In Desktop CPUs, AMD Is Doing These 5 Things Better Than Intel
Intel announced its 10th Generation Core “Comet Lake-S” line of processors today, and it exposes a few crucial areas where the chip giant needs to catch up to its scrappy, nimble rival.
This week, Intel announced the launch of its new set of mainstream-focused desktop processors, the 10th Generation “Comet Lake-S” chips, which are set to go head-to-head with AMD’s similarly positioned “Zen 2” line of Ryzen CPUs. On its surface, this announcement from Intel is familiar and predictable, lacking the revolutionary flair that came with the launch of Zen 2 last year, leaving AMD in a prime position to expand its footprint in the desktop-CPU market.
On the surface, the launch of Comet Lake-S may have the appearance of keeping Intel competitive with the performance hounds out there (gamers and creators alike). But once you start to scratch through the PR crust, the core (no pun intended) of what’s on offer here may not be as substantial as the chip giant would have you believe. Despite offering the newer product (AMD’s Zen 2 will hit its first anniversary on shelves this coming July), Intel still lags behind its chip rival in a few key areas that warrant looking into before you decide which processor will live in your next desktop PC, whether it’s a build-your-own project or a prebuild from a system OEM.
Thing 1: AMD’s Lithography, or “14” Is Not “7”
To get it out of the way early: There’s no escaping the 14nm elephant in the room. The 10th Generation Intel desktop processors, releasing in 2020, are still based off a lithography that likely saw its first draft on the drawing board sometime in 2010, right around the time the Apple iPhone 4 was hitting shelves.
Now in fairness, it’s possible that AMD’s plan for 7nm wasn’t much further behind that. (The company unofficially uses the line that Zen 2 was “eight years or so” in the making.) True roadmaps for CPUs, from the initial brainstorm meeting to the first chips leaving the lines, can be anywhere between a few years up to potentially a decade in Intel’s case. It’s understandable that these things take time, and new process technologies are ever harder to execute on.
The reasons why Intel finds itself so far behind in a race it’s been dominating for the better part of a quarter-century are multiple, diverse, and not all the company’s own fault. Intel has had a litany of publicly known issues with getting certain fabrication processes off the ground (the move to 10nm has been fraught with problems from the start), and the company has even openly admitted that its own 7nm processors for mobile chips are still not due to arrive, at minimum, before the end of 2021. It’s also possible COVID-19 may delay (or already have delayed) this timeline even further, though Intel hasn’t made any public statements about the effect the pandemic has had on production times or process roadmaps.
Thing 2: The Platform, or Yes, Another New Socket
Aside from the modest gains being made on an aging (downright ancient, by modern standards) architecture, Intel is making the switchover to 10th Gen a tad less appealing by requiring all-new motherboards: That is, a platform switch to a new socket (this new one is called “LGA 1200”) and the need for new 400-series chipsets.
While there’s nothing new about platform switches, requiring a third in the same lithography (14nm) seems…let’s say gratuitous, and doubly so for anyone who might want to upgrade from a 9th Generation part. That’s because all new 10th Generation Intel chips moving forward will only be supported on the new socket, using boards with the Z490 (high-end), B460 (mainstream), or H410 (budget) chipsets.
Z490 boards will launch first, with the cheaper ones to follow soon after. (You can read more about the Z490 platform and the full list of upgrades it has over Z390 in our Z490 board mega-guide. It details the around 40 different motherboards that are making their way down the platform pipeline over the next few months.)
Some Z490 boards from the likes of MSI, Asrock, and Gigabyte will come with latent support for the new PCIe 4.0 standard (i.e., touted as PCI Express 4.0 “ready”), however Intel notes that this batch of 10th Generation chips supports only PCIe 3.0, and the chip maker was mum about future possible developments on that front. So, what gives? Well, the speculation is that Intel plans to use Z490 as the platform for its next generation of chips, too, though when those are due out (and along with it, the option to run any kind of Intel-based PCI Express 4.0 system) is anyone’s guess.
The problem with pushing 10th Generation to a new platform without support for some of the main features that justify the cost of upgrading? If you’re, say, an Intel Core i9–9700K owner, and Intel is only promising an average performance boost of around 15 percent, that means you’ll be investing potentially half the cost of upgrading your machine (depending on the motherboard and CPU combo) into just a few frames per second extra on your favorite game.
