Two years ago, AMD’s Zen3 Ryzen 5000 CPUs took the desktop gaming world by storm, as the last bastions of Intel superiority fell and AMD stood uncontested with the best CPUs in essentially all desktop, workstation, and server applications and workloads. A few months later, Intel attempted retaliation with the Rocket Lake Core 11000 CPUs, and stumbled as they remained on 2015-era 14nm production technology. However, one year ago, Intel’s Alder Lake Core 12000 fired back hard, regaining the top of the performance charts, but that power came at a price (or two, literally; between new CPUs and new boards they could get expensive, and the power demands were even higher than Rocket Lake’s.)
Alder Lake’s volley left AMD stumbling, releasing several lower-end CPUs and the impressive but curious and not quite in the spotlight Ryzen 7 5800X3D, and left a lot of questions for how AMD would recover itself.
As a single grain of salt, remember that all the performance numbers in AMD’s announcement are from their own benchmarks, and will need to be validated by third-party reviewers before preaching them to the masses BUT…
September 27th, 2022, marks AMD’s response. Zen4 architecture, and Ryzen 7000 desktop CPUs. TSMC 5nm production technology, specially enhanced by AMD materials engineers, PCIe Gen 5.0, DDR5 RAM support, in a new LGA 1718 socket. AMD announced four new CPUs in this new generation, matching their initial release from Ryzen 5000:
AMD are releasing Ryzen 7000 using a similar pattern to what they used for Ryzen 5000. The 7700X is at face value roughly equivalent to the 5800X.
Ryzen 7000 CPUs have significantly higher rated power draws, and thus heat outputs, than Ryzen 5000 CPUs. This is concerning, given that AMD have offered hugely better thermal efficiency and ease of cooling than competing Intel CPUs.
Single-core boost clock speeds have increased by 700-800MHz across the board, and all-core-sustained base clock speeds have increased further. Because this is a new architecture generation it can be difficult to translate that into real-world performance, but AMD have more to say about that.
Cache sizes have increased across the board, especially on the higher end. This is likely related to the engineering discoveries made surrounding AMD 3DVCache and the 5800X3D, in that in particular many games benefit hugely from increased cache sizes.
Initial MSRPs are the same, or slightly below, the initial MSRPs of Ryzen 5000. This is good because it means they aren’t trying to push at least these components to higher prices, in a time of ever dreaded inflation. However…
Much like with Ryzen 5000, the lowest-end CPU offered is at the solidly mid-grade $299 price point. This is a pattern Intel also fell into with last year’s 12000 releases, only initially offering the high-grade and mid-grade CPUs, and then several months later offering lower-priced options.
AMD have lofty claims about their new CPUs, in comparison with the previous Ryzen 9 5950X. AMD’s internal benchmarks are claiming a 13% uplift in instructions-per-clock, and a 16% uplift in peak sustained clock speeds, which combine to form an estimated 29% uplift in single-threaded performance over the 5950X. That 29% uplift is realized as an estimated 15% average framerate uplift in gaming, up to a staggering 35% framerate uplift in Shadow of the Tomb Raider. And as a result of the greatly increased all-core base clock speeds, the uplift is realized even more strongly in content creation tile-based rendering workloads, with a claimed 30% performance uplift in Blender rendering, up to a 48% uplift in V-Ray benchmarks. (It is worth noting that % performance uplifts in time-consuming production workloads are a little awkward to translate to reality because what you’re really interested in is the time cost reduction but let’s be honest half the purpose of these press conferences is to make their investors feel good and investors like it when bar goes up.) AMD have made it clear that their major design focus over the years of Ryzen has been single-thread performance, and then allowing multi-threaded performance to follow implicitly. This is a somewhat striking difference from Intel’s design philosophy of Alder Lake, with Intel’s introduction of P-Cores (for high-performance work) and E-Cores (for lower-level and more background tasks).
