Say hello to Sky Lake Refresh Refresh Refresh Refresh

We've been talking for years about the travails of Intel's 10 nm process node.  Under the old tick-tock model, Intel would have two big core (i.e., not Atom) architectures on a given process node before moving on to the next.  The first architecture on a new node would be a die shrink of the second on the previous node.  There could be some relatively minor enhancements, but nothing revolutionary.  The big architectural changes would come from the second architecture on the same node.  That way, you're constantly either troubleshooting a new architecture or a new process node, but not both at once.  The "not both at once" makes things easier.  And for quite a while, it worked well.

On 14 nm, the plan was for Broadwell to be the first architecture, and then Sky Lake the second.  Then it would be on to Cannon Lake on 10 nm, while 14 nm would be relegated to chipsets and older parts.  Broadwell came and went.  Sky Lake came and was good, especially as compared to what AMD had to offer at the time.  And then the 10 nm process node wasn't ready in time.

Rather than having nothing new until the 10 nm node was ready, Intel went back and did a refresh of Sky Lake.  This gave them a chance to fix whatever mistakes they had made, and design against a more mature process node.  Normally, you wouldn't do that, as you'd get bigger gains by going to the next new process node.  But with no new process node available, doing a refresh of the old architecture on the old node was better than nothing, at least if you have the volume to justify it.  And Intel has a very large volume.

The refresh was Kaby Lake, which was still effectively Sky Lake cores, but now clocked higher.  Then 10 nm still wasn't ready, so they did another refresh, this time calling it Coffee Lake.  By this time, AMD was pushing 8-core Ryzen processors that were at least somewhat competitive, so Intel felt the need to offer more than their traditional four cores in a mainstream desktop.  In addition to higher clock speeds, Coffee Lake offered up to six cores.

And then 10 nm still wasn't ready, so Intel did yet another refresh, this time calling it Coffee Lake Refresh.  Apparently they gave up on new names or something like that.  Coffee Lake Refresh would go up to eight cores, and again, clock speeds went up.

Now it's time for another new architecture, and Intel finally does have 10 nm at least kind of working.  You can buy Ice Lake laptops now.  And they're so awesome that Intel felt the need to launch yet another architecture on 14 nm.  Yields tend to get worse as you go to larger dies, and for now, Intel doesn't have any Ice Lake chips with more than four CPU cores.  Quad core CPUs are so 2007, so for now, Intel is pushing yet another refresh of Sky Lake as the way to get more than four cores.  This time, they're calling it Comet Lake, and it's the fourth refresh of Sky Lake.

For now, Comet Lake is only available in laptops.  There are rumors that it will come to desktops, but that's not available yet.  The laptop version again tops out at eight CPU cores, but clock speeds went up once more.  Now you can get up to 5.3 GHz in a laptop under certain, very restrictive conditions.

AMD has done well in the enthusiast desktop CPU market in recent years, but in the three years since the launch of Ryzen, the laptop market has mostly been dominated by Intel, and for good reason.  Intel CPUs could clock higher, and at least as compared to first and second generation Ryzen, had better IPC.  Higher clock speeds while doing more per clock gives you better performance.

Intel CPUs typically had lower idle power consumption than AMD, which extends laptop battery life.  In addition to the CPU itself, Intel laptops supported various generations of LPDDR*, while AMD didn't, relying instead on normal DDR3 or DDR4 that used more power.  Shoddy design from laptop OEMs made the difference look larger than it really was, as they'd put in the work to make a premium laptop with other parts efficient at idle for their best Intel laptops, but go cheap for AMD.  That was a sensible thing to do, as Intel had the better CPUs, so if you're going to make a premium laptop design, you want a premium CPU, and that meant Intel.

But now with the launch of the Ryzen Mobile 4000 series, Intel's low idle power and hence longer battery life advantage might well finally be gone.  Even if it's not completely gone, it is certainly greatly diminished, as AMD is by far the most competitive that they've been on this count since Intel launched Merom (laptop version of Conroe) way back in 2006.  Tom's Hardware measured the new Asus ROG Zephyrus 14 as having a battery life over 11 1/2 hours.

With AMD now a process node ahead of Intel for the first time ever, AMD also has a load power consumption advantage over Intel, in the sense of needing less power for a given amount of work.  AMD still has much better integrated GPUs than Intel, for all of the noise Intel has made about how they're going to change that soon.

That leaves higher single-threaded performance due to higher clock speeds as Intel's main remaining technical advantage.  The key way that Intel has been able to increase performance with successive refreshes of the same CPU architecture on the same process node is by increasing the clock speed.  That has consistently come at the expense of higher power consumption.  Sky Lake wasn't stock clocked all that aggressively, so you could overclock it if you wanted.  By Coffee Lake, Intel was clocking their CPUs far more aggressively, so that the stock turbo would use up most of the headroom.  That's not a bad thing, but it did come at the expense of Coffee Lake Refresh largely ignoring the nominal TDP of 95 W and having no qualms about burning 150 W indefinitely.

You can do that in a desktop, as cooling 150 W with a beefy tower cooler in a mid-tower desktop is pretty easy to do.  Indeed, video cards commonly burn a lot more than that.  Now with the need to bring higher clock speeds to laptops, Intel is bringing that approach to laptops.  If a laptop vendor is willing to do all the design work necessary to make a laptop able to feed 130 W to the CPU, then the new Core i9-10980HK can turbo up to 5.3 GHz.  And if the laptop vendor would like the CPU to stay closer to its 45 W TDP, then the CPU won't dare go over 5.1 GHz.

For comparison, the previous generation went up to 5.0 GHz.  A 100 MHz speed bump from a new generation is better than nothing, but it's really not that impressive.  That's why Intel went through the games to make it offer a 300 MHz speed bump.  Don't expect this to be widely used.  It's largely targeted at the sort of laptop vendors that build top end gaming laptops by using a higher clocked desktop CPU.  Intel is making it so that their flagship laptop CPU can run at desktop-like speeds, at the expense of needing desktop-like power consumption.

The upshot is that Comet Lake ensures that if you want the highest single-threaded CPU performance you can get in a laptop, Intel will be comfortably ahead for another generation.  For comparison, AMD's Ryzen 9 4900H tops out at 4.4 GHz, as compared to 4.7 GHz for the desktop Ryzen 9 3950X.  But there is a considerable price to be paid in power consumption, and hence heat output, and the size and weight needed to properly cool so much heat.  I'd expect Clevo to offer it, as they've been willing to offer 130 W desktop CPUs in some big, thick, and heavy laptops in the past.  This isn't just a dumb marketing gimmick, but it is a pretty narrow niche.