Monday, May 24th, marked the first day of COMPUTEX 2022, and with that came the AMD Keynote address, discussing more of the design and specifications of the new upcoming Zen 4 Ryzen 7000 “Raphael” desktop processors. And while much is still unknown about these new CPUs, what is known has serious implications for the coming years of both enthusiasts and system integrators.
AMD’s first point of news comes following AMD’s success in 2020 and 2021 of the Zen 3 Ryzen 5000 “Vermeer” desktop processors. After their success in improving Ryzen’s Instructions-Per-Cycle (IPC) performance, AMD announced their continued commitment to this goal, with Ryzen 7000’s Zen 4 architecture poised to offer a further 15% improvement in single-threaded performance, 2x the per-core L2 Cache of today’s current Zen 2 and Zen 3 cores, and boost clock speeds reaching 5.5GHz. (AMD’s current flagship, the Ryzen 9 5950X, brings 16 cores and 32 threads up to 4.9GHz max boost.) Zen 4’s flagship has not been formally announced, however from reading the fine print on AMD’s slide deck, we can see that there will be a 16-core CPU somewhere high in Ryzen 7000’s lineup. This strategy of focusing on single-threaded performance and IPC is a change from their early strategy in Ryzen 2000 “Pinnacle Ridge” and Ryzen 3000 “Matisse“ of engineering higher core counts, although AMD did still hint that some of Raphael’s processors will feature more cores than today’s Vermeer equivalents.
Zen 4’s clock speed and performance uplifts will be enabled by AMD’s continued collaboration with the Taiwan Semiconductor Manufacturing Company (TSMC), and TSMC’s new 5nm production node, offering even greater transistor density in exchange for lower power draw and lower heat output than the 7nm production node currently used in Zen 3. And given that Zen 3 was already renowned for its electrical and thermal efficiency, this can only be good news for gamers’ continuing battles with the fires of overclocking. However, it also appears AMD has decided to take advantage of these improvements to efficiency, and will be offering higher-powered CPUs in Ryzen 7000; while the Ryzen 9 5950X operates at a listed TDP of only 105 watts, Ryzen 7000 CPUs will operate at up to 170 watt TDP, a drastic increase in power and heat. One hopes that the performance improvements will be worth the thermal cost.
As for motherboards and Socket AM5, it was announced last year that Zen 4 and Ryzen 7000 would herald the end of Socket AM4, and thus force the creation of new motherboards to go with the new CPUs. These motherboards will follow similar patterns to AMD’s past motherboard models, however AMD have shifted the performance window up: there will (initially) not be an A620 chipset, there will be a mid-range B650 chipset, an upper-range X670 chipset, and a new overclocking-focused X670 Extreme (X670E) chipset badge. These boards will come with a range of new technologies: PCIe Gen 5 graphics support (on X670 and X670E), PCIe Gen 5 NVMe support, Wifi 6s, and most importantly, the next generation of RAM: DDR5. What was not previously known, was that Ryzen 7000 will support only DDR5 memory, and not DDR4. (Intel’s Alder Lake CPUs allow the choice between the two memory technologies.) However, given the great longevity of Socket AM4 (stretching all the way back to February 2017), the fact that even the earliest A320 motherboards can be brought forward to use the Ryzen 7 5800X3D, and AMD’s past hopes that AM5 will have a similar lifespan, it does make sense that AMD would lock AM5 to DDR5 memory support. The most unfortunate part of the new motherboards is that the product naming may become slightly amusing, as many motherboard partners already produce boards with the word “Extreme” in their name and the new X670E chipset may result in the word “extreme” appearing multiple times in the product name, much to the chagrin of anyone trying to write product reviews and articles and trying to maintain a formal and professional tone of voice.
Despite the new socket and motherboards, AMD did offer an announcement of relief and good news: Socket AM5 motherboards will use the same cooler mounting brackets as today’s Socket AM4 motherboards. This is actually very surprising given that Socket AM5 will be a Land-Grid Array (LGA) socket, LGA1718, a drastic change from Socket AM4, AM3, and even AM2’s Pin-Grid Array (PGA) designs. (In contrast, Intel’s CPUs and sockets have used LGA designs for the past several decades.)
In further good news for both gamers who like to push the limits, and for businesses looking for cost-effective solutions (an odd pairing of parties to celebrate, all Raphael CPUs will include an integrated graphics processing (IGP) chip. AMD clarified that Ryzen 7000’s initial lineup will not contain G-series APUs, and that while Raphael’s I/O die and IGP will be built using AMD’s RDNA 2 architecture and use TSMC’s 6nm process node, they are primarily intended to be used for testing purposes (helping ensure that a system can be used and tested in the face of a suspect, unstable, or failed graphics card), or for simple video output; they are not intended for use in mainstream gaming. However, AMD’s inclusion of hardware support for video encoding and decoding on these IGPs means that there are potential uses for them in video editing and production, an advantage long held only by Intel with its use of universal IGPs; and these graphics processors will easily be powerful enough to handle ordinary office tasks, allowing the construction of inexpensive or compact GPU-less office computers.
Jarringly, absent from COMPUTEX this year was Intel. Following 2021’s “Innovation at Intel” keynote, Intel elected to not give a keynote speech at this year’s COMPUTEX. While Alder Lake stunned the gaming world last year with its move to building CPUs containing both high-performance “P-Cores” and high-efficiency “E-Cores”, and the i9-12900K marked Intel’s return to dominance in desktop performance, Intel’s absence this week leaves many questions. The 12th gen Intel Core CPUs were very powerful, but that performance came at a cost: new motherboards with new cooler mounting requirements, the move forward to DDR5 memory, and no reduction in the concerns of high TDP that plagued past i9 CPUs. And with Intel not bringing a keynote to COMPUTEX to match AMD’s announcements, we are left to wonder, “what’s next from Intel?” “What will Raptor Lake bring, and how will it contest the new technologies of Zen 4?”
But until Intel’s next appearance, AMD is looking to answer Alder Lake’s strength, and Raphael is poised to make that answer.