After touting how well they had out-performed their own previous work, AMD CEO Dr. Lisa Su went on to tout how well they are out-performing their CompetitionTM. (And by CompetitionTM we mean the Intel i9-12900K, the slides shown say that but they never say it out loud.) Her claims for gaming are much less impressive, showcasing only a 20% average uplift over the 12900K (with a few titles showing ties or even narrow disadvantages), but her claims for content creation benchmarks are even more lofty, up to a 36% Blender uplift and 62% V-Ray uplift. Given that the two-year-old 5950X still holds a small lead over the 12900K in those benchmarks, AMD are looking to expand to a very significant lead, in a territory that for years was one of Intel’s bastions. And their final parting shot before going into other territory, was to claim that all four of their newly announced CPUs defeat the 12900K in the Geekbench single-core benchmark. (As Geekbench is a purely synthetic benchmark this might not be representative of real-world performance but it is still impressive if even the 7600X can pull better single-threaded performance than the 12900K. This also completely flips the script on Intel, because last year Intel was able to make the same claim, with even the 12600K beating the 5900X and 5950X in certain workloads.)
But where AMD spent the most time and effort talking about their improvements, was in power efficiency improvements, and Performance-Per-Watt, and to talk more about both that and architectural changes, Dr. Lisa Su brought out a man named Mark Papermaster, who looks exactly like his name sounds. AMD are claiming huge uplifts in this Performance-Per-Watt metric, in the realm of just below 50% more performance than either the 5950X and the 12900K at the same power level, or being able to achieve the same performance at less than half the power draw. However here’s where the comparisons get a little awkward, and something that AMD did not speak about in their presentation. They’re claiming huge uplifts in performance, and performance-per-watt, but, these new CPUs also have greatly increased rated power draws than their current CPUs. So it’s difficult to see where that improved performance-per-watt is actually realized given that the wattage headroom is so greatly elevated. They did show several charts of the increased performance-per-watt at different power levels, but those charts are very awkward to understand because they compare at 65W, 105W, and 170W; and there are no Ryzen 5000 CPUs are rated at 170W and no Ryzen 7000 CPUs that are rated at 65W. So both the lower and upper bound charts are very unclear on what components are being compared. And on top of that, the 65W range is where the TDP improvements are most realized, but once again there is no 65W Ryzen 7000 CPU to take advantage of that. Perhaps the real-world implications of this is that Ryzen 7000 will be much more electrically and thermally efficient in low-demand workloads than Ryzen 5000 was, which would be good for long-term low-level power draw (which is something that has been very present in public, and lawmaking, conversations in recent years), but that’s a hard benchmark to quantify right now.
As for physical and architectural changes, AMD have made some very interesting achievements relative to their own past work and especially relative to Intel’s work. (This paragraph will mostly compare the AMD Zen4, AMD Zen3, and Intel Alder Lake architectures, rather than talking about CPU product names.) Zen4 is claiming an 18% smaller physical CPU die than Zen3, and a full 50% smaller CPU die than Alder Lake. That has huge implications for manufacturing, as (1) larger dies are more prone to errors and failures, and (2) fewer large dies can be cut from a single silicon boule. AMD have worked directly with TSMC on the design, materials engineering, and metallurgy of the new manufacturing processes, and so the hopes are that this will ensure plentiful supply of the new components.
Looking forward to the coming months, AMD’s roadmap includes a new line of server and cloud-optimized CPUs under the architecture Zen4c, slated for 2023, and a follow-up to this year’s 3DVCache 5800X3D, with a 7000-series 3DVCache X3D CPU. This mythical 7000 X3D has not been formally announced, but it appears on the roadmap before the Zen4c entry, which inspires hope for an early 2023 release.
