Sidenote: This Week’s AMD Ryzen Fourth Generation Unveiling and The CPU Landscape

I thought about putting this out closer to Thursday, the actual event day, but ideas I wanted to share kept popping up, and the WoW posts I have in the works need a bit more play in the beta to qualify some observations I’ve had. So, as a result, let’s talk gaming tech!

The biggest turnaround in the gaming market the last few years has been the re-emergence of a highly competent and competitive AMD in the CPU market. After spending much of the 2010s flat on their asses due to a mix of over-commitment to a design that did not work (Bulldozer and the FX series CPUs) and the technical advances of Intel’s lineup with Sandy Bridge in 2011. Ryzen’s first generation lineup in 2017 went a long way towards building goodwill with the enthusiast PC community, but AMD had a lot of work to do to get anywhere near their previous heights as a CPU brand and to re-establish themselves in the mainstream (where AMD has always had a perception problem as the sort of Great Value, store brand version of an Intel CPU, regardless of how little that has tracked with reality for much of the company’s history after the 1980s).

Ryzen’s story through to today has been a similar one each year. At their initial launch, Ryzen beat Intel’s offerings in multimedia work quite handily, with some exceptions, but Intel has retained the gaming crown due to faster single core performance, an advantage they only hold at present due to much higher clock speeds and power draw (the i9-10900k is, technically, the best gaming CPU you can get, but only because it offers a 2-core boost clock of up to 5.3 GHz, compared to AMD’s Ryzen 9 3950x’s modest 4.7 GHz). Because of this, even as AMD has taken the lead in IPC even today when comparing the two brands best at the same clock speeds, if literally all you want to do with your PC is game and consume content, Intel still, technically, wins. Intel also has entrenched supply deals with most major system integrators, and so even as AMD has been on the upswing against nearly 5 years of Intel’s technological stagnation, AMD has had to fight against inertia to get design wins as their performance has improved and started to beat Intel. Only in the last year with the 7nm Zen 2 core offerings has AMD started to be seen in more laptops, desktops outside of major gaming system integrators, and in advertising for those same system integrators.

AMD has their best shot since the early 2000s to fully dethrone Intel with this week’s Zen 3 design unveiling. Intel is in their weakest state ever, shuffling through management personnel, delaying their next major silicon manufacturing process even as the 10nm node that was supposed to carry them into the future remains largely MIA in most market segments, and until we see the Sunny Cove core design wind its way into a desktop or datacenter CPU, they are still selling derivatives of the same Skylake architecture they’ve had since 2015, security flaws and all. AMD, meanwhile, is currently ahead on IPC and could overtake Intel in gaming performance with a combination of higher clock speeds, better IPC, and/or reduced latency. The good news for gamers with AM4 motherboards in their machines is that it looks like they are going to take all of the above, and depending on the scale of it, they’ll win over Intel just overall.

If AMD can win the overall performance crown, it makes the Ryzen story simpler and more compelling, and should, in theory, prompt a stronger response from Intel when they pull their heads out of their asses and remember how to advance their technological assets again.

But I want to focus on the raw features of what I expect Ryzen fourth generation to offer, based on a mix of confirmation via AMD slides or materials, rumors and leaks, and pure speculation. I’ll do my best to ensure that each point is properly labeled and annotated as such and when the stream is over on Thursday, it’ll be fun to see what I got right and/or wrong. Here goes!

IPC Increases: Higher for Single-Threaded than Multi-Threaded, around 15% Overall Uplift: A mix of speculation and leaks here – a couple of leaked benchmarks (Ashes of the Singularity and a CPU-Z score) show around a 20% uplift single-core and a 15% multi-core improvement. While this seems weird at first, it makes sense with a few things we’ll discuss in more detail. For now, the summary – higher and better-sustained boost clocks would proportionally affect single-threaded performance more, multi-threaded performance would be boosted by a mix of sustained boost clocks and the reduced latency of the CCX redesigns, and both benefit from higher IPC – rumored to be a bit lower at around 10-12% and mostly in integer workloads. Combine all of those, and this result makes sense!

CCX Redesign Confirmed – Lower Core-to-Core Latency Improves Multi-Core Performance: The confirmed details we’ve seen in prior AMD slides about Zen 3 is that the generation entails a full redesign of the Zen CPU core rather than tweaks and modifications as we’ve had for the 2nd and 3rd gen parts. The key takeaway that AMD communicated in those materials is that the design is changing from a 4-core-centric design to an 8-core one. Turbo-explainer: Ryzen CPUs through today are sorted in logical groups of 4 physical cores, in units called CCXes. A CCX is 4 cores and a shared L3 cache between them, and any communication with cores outside of a CCX requires a higher-latency trip over AMD’s proprietary interconnect mesh, the Infinity Fabric, which, in some benchmarks, takes 5 times as long as a same-CCX communication between cores. A current Ryzen 8-core is 2 clusters of 4 cores each, my 12 core 3900x is 4 groups of 3 cores each (because AMD disables cores in CCXes symmetrically for the most part to bin lower-end SKUs), and the 16-core 3950X is 4 clusters of four cores each.

