Intel’s Comet Lake-S Could Push Even More People Toward AMD

Intel’s next-gen mainstream CPUs, dubbed Comet Lake-S, aren’t expected land in desktops until sometime in 2020. But the rumors and leaks about the company’s upcoming 14nm (++++?) processors have started to pile up. And boy, it’s hard to see how these chips -- which look to be a fifth (if you count Coffee Lake Refresh) re-warming of the company’s 2015-era Skylake architecture -- are going to find favor among just about any enthusiast or system builder next year.

Assuming the leaks hold true --  and we shouldn’t assume that they all will -- Comet Lake-S looks like the kind of lineup that would have been seen as a tepid, skippable update five years ago when Intel had little real competition in the desktop space. Here in the back half of 2019, AMD’s Ryzen 3000 processors (following on the well-received previous-generation Ryzen parts) have injected excitement into the desktop world, thanks to substantive performance (and performance per dollar) gains the likes of which we haven’t seen in going on a decade. Add in the bandwidth-doubling PCIe 4.0 interface packed into new AMD X570 motherboards and backward-compatibility with previous-generation, lower-priced X400-series boards and what Intel looks to be readying for early next year honestly borders on insulting.

What We Think We Know

Before I go any further, let’s take a quick recap of what we know -- or at least what we think we know -- about Comet Lake-S so far. According to recent reports from Hong Kong’s XFastest, at least partially corroborated by an ECS slide posted by Japan’s ASCII (both of which are covered in more detail here, Intel’s new mainstream desktop CPUs will have up to 10 cores, with a max TDP of 125 watts, while sporting the same 16 lanes of PCIe 3.0 bandwidth (a spec that’s been around since Ivy Bridge back in 2012) and the same (official) DDR4-2666 memory support of 2015’s Skylake chips.

Here’s a quick rundown of the leaked (again, unverified) specs of Comet Lake-S, according to recent reports, compared to last year’s Coffee Lake Refresh:

*specs in chart are unconfirmed

There’s no mention of the all-important clock speeds here, which would give us a solid sense of just how much better Comet Lake-S will perform compared to Coffee Lake. But the Core i9-9900K is already a very power-hungry, hot-running part with its eight cores--especially when overclocked.

If Intel is going to add 25% more cores and (presumably) increase clock speeds to some extent while sticking to its well-worn 14nm process node, that reported 125W TDP rating is likely conservative in the extreme. After all, at stock speeds, we saw the “95W” 9900K consume upwards of 145W at stock settings when running Blender. Granted, TDP ratings are a measurement of required heat dissipation at base frequencies, not power consumption. But if the 8-core 9900K is any indication, its 10-core replacement running on effectively the same architecture and process node is going to need some serious cooling, even at stock settings.

So even if Intel adds a couple of cores and keeps pricing relatively steady, as it’s done in the recent past, it’s still tough to see how, say, a 10-core K series chip in the roughly $500 range is going to make a whole lot of sense when a 12-core Ryzen 9 3900X sells for about that price (and will probably be less by early next year). With that Ryzen chip, you get a good in-box cooler, plus backwards compatibility that means you can install it in a sub-$70 B450 motherboard, all the way up to a high-end X570 model. With Comet Lake-S, you’re likely going to need a brand-new board with a new chipset.

Yes, Intel looks to be implementing yet another socket change for these new parts, requiring new motherboards for anyone who wants to swim in Comet Lake’s waters. This is hardly surprising given the company’s track record in that regard, but again Intel now has very strong competition from AMD -- which has been pushing out faster, bigger, more powerful CPUs on the same AM4 socket since September of 2016.

I’m sure Intel’s tactic makes motherboard vendors happy, but especially given what AMD has been able to push on AM4, Intel’s rumored move to yet another socket feels like a kick in the face (or at least the wallet) to its loyal customers. At the very least, it will likely push more frustrated consumers toward Team Red.

Stiff Competition

Speaking of AMD, let’s look at how the above rumored Intel specs compare to what’s available from Ryzen 9 3900X today, as well as what will be available soon (AMD says September) with the 3950X.

AMD in 2019 is offering more cores at a lower TDP (thanks to that sweet 7nm TSMC manufacturing process and AMD's different method of measuring TDP), and faster officially supported memory speeds than what it looks like Intel will be bringing to the table next year with Comet Lake-S.

And I barely mentioned the bandwidth-doubling PCIe 4.0 interface yet. We’ve covered it in detail and from different angles before, so I won’t go into detail here. But if Intel is indeed sticking with 16 lanes of PCIe 3.0 for its next-gen CPUs, AMD’s 24 lanes of PCIe 4.0 (which you’ll admittedly need a fairly expensive X570 motherboard to officially make use of) represents a 200% bandwidth increase compared to what Intel will reportedly offer with its new CPUs. Add the lanes from the chipset as well as the CPU, and that disparity is almost certainly going to increase even more. Granted, many won’t take advantage of all that bandwidth, but with SSD prices continuing to fall, more and more people are likely to run up against the limitations of Intel’s PCIe 3.0 lanes.

Zen 2 Isn’t Perfect

Of course, not everything is Zen in the world of Ryzen. Shortages of AMD’s high-end parts persist nearly two months after their launch, our testing has shown that not all Ryzen 3000 cores can hit the top advertised speed of the CPU, and almost all X570 motherboards require active cooling -- something that hasn’t been common since the days of dedicated northbridge and southbridge chips.

But if you don’t want to deal with the whir of an extra fan (and an extra potential point of failure), you can of course opt for a lower-cost previous-generation 400-series motherboard with AMD, although you’ll then lose PCIe 4.0 support. But it’s certainly nice to have that option, flexibility, and upgradility. Yes, I’m looking at you, Intel.

If we’re indeed several months out from new competing parts from Intel, and if all the company manages to deliver in early 2020 is yet another set of re-tweaked, hot-running 14nm CPUs with at best a couple of extra cores, AMD doesn’t have much to worry about in the short term, to say the least.

Granted, if 10nm still isn’t delivering the yields necessary for mass rollout, Intel likely feels like it has to release something. And there’s no doubt that Comet Lake-S will sell well enough to system integrators for new Intel-based desktops in big-box stores and for compact systems where Ryzen’s lack of integrated graphics (outside of its lower-end APUs) is a problem. New Intel chips will help move new PCs, especially in a world where the blue team’s long dominance means many mainstream customers don’t know enough to consider alternatives.

But with all the positive (and mostly well-deserved) press that AMD has been getting this year, even mainstream consumers are soon likely to start considering alternatives, as more and more AMD machines show up in Best Buy and on Amazon’s curated choice lists. Unless Comet Lake-S somehow turns out to be far more impressive than it looks on paper (granted, again from unconfirmed leaks), Intel’s next-gen desktop chips will be a Broadwell-like stop gap at best. And for enthusiasts -- even those who have been long-time Intel fans -- another round of modest performance gains packaged with yet another round of required motherboard upgrades is only going to increase the tide of consumers turning toward AMD.

Granted, the enthusiast and gaming desktop market isn’t the biggest focus for Intel, but the company is arguably facing even tougher competition from AMD on the server space with EPYC Rome, where margins are much, much higher. And if Intel starts losing customers en masse in the enterprise IT space as well as among gamers and enthusiasts, it will take some serious technology advancements to turn things around.

Note: As with all of our op-eds, the opinions expressed here belong to the writer alone and not Tom's Hardware as a team.