December 5, 2022

Zen 5 in 2024 with an all-new micro-architecture

Zen 5 in 2024 with an all-new micro-architecture

Today is AMD Financial Analyst Day, the biannual gathering focused on analysts. While the primary purpose of the event is for AMD to reach investors, analysts, and others to demonstrate the company’s performance and why they should continue to invest in the company, FAD has also become a physical event for AMD’s product roadmap. After all, how can you invest wisely in AMD if you don’t know what will happen next?

As a result, the half-day series of presentations is packed with a small set of information about products and plans across the company. Everything here is top-notch — don’t expect AMD to distribute the Zen 4 transistor scheme — but it’s easy to be our best look at AMD’s product plans for the next couple of years.

Kick off FAD 2022 with AMD’s most intriguing update always being the Zen architecture roadmap. A cornerstone of AMD’s resurgence and re-emergence as a competitive and capable player in the x86 processor space, Zen architecture is the foundation of everything from AMD’s smallest integrated CPUs to its largest enterprise chips. So what’s going to happen in the tube over the next couple of years is a pretty big deal for AMD and the industry as a whole.

Zen 4: Optimizing performance and performance per watt, shipping later this year

Diving right in, AMD is currently in the process of ramping up Zen 4 architecture-based products. This includes Ryzen 7000 (Raphael) client CPUs, as well as 4The tenth EPYC (Genoa) Server CPU Generation. Both are due to be released later this year.

We’ve seen bits and pieces of info about Zen 4 so far, and the most recent one is with Ryzen 7000 announced at Computex. Zen 4 brings new CPU chips as well as a new I/O die, adding support for features like PCI-Express 5.0 and DDR5 memory. On the performance front, AMD is aiming for significant improvements in performance per watt and clock speed over its current Zen 3-based products.

Meanwhile, AMD is following up on this Computex announcement by clarifying a few things. In particular, the company is addressing questions about hourly instruction forecasts (IPC), noting that they expect Zen 4 to deliver an 8-10% increase in IPC over Zen 3. Computex’s initial announcement and demo seem to indicate that most of AMD’s performance was The gains are from Clockspeed improvements, so AMD is responding to that without showing much of their hand months after the product launch.

Alongside this, AMD also reveals that it expects an overall single-threaded performance gain of over 15% – with an emphasis on “greater than”. ST’s performance is a combination of IPC and clock speed, so at this point AMD can’t be overly specific because they haven’t locked down final clock speeds. But as we’ve seen with their Computex demos, for light-threaded workloads, 5.5GHz (or more) is currently on the Zen 4 table.

Finally, AMD also confirms that there are ISA extensions for AI and AVX-512 coming for Zen 4. At this point, the company hasn’t said if any (or both) of these extensions will be in all Zen 4 products or just a subset – AVX -512 is a bit of a space and a power hog, for example — but at least, it’s reasonable to expect it to appear in Zen 4 server parts. The addition of AI instructions will help AMD keep pace with Intel and other competitors in the short term, as CPU AI performance has already become a battleground for makers chips. Although what this does is only for AMD’s competitiveness, it will largely depend on the instructions (and data types) that are added.

AMD will produce three flavors of Zen 4 products. This includes the vanilla Zen 4 core, in addition to the previously announced Zen 4c core — a compact core intended for high-density servers that will go into 128-core EPYC Bergamo . processor. AMD is also confirming for the first time that there will be Zen 4 parts equipped with V-Cache as well – which despite the new information, comes as no surprise given the success of AMD’s consumer and server V-Cache parts.

Interestingly, AMD plans to use both 5nm and 4nm processes for the Zen 4 family. We already know that Ryzen 7000 and Genoa are slated to use one of TSMC’s 5nm processes, and that Zen 4c chiplets are set to be built on the HPC version of N5. So it’s not immediately clear where 4nm fits into AMD’s roadmap, although we can’t rule out that AMD is playing around a bit quick with terminology here, given that TSMC’s 4nm processes are a branch of 5nm (rather than an entirely new node) and typically They are categorized as 5 nm variants to begin with.

At this point, AMD expects to see a 25% increase in performance per watt with Zen 4 over Zen 3 (based on 16C desktop chips running CineBench). Meanwhile, the overall performance improvement stands at >35%, undoubtedly taking advantage of both the higher performance of the per-thread architecture, and AMD previously revealed higher TDPs (which are particularly useful for decoding more performance into workloads). MT). And yes, those are awesome graphs.

Zen 5 Architecture: All-New Microarchitecture for 2024

Meanwhile, carrying AMD’s Zen architecture roadmap to 2024 is the Zen 5 architecture, which is announced today. Given that AMD hasn’t shipped the Zen 4 yet, its detailing on the Zen 5 is understandably at a very high level. However, they also point out that AMD won’t be resting on its laurels, and they have some solid updates planned.

The big news here is that AMD describes Zen 5 as an “all-new microarchitecture”. This means that it won’t just be an incremental improvement over Zen 4.

In practice, no major vendor designs the CPU architecture completely from scratch – there will always be something good enough for reuse – but the message from AMD is clear: they’ll be making some significant tweaks to their CPU core architecture in order to further improve its performance As well as energy efficiency.

As for what AMD will reveal at the moment, the Zen 5 will redefine the front end and increase the version width again. The devil is in the details here, but coming from Zen 3 and its four instruction/cycle decoding rate, it’s easy to see why AMD would want to focus on that next — especially when you’re on the backend, the right modules already have a 10-width wide case.

Meanwhile, in addition to Zen 4’s new AI instruction, Zen 5 is incorporating more AI and machine learning improvements. AMD doesn’t say much here, but it does have an important library of tools to choose from, covering everything from AI-focused how-tos to adding support for more data types.

AMD expects the Zen 5 chipset stack to be similar to the Zen 4 — meaning they’ll have the same three designs: a vanilla Zen 5 core, a compact core (Zen 5c), and a V-Cache-enabled core. For AMD customers, this kind of continuity is very important, as it gives customers a guarantee that AMD’s most detailed configurations (Zen 4c & V-Cache) will have successors in the 2024+ timeframe. From a technical perspective, none of this is very surprising, but from a business point of view, customers want to make sure they’re not adopting clogged devices.

Finally, AMD has an interesting manufacturing mix planned in Zen 5. The Zen 5 CPU cores will be built on a mixture of 4nm and 3nm processes, which unlike the 5nm/4nm mix of Zen 4, the two 4nm TSMC nodes and 3 nm is completely different. 4nm is an improved version of 5nm, while 3nm is an entirely new node. So, if AMD’s manufacturing plans go ahead as currently shown, the Zen 5 will cross a huge leap in the node. However, it’s not unreasonable to suspect that AMD is hedging its bets here and leaving 4nm on the table if 3nm isn’t as much as they like.

Ultimately, the architecture of Zen 5 is slated for 2024. AMD is not giving any further information about when that year might be, although looking at Zen 3 and Zen 4, both have been/will be released later in 2020 and 2022 Straight. So H2/EOY 2024 is as good as guessing anything else.