With the redesign, an 8 core CPU in the 4th gen parts will be a single 8-core cluster. This has two major effects that improve performance. First, each core gains access to the L3 cache that was previously kept for the paired cluster of cores, which results in an effective doubling of available cache per core, which can help single core performance or lightly-threaded performance. Secondly, and more profound, having an 8-core cluster means that core-to-core latency is drastically reduced, improving multi-threaded performance sharply and removing one of the common causes of microstutters and small dips in performance with the current Ryzen lineup. There will still be slower core-to-core communication in the higher core count parts (anything over 8 cores) but for gaming, fourth-gen Ryzen will be substantially better as gaming is latency-sensitive, while the kinds of applications that already do well on Ryzen aren’t latency-averse and will see smaller gains, but should still benefit from the improvements made (and especially the available cache per core advantage this change offers).

Higher, More Steady Boost Clocks: Ryzen 3rd gen did away with a lot of the branding terms around AMD’s opportunistic boost for Ryzen. Instead of “Precision Boost” and “XFR” offering increases with defined terms and documented behavior, there was simply a printed maximum boost clock, which, depending on a myriad of conditions, could end up never being met. My 3900x, for example, has not shown me a 4.6 GHz clock speed once, at least for as long as I have been actively monitoring it – which is not to say it hasn’t happened, but rather, that the boost behavior is very granular. The CPU can (and does) boost to higher speeds for fractions of a second to power through a task and then immediately drop back down.

Rumors are that a key target is to increase peak boost clocks, with a rumored 4.9 GHz 12-core floating around. However, the better rumor I saw is that not only is that the case, but that AMD has targeted improvements in boost stability, trying to push the silicon to maintain the higher boost clocks for longer periods of time with less deviation. Intel’s Turbo Boost tech is a “dumb” boost in that as long as the CPU isn’t melting itself to death, it can run at maximum boost indefinitely. The i9 CPUs in the current gen do have an additional opportunistic boost if they maintain temperatures below 70 degrees Celsius, which can be a problem due to the higher power draw. If AMD can manage to stabilize their boost to maintain the frequency for longer without dips (no matter how intelligent and workload driven the dips are), that will offer an improvement that can be easily seen and measured.

There is one other piece to that puzzle, however. The Ryzen 3rd gen CPUs are very power-conservative, with rated thermal design power of 105w for even the 16-core! AMD and Intel measure this differently (Intel’s is the rated power at base clocks only, while AMD’s is an average which includes boosted frequencies) but Intel currently has higher TDP for their top-end parts (125w for the i9-10900k). The rumored TDP for the 12-core 5900x (more on that in a moment) is 150w, which, while it should include boosting in the average, means that the CPU will have access to greater power draw which will directly translate to higher clocks, higher average clocks, and greater performance.

Advancements to the 7nm Node: AMD has been one of the biggest beneficiaries of TSMC’s 7nm manufacturing process, with both Ryzen 3rd-gen and the Radeon RX 5xxx GPUs made on it. TSMC has newer versions of the same process which offer higher transistor density and improvements to clockspeeds and power usage. While AMD’s slides originally confirmed using the N7+ node, they have since backed off of that claim. However, I would still not be surprised to see them confirm use of N7+, as it offers a 20% increase in transistor density and can result in higher clockspeeds as well. Both of these changes would allow AMD to pack slightly more hardware into an equally-sized chip, and the clockspeed improvements are self-explanatory.

Ryzen 4th Gen is…Ryzen 5xxx?: Seems that way. AMD has been, confusingly, shifting generational monikers around due to the APUs and mobile lineup, so the current Ryzen 3rd-generation mobile CPUs have 4xxx model numbers. In order to consolidate this and correct the naming bloat, they seem to be making one last confusing move – to call the next generation 5000-series, despite being the 4th generation. This one doesn’t mean much, but fun trivia!

