Editor’s Note: this post is largely just me ranting about Intel’s garbage-tier marketing and childish strategies for a few thousand words. I think it’s a fun ride, but may not be your flavor! On the MMO side, I finished a complete Allied Race level under the new WoW systems and that’ll be up tomorrow.
It’s been a long, long time since Intel has been forced to compete, and it shows.
Over the month of October, AMD held two major events, prescheduled and preannounced, with the intent of each clear – the announcement of details and launch dates for Ryzen 5000 and Radeon RX 6000, respectively.
With what we know right now, Intel is done releasing desktop enthusiast parts for the year and their next major CPU launch for desktop is only penciled in as “Q1 2021.” No big deal, details aren’t really needed until Intel is ready to share a product stack with pricing, SKUs, first-party performance benchmarks, and then maybe some demos or side-by-side comparisons with AMD parts or with 10th gen Intel Core-series desktop parts.
Or so you would think.
Intel, in their infinite wisdom, made two blundering moves, the latest in a long string of them, during October that have left me scratching my head and revealed how little preparation they have actually made for AMD taking the performance crown from them. The night before AMD’s Ryzen 5000 event, the company posted a blog on the site Medium, talking about how excited they are for their Rocket Lake CPUs to launch early next year – and that was it. The other move was to release a press deck this last Friday, adding some feature-list bullet points to what Rocket Lake (the 11th generation Intel Core-series) will offer. I would have called what they did “detailing” the lineup, but that would be overly generous. Intel shared precious little actual information, other than confirming some rumors – it is a backported 10nm architecture on 14nm, 8 cores max instead of 10 (a downgrade!), with new Xe graphics on-die (they’re good for integrated but not a hype-worthy feature to the enthusiast audience), PCIE Gen 4.0 support (only a year and a half late!), and a new chipset (so buy a new motherboard again, oh joy!).
As an enthusiast, there are only a few bullet points worth getting even half-excited about – the official memory support goes up to DDR4-3200 (Intel considers XMP an overclock as of 10th gen and only certain motherboard/CPU combos support using a standard XMP profile past certain memory speeds so this is good news for buyers who get stuck in that morass of bullshit), it gets both PCIE Gen 4 and now has 20 lanes instead of 16 (Intel has been behind on reduced-latency direct CPU attach for NVME SSDs since AMD launched Ryzen), and AV1 support in the IGPU (in case you can’t get an RTX 3000 or Radeon RX 6000 card, you’ll be ready for more efficient, high quality web video). At least Intel had the decency to space this say-nothing deck out a week from Ryzen 5000 benchmark embargo instead of literally dropping the night before (but who knows, maybe they’ll pull that too!), but it is painfully obvious that Intel is flailing about for an answer to AMD at this point.
To be clear, Intel is a multi-billion dollar corporate entity whose interests and business activities stretch far beyond desktop CPU and GPU markets. When I say they’re being “beaten” by AMD, I only mean in the desktop market – because Intel has networking, AI, FPGA, graphics (lol, sort of), foundry and manufacturing, and software businesses heavily entrenched. They aren’t in any immediate risk of flaming out and dying, although looking at how seemingly off-kilter the company is these days, perhaps that is premature to say as well. I just want to make that crystal clear before this next part!
AMD has, since Ryzen came out in 2017, been rather quickly encroaching on Intel’s enthusiast mindshare. As a gaming CPU, first and second generation Ryzen weren’t anything incredible to write home about, but they were solid, competent steps that represented a massive leap over the FX series parts. However, the cadence of Ryzen has made Intel’s unprepared nature over the last few years incredibly clear to all. When Ryzen 3000 launched last summer, Intel’s marketing team was bellyaching to anyone who would listen about “real-world performance” and complaining that benchmarks like Cinebench don’t represent any actual use case to any real user. Never mind that Cinebench was one of many benchmarks that showed AMD ahead in the productivity side of things, because Intel tried to zero in hard on that one specific use case. They did, however, make a second, more valid argument – that they were winning in gaming. Because, at the top of the stack, they were. Until Ryzen 3000, Intel had an IPC (instructions per clock) advantage over AMD, and while Ryzen 3000 did beat Intel’s current designs at IPC, they were clocked much lower than the Core-series parts, which, coupled with core-to-core latency in the CCX design of Ryzen, meant Intel had a stranglehold on the tippity-top of the PC gaming market and used that to insinuate down the product stack that they were winning, even as in the mid-tier, they were not competing effectively against products like the Ryzen 5 3600.
With Ryzen 5000, AMD is going to claim an overall lead on single-core performance, if their benchmarks hold true (and generally, AMD does have a good track record here). Intel right now has nothing to show or counter with, and as the rumor mill has spun up since the announcement, there are tests in the wild showing even the bottom Ryzen 5 5600X beating the i9-10900k in single-threaded benchmarks! The fucking top of Intel’s stack is being walloped by a $300 6-core in single-core performance?! This is, to Intel, an obvious problem.