AMD, meanwhile, is still allowing users on its platforms based on the AMD AM4 socket and earlier chipsets to install top-of-the-line 16-core Ryzen 9 3950X chips into some boards (depending on BIOS-update support) that cost $100 or $150 four years ago. (That said, I’m sure the company wouldn’t expressly recommend it.) But the point is: wide backward compatibility with many old boards and new chips.
Sure, upgrading to boards based on AMD’s X570 chipset yields you cutting-edge tech like PCIe 4.0, but if you’re on a budget and just want to upgrade an older AMD AM4-socket-based system you already own, the company has cut a clear path for you to do so, and you might have to spend substantially less than those folks upgrading to an Intel equivalent would.
Thing 3: Pricing Remains an AMD Edge, Depending on How (and When) You Look at It
Let’s be clear: It doesn’t all appear to all be grim news for Intel. At the top end, some value gains are apparent in the Core i9–10900K, which succeeds the Core i9–9900K and adds two more cores at a price reduced from its predecessor’s by…$11. Yes, $11. Why $11? Not sure, but it’s $11 less.
This almost militarily-precise pricing structure continues down the stack, with the Core i9–10900KF costing $472 MSRP, the Core i7–10700K costing $374, and so on. While seemingly strange at first, one only needs to check the current street prices of current AMD processors for a sense of where Intel may have gotten its influence…
For a moment, some of the chips in the KF and F lines (those are the Intel chips without integrated graphics, like most of the equivalent Ryzens) almost look like a good deal. But…
Also consider this. If these new Intel chips are just now debuting at the street price of AMD chips that were launched in July of last year, the basic math tells you most of what you need to know in this story. Once AMD’s expected Zen 3 architecture makes its debut sometime this year (AMD’s latest earnings call told investors to expect a “late-2020” launch alongside RDNA 2), with it will almost definitely come price-slashing across the board for the current crop of Ryzen processors. If Intel is still asking full price for this line of 10th Gen CPUs by the time that happens-and it probably will, as the company is known for almost never dropping the price of its processors until years after launch-every processor in the stack will be left in the cost-per-core dust by AMD yet again, two years in a row and counting. (Plus most AMD Ryzen processors come with nice stock coolers, which aren’t free. Just saying. Intel’s are a mixed bag of cooler-versus-no-cooler, and the stock fan is, shall we say, homelier?)
Thing 4: Chiplets Can Help With Yield
When Zen 2 launched last year, one of the most surprising elements of the whole reveal was just how many cores AMD had managed to cram into one die at seemingly impossibly low prices. The pricing undercut the market of the time by nontrivial percentages across every vertical, including desktop gaming and content creation.
This is because AMD’s new lithography allows for much more flexibility in how many dies can be utilized per wafer during the CPU production process. Because the entire stack of Zen 2 chips relies on the same underlying “chiplet”-style design, AMD (or more specifically its chip fabricator, TSMC) has been able to achieve very good usable chip yields when spread across the full product line: One anonymous source claims 70 percent yield to Intel’s 35 percent with certain Zen 2 versus 14nm chips. Take those exact numbers with a big grain of salt, but given the core/thread count for the money AMD has been offering, and the possible ability to down-bin partly functional chips to lesser SKUs, AMD is likely in a strong position here.
The more yield AMD gets per wafer, the less it spends on production of the same number of chips, and the more savings it can pass onto buyers.
A major point of foreboding in this department: Intel hasn’t announced any concrete plans to adopt a chiplet microarchitecture approach for its desktop processors anytime in the near future. And that could mean AMD (again, read: TSMC) might maintain this ostensible yield advantage over Intel until the company switches either to that kind of design or to any of the other newer technologies that have been proposed to speed Moore’s Law back up a bit for the next decade or so (such as chip stacking, or new lithography processes).
Thing 5: Content Creation Work? It Sure Loves Cores and Threads
We should preface this section by saying that AMD processors aren’t specifically better than Intel for running creative tasks on a core-to-core basis. But in this instance, it’s just the cost-per-core ratio that gives AMD a big edge.