After Papermaster’s speech on architecture and energy efficiency, a third speaker, David McAfee, came out to talk about how good the AM4 socket and motherboard platform were and to talk about the AM5 socket and platform. The AM4 section was a bit of self-congratulation on the (almost) six years, five generations, four manufacturing techs, 125 CPUs, and 500 motherboards, that have been covered by the AM4 socket. And AMD claim they want to do that again with AM5. The new AM5 socket is (as we know) shifting from flip-chip PGA to LGA (moving the contact pins from the CPU to the motherboard socket), the new socket is a 1718-pin (up from AM4’s 1331-pin and Intel Alder Lake’s 1700-pin LGA), with a 230W max rated package power delivery (up from 142W rated max for AM4) (although this number is in no way related to the actual upper limit for overclocking; this just tells us how high of TDPs AMD is expecting to put on their next several years of CPUs). The new AM5 platform will support DDR5 RAM only (much as AM4 supported DDR4 RAM only), and will support a mix of PCIe Gen 4.0 and 5.0 depending on the grade of motherboard. Most surprisingly given the new LGA design, existing AM4 CPU coolers will be directly compatible with new AM5 motherboards. (However, the initial line of Ryzen 7000 CPUs will not include any version of the Wraith stock cooler, meaning if you are building a new system from scratch you will need to buy a cooler as well.) And, AMD are promising to support AM5 through 2025. This is in line with their original 2017 promise to support AM4 through 2020 (which in reality was extended out through 2022, with even the low-end A320 motherboards receiving BIOS updates to support the 5800X3D). This is also a point that has set AMD apart from Intel, as Intel have for years been introducing new sockets every 1-2 years, versus AMD’s much longer socket lifespans. (For further comparison, AMD’s previous socket, AM3+, was launched in 2011, was backwards-compatible with the 2009 AM3 socket, and stood six years until AM4’s launch in 2017.)
There will be four motherboard chipsets launching alongside Ryzen 7000: B650, B650e, X670, and X670e. X670/e will launch in September directly alongside Ryzen 7000; B650/e will release a month later in October. The “e” suffix on the chipsets claims to denote that these are higher-grade boards with an overclocking focus, but in practice represent that the PCIe graphics card slot will run PCIe Gen 5.0 wiring. The non-”e” chipsets will offer only PCIe Gen 4.0 graphics slots. In the present day, this is not at all a significant drawback, as at present there are no graphics cards that operate up to the PCIe Gen 5.0 specification, and there are also none that even push against the current data bandwidth limits of PCIe Gen 4.0. So while pushing for PCIe 5.0 graphics compatibility is a good move for graphics cards that will be released in coming years (say, in 2025 maybe), it will not be a meaningful feature for some time. What will be a meaningful feature however, is the presence of one or more PCIe Gen 5.0 NVMe storage slots. There are no PCIe Gen 5.0 NVMe drives today, but there are slated to be a variety of them released in November. And all four AM5 chipsets will feature one or more PCIe 5.0 NVMe slots to use them.
For further overclocking value, Hey everyone remember the XMP and DOCP memory overclocking profiles? Well AMD have come up with a new one they’re calling EXPO (EXtended Process for Overclocking), a “one-click overclocking” solution for DDR5 RAM, and there will be an assortment of EXPO-supporting memory kits launching alongside AM5 and Ryzen 7000. The top of these kits are expecting to run at 6400MHz, matching the current high end of Intel XMP-rated memory kits.
Dr. Lisa Su returned to give us one final teaser at the end of the presentation: a prototype RDNA3 graphics card, in the testing labs, and (once again) offering a 50% Performance-Per-Watt uplift over the previous generation’s RDNA2 (Radeon RX6000 series) graphics cards, and the teaser that these cards will release Later This Year. Given that NVIDIA is expected to release RTX 4000-series cards later this year, AMD are telegraphing that they do still intend to try to compete solidly on that front, and in the same release time frame.
Now, as this is a new technology generation, there is one huge question: what does it cost to buy in? Well, let’s go through that. The Ryzen 7000 CPUs themselves are looking to be the same or lower release price as Ryzen 5000 CPUs were. However, as this is a new socket (AM5) and a new memory generation (DDR5), you cannot just buy a Ryzen 7000 CPU as an in-place upgrade. A rebuilt computer is going to be needed, with a new motherboard and memory kit as well. Fortunately, you don’t need to be alone in figuring that out: VRLA Tech is here to help, and is here to offer you the latest and greatest in computing technology today, tomorrow, and for years to come. Check out our gaming PCs now!