Before we dive into the Rocket Lake problem, this isn’t even the first time in the last 12 months Intel has been bested so badly in this way. AMD set an embargo date late last year for Threadripper 3000 CPU benchmarks, CPUs that absolutely trounced Intel’s high-end desktop lineup. Intel, in response, launched their rebadged “new” high-end parts with an embargo literally 6 hours prior to AMDs, changing it at the last minute and pissing off tech reviewers so bad that some of them chewed Intel out, changing the flavor of the review, and Linus Tech Tips even did the bold step of including ghosted bars for the Threadripper performance (violating the spirit if not the letter of AMD’s NDA) in the Intel review so that Intel’s gambit of not wanting to be compared against the new AMD parts did not pay off.
Intel behaving this way isn’t particularly new, but in the last great CPU competition between them and AMD in the early 2000s, Intel’s scummier practices were largely at an operational level rather than in marketing, and it was accompanied by consistent climbs in processing power (save for the introduction of the Pentium 4 compared to late-era Pentium III, which was…eh…not great for Intel trying to sell P4s!).
This time out, there are some particularly troubling signs that Rocket Lake is going to be a dud, though.
Firstly, the deck itself was pushed to US media outlets late on Friday, with less than an hour prior to Intel’s industry contacts closing business for the week. This doesn’t inherently mean much, but it says a lot when noting the lack of information in the deck and the difficulties many outlets had in getting answers from Intel to fill in the gaps.
Secondly, the core count decreasing by 20% at the top generation over generation is…troubling. It does make logical sense for a simple reason. Intel backported the design of their “Cypress Cove” core from their 10 nanometer manufacturing node to their larger 14nm one. What this means functionally, is that the transistor count, made assuming smaller transistor size and higher density on 10nm, is brought over fully intact to 14nm, making each CPU core physically larger, along with all other new elements of the core (the Xe graphics, the memory controller, the PCIE Gen 4 logic). Because Intel is using the same LGA 1200 physical socket as their current 14nm 10th gen parts, there is a limited amount of physical space on the substrate available, and they have to solve for routing challenges (making sure that components of the die line up with the interconnects that bridge into the socket or can be routed without signal degradation), so given those challenges, it does make sense. However, it still is a bad move, and one that signals a degree of worry to enthusiasts. Losing 2 cores means that the rest of the CPU must make up the performance deficit. Does it?
Thirdly, Intel is bragging about an unspecified “double-digit” IPC increase. Now, if Intel managed to get 19% like AMD with Ryzen 5000, or even higher over their Skylake design, they’d be shouting it from the rooftops. However, they aren’t – and that tells us a LOT right off the bat. Theoretically, double-digit means anywhere from 10% to 99% IPC uplift, and my guess is that it is sub-19% (I’ve seen 14% discussed in rumors, but I can’t fully remember the architectural comparison because Intel’s naming is difficult to follow). While 10-19% range is a good and solid increase, it means that Intel is likely still losing to AMD. Why is that?
Fourthly, Intel lists no clock speeds at all. When announcing new parts for the last handful of years, all Intel ever fucking talks about is 5GHz+ clock speeds. And to be fair, that’s their strength with Skylake and 14nm – they’ve wrung out every drop of performance from the core and process design and since AMD wasn’t even sniffing at 5 GHz parts, I think this was a good strategy in the past for Intel. The fact that they aren’t saying anything about clock speed ranges or an upper bound tells me they’re in trouble. All the optimizations Intel has made on 14nm have been Skylake specific, largely – process yield improving is design agnostic (mostly) but the high clock speeds are Intel’s engineers getting very good at pushing the Skylake design on 14nm very high.
Rocket Lake is a new core design and layout, though, and this means that all the clockspeed optimizations Intel has made over the last 4 years go out the window. In mobile on even the intended 10nm process, the “cove” series of core designs have struggled to go over 4 GHz, even when Intel has been able to push mobile Skylake-derived parts near or to 5 GHz+. With the inherent inefficiencies of the backport, Intel may struggle to push clockspeeds higher than 4 GHz on desktop Rocket Lake at all. The bigger problem in this issue is this – right now, using the AMD first-party numbers for Cinebench as a starting point, Intel is about 10% behind the Ryzen 5000 parts. If they gain 20% IPC but lose 20% clockspeed at max single-core turbo boost, then they end up…right where they started! The bigger problem is that the i9-10900k has a 5.3 GHz boost for a single core, so if a Rocket Lake part is only 4 GHz, it loses more than 20% – and it doesn’t take a math genius to realize that this is a nightmare if you are Intel.