In almost every modern creative application, more cores and threads supported generally means more performance, all else being equal. That’s because unlike most games, which max out at using four cores at a time, creative applications that need to complete tasks like rendering complex scenes in 3D worlds, layering dozens of music tracks on top of each other, or encoding videos into new formats will take every available ounce of processing power they can get to finish the job.
If the pricing trends of Zen 2 continue on the same path that was first established in July of 2019, AMD and its coming Zen 3 architecture could bring the per-core prices even further down a sinkhole than it already has. And that’s a daunting prospect for any other companies that might dare challenge them.
Processors like the Ryzen 9 3900X and Ryzen 9 3950X have begun to bridge the gap for what kind of power is available to amateur and professional creators alike, giving even bedroom artists the opportunity to design, draw, animate, and render to their heart’s content for a price that’s far lower per-core than what Intel offers. And this is to say nothing of the raw power that 32-core monster CPUs like the Ryzen Threadripper 3970X bring to the table.
Now, the Intel story isn’t so bleak here as it may have been with the 9th Generation desktop lineup, looked at as a whole. Most of the chips in the new Comet Lake-S line support thread-doubling Hyper-Threading, which is a big boost from the last generation, in which only some chips, mostly at the high-end, supported it. That will help in the content creation market up and down the stack, no doubt about it.
Intel claims the new breed of 10th Gen chips will be “Up to 18 percent faster than previous generations,” and in checking the fine print, we see this statistic is pulled from an “Adobe Premiere Pro CC RUG [run that] measures the time it takes to export a Premiere Project timeline to 4K HEVC video using the ‘Software Encoding’ option which is CPU only.”
The bad part? In tests like Cinebench R15 and R20, which stress all cores, as well as Handbrake video encoding, the Ryzen 9 3900X performed anywhere between 25 and 45 percent better than the processor Intel is using for a point of comparison here, the Core i9–9900K.
Run the math at each end of the spectrum, and on average the near-year-old, 12-core/24-thread AMD Ryzen 9 3900X could, in theory, still produce results between 7 and 38 percent faster in content creation than the spanking-new 10-core/20-thread Intel Core i9–10900K. We’ll have to see when we test the latter if that projection holds.
Again, the lithography, die design, and production process are all part of the advantage for AMD here, and until Intel makes some serious changes to the underlying architecture of its desktop chips, that lead isn’t going anywhere anytime soon.
The Last Thing? Intel’s Got a Long Road Ahead
These days, a lot of the world can feel like we’re living in Opposite Land, and perhaps nowhere is that more apparent than in the current standings between Intel and AMD. AMD has struggled for much of its recent history to beat Intel at its own game, and up until the middle of last year, there looked to be no end to that power dynamic in sight.
Now, though, Intel is grasping at increasingly fine-sliced straws to remind you why you should still insist on Intel Inside in your next PC build or upgrade. Much of it revolves around single-core performance or the ability to boost a few cores rather than all of them in applications that use just a few.
This launch is more a story of what’s missing than what’s there. I suspect that gaming performance with these new 10th Generation CPUs will continue to outpace AMD on around the same level it did for the Zen 2 launch, and there will assuredly be some Intel wins in overclocking potential. (The chips in this current crop of Ryzens aren’t famed for their overclocking headroom.) But despite all of that, in many ways, I’m still reminded of…the game Fallout 76.
Even though it was released in 2019, the game runs on an engine that was technically designed and built in 1998 for the development of Elder Scrolls: Morrowind. Now, in the 22 years since the company made it, Bethesda has become an all-out master of that engine; its developers know it back to front and in and out. But ultimately at its core, it’s still an engine from 1998 trying to run a modern AAA game (online, no less).
The case of 10th Generation Comet Lake-S from Intel feels similar. Due to many factors that were in its control and a few that weren’t, the company is still at 14nm++, and there are only so many ways you can try to put a shine an old lithography before the cracks start to show, especially when AMD has been blasting such a bright spotlight on them for getting on a year now.
We haven’t had an opportunity to test any chips in the new Intel stack just yet, but rest assured that as they come in we’ll be benching them to see if Intel’s claims of gaming supremacy hold fast. With Comet Lake-S, the chip giant has been more aggressive with its pricing structure this time around, and its new offerings on overclocking headroom are enticing, but we’ll have to wait to see how the chips do under scrutiny once they arrive.
Originally published at https://www.pcmag.com.