There’s a 10-Core SKU Rumor and I Don’t Think So: Some pundits were discussing a 10-core variant of the next-gen CPU, and the thought in my mind is…why? It would allow AMD to offer a direct i9-10900k competitor which they could massage into being dominant all-around, and I think that isn’t altogether a bad idea, but it just seems really weird. Unless AMD offers only one SKU per core count (instead of the 3rd gen, which has two different 8 core CPUs), it would start to get really weird. So while they could have an 8-core 5700x, a 10-core 5800x, and the 12-core 5900x, I don’t think they’ll go that route. But, I could be wrong – the 3800x 8-core made almost no sense compared to the 3700x, as it offered a very small improvement in performance at an $80 premium in price, and both even included the same cooler design despite the 3700x being a 65w part and the 3800x being 105w. So, if they have a 5800x 10-core that slots in better, it could make that part worth getting over the 5700x for multimedia work, but it seems like it would be such a relatively small difference in the lineup. At that point, it would need two CPU chiplets, and depending on how they handle binning this generation, it’ll either be dual 5-core chiplets (which, uh…huh?) or an 8-core and a 2-core, which is also a headscratcher for me.

No New Motherboard Chipset: AMD has a great, really fully-featured top-to-bottom lineup with the 500 chipsets that came out over the last year, and with no new features that would interact with the motherboard rumored, I think X570, B550, and A520 will keep on keeping on. I have some concerns about how BIOS updates would work there, but the motherboard makers made more boards last generation with BIOS flashback buttons to allow BIOS updates without a CPU socketed (or recognized) and that might help, or hell, they could have figured out a way to allow the CPU to boot without a current microcode just to get users into the system. I do expect that motherboard vendors, eager to make a buck, will refresh their lineups and offer a mix of existing designs with updated BIOSes out of the box and brand new designs made to capitalize on the new CPU lineup. For my money, more manufacturers should come out with X570 boards that don’t have little annoying chipset fans – only one board last generation made a fanless design work, and it commands a $400 price premium over its next of kin from the same manufacturer!

Staggered Rollout: The word on the street is that AMD is targeting a staggered launch, with the first lineup being an 8 and 12 core model, with 6 and 16 cores following later. This sort of sucks, as it would be nice to have the top end 16 core and the typical best mainstream choice in the 6 core out at launch, but AMD has a lot of 3950X 16 cores and 3600/3600X/3600XT 6-cores in retail today that probably need to sell through. Depending on pricing, the 3600 lineup may remain attractive as a budget choice, because…

AMD Is Coming for that Wallet: Pricing is expected to go up, to reflect the improvement in AMD’s brand perception and because they retain the value proposition against the still-higher priced Intel CPUs, especially the i9 lineup this generation, which is nearly impossible to find at retail pricing due to demand. I haven’t seen anyone put good numbers to this idea yet, but I would suspect 6-cores starting at $250, 8 cores at $350, 12 cores at $500+, and the 16-core model could be closer to $1,000, as even the 3950X 16-core beat Intel’s i9-10980XE in nearly all benchmarks, despite the Intel CPU being an 18-core. I’d also expect the X series CPUs past the 8 core to not include coolers anymore (the 3950X, 3900XT, and 3800XT already do not, but the 3900X did and everything below it did). In fact, I might even suspect the 8-core could forgo a cooler. Intel already does this with their K-SKU CPUs, and for a fair number of people building in the upper-midrange pricing tier, a separate cooler is usually on the list anyways. I personally used the Wraith Prism that came with my 2700X for my wife’s 2200G system (which definitely did not need that much cooling!) and the one that came with my 3900X is currently disassembled on my desk as a trinket for display. The coolers are cheap but removing them increases AMD’s margin, and when investors talk about AMD, their low margin on products is a chief concern for their long-term market viability. So, while they actually have excellent stock coolers that work exceedingly well, I fully expect that any part that is juiced past a 105W TDP to simply exclude the cooler and leave it up to the owner to furnish. For my part, that’s fine enough – I’m planning a 12 or 16 core build with open-loop water cooling, so that is one less bauble to pollute the planet – but that is a hit to a lot of builders, and I have seen many builds online with the Wraith Prism peeking out over the last year!

Overall? I’m really excited for this announcement and launch (he says, 2,600+ words deep into a nerdy post about a microprocessor lineup). When I first got into tech as a teenager, AMD was dominant on performance but Intel maintained their marketshare through shitty tactics even as they lost and continued to lose on performance and pricing for nearly 6 years. With Intel delaying their next-generation manufacturing and that causing huge product delays and reshuffles, this is the best chance AMD has ever had to offer meaningful competition, and another stellar lineup after the third-generation parts brought Ryzen to mainstream prominence would help cement them in the current stack as leaders. And as a competitive CPU market has benefits to the whole of technology, I am thrilled to see an actual fight brewing between these companies.

Let them fight.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

This site uses Akismet to reduce spam. Learn how your comment data is processed.