Of course, the mobile parts don’t neatly map to what we can expect on desktop, but Intel’s Skylake cores and Cove-series cores on mobile could represent a sort of worst case, since Skylake mobile does scale nearly identically to desktop. The bigger problem is that the Cove designs on mobile at least get the proper 10nm process design and the reductions in power consumption that causes. Intel’s Rocket Lake slide deck for desktop shows that their “projections” for Rocket Lake assume the same PL (Power Limit) 1 and 2 targets as the current 10th-gen parts, and if the core design is less-optimized for 14nm and thus unable to reduce power draw as much as is needed even after cutting two cores, well, that could be a bloodbath in the benchmarks. Intel already loses multi-threaded benchmarks today, and by the end of this week they’ll do the same in single-threaded. If their next generation part is a step back in single-core performance AND a step-back in core count, well, that makes it damn near impossible to imagine buying anything but Ryzen 5000.
Fifthly (this format was not meant to count this high, was it?), the whole move just stinks of desperation, like a lonely guy at last call at a bar. Nothing about this announcement actually gives anyone anything to latch onto and say “hey, great, Intel is still competitive!” In fact, the whole slide deck goes incredibly far out of its way to avoid saying anything at all – just a bunch of acronyms, buzzwords, and “NEW” written all over it in sickly yellow font, trying so hard to pull any attention away from what is to come later this week from AMD. If I was starting to lose my foothold in a market that I’d dominated for nearly a decade and I didn’t have any way to actually meaningfully counter my opponent’s new product, I would simply shut the fuck up and hold until I have something specific and interesting to share.
I do suppose it kind of works for two audiences – one is investors, who may not be tech enthusiasts and to whom such a slide on a week like this shows stability and direction which that specific audience needs, and two is for fanboys on both sides – Intel fanboys to say “see, they’re gonna win again, just wait!” and AMD CPU fanboys (like myself, no denying it) to say, “ha, this is pathetic and Intel is going to keep on losing!” And I mean, for me, it was kind of funny to turn on my analytical brain and think through all the gaps in the slide deck from Intel and why they would be missing, and I’m sure on some level the company just wants people talking about them. Hell, if Rocket Lake surpasses the subterranean expectations this “news” sets for it, the product probably erases any criticism from the ledger and most people will talk it up more if, say, they ship a 25% IPC increase at 5 GHz and retake single-threaded performance from AMD in a matter of months, and people will (rightly) point out that a crown that took AMD over a decade to get back was taken away from them again in a fraction of the time.
For me, here is a bit of person Bobby, not writer Kaylriene – I’m an AMD fanboy in the CPU space, but I do like to keep that grounded. Intel has, for the last several years, been the better choice for gaming CPU once you start looking to spend more than $300~ dollars. Ryzen only made sense past that point as a productivity or mixed-mode usage, and/or as a vote against Intel. I’m extremely excited that Ryzen 5000 looks to give that crown to AMD and make it so that, no questions asked, in nearly every price bracket, AMD is the clear winner at gaming and continues to be strong in productivity. I like that AMD has focused on keeping an open-ish platform (all the fuss about delayed 400-series chipset motherboard support for Ryzen 5000 notwithstanding) and one that doesn’t arbitrarily lock off features like overclocking in order to squeeze an extra dollar out of the consumer. I’m extremely glad that the Radeon team pulled their head out of their ass at the high end and made what appears as though it will be an extremely compelling graphics card lineup.
However, since I turned 18 (2003, if you’re keeping score at home), for those 17 years, I’ve only owned an AMD system for 4 of them on the CPU front, and had a Radeon in my system for 5 of those years (actually, this genuinely surprises me, since I thought for sure I had more time with AMD CPUs than GPUs, but hey). When I build a system or recommend a build for friend/family/whomever else, I always focus on what works the best in that price range, and for far too many of my system builds, Intel CPUs/Nvidia GPUs tended to win. What I am most excited for is to see what meaningful competition does to the market, as AMD rising up huge in both of their target markets is a fascinating move. What I find interesting, perhaps bad, even, is that Intel has failed to meet the challenge on the CPU front, brushing AMD off, then complaining that people aren’t using the benchmarks they want you to use (which, coincidentally, are often Intel investments with Intel codepaths), and now are just posting Powerpoints and blogs on third-party sites just begging for a crumb of attention.
It is sad and pathetic in a lot of ways, and coupled with the announcement that Intel’s 7nm process is also delayed, I can’t help but worry if AMD eventually ends up being like Intel – slow, lazy, complacent, and fucking up once finally challenged. Nvidia has generally done better, but the flurry of rumors and unannounced cards being cancelled around the RTX 3000 series, coupled with the godawful lack of availability, suggests that they may also have been a bit too complacent.