUs also tend to be pretty severable, as if a chip has eight compute units and three of them don't work, you can disable the ones that don't work and sell it as a GPU with five compute units. The only time you can't do that is when the defective part is something that the chip only has one of, or when you have multiple copies of some piece and they're all defective. Most of the chip consists of some pieces that are copied a bunch of times.
The specs on the Core i3-8121U demonstrate that yes, Intel does have working Cannon Lake 10 nm chips, for some loose definition of "working". But they also demonstrate that yields are so awful as to make the chips nearly unusable. Most likely, Intel cut a deal with Lenovo to take some early chips off their hands very cheaply so that they'd at least get more money for them than throwing them in the garbage. But those specs don't imply yields that are close to having a commercially viable product.
This does not mean that Cannon Lake is forever doomed. Process node improvements or respins could fix yields and ultimately make it into a terrific product, or at least a viable one. Many chips that go on to be very good had terrible yields early on before problems got fixed. But for most of them, "early on" doesn't mean "two years after the chip was originally supposed to launch". Intel has surely had time to do plenty of respins of Cannon Lake by now, meaning that it's almost certainly a process node problem.
Of course, a product that would have been excellent for its day if it had launched in 2016 might be mediocre for its day if it doesn't arrive until 2019. That's the problem with lengthy delays: even if you can eventually fix it, by the time you do, it's often too late for it to still be a good product.
Post edited by Quizzical on
Comments
This block Intel seems to have perfecting 10nm is a bit shocking. All (early) signs seem to point to Global Foundry's 7nm process doing very well in comparison.
Lenovo probably only pushed this out because some contract somewhere obligated it to be created.
It's possible that they simply ran into a wall due to physics. While TSMC has announced that their 7 nm process node is ready for mass production and Global Foundries is supposedly going to have their own 7 nm node soon, we don't really know how good those nodes will be. If they offer the sort of gains we'd hope for, then all is well and Intel simply had problems of their own. But if they offer dismal results comparable to Cannon Lake or are only really useful for <50 mm^2 chips, then we might be running into the end of useful die shrinks.