by Paul Alcorn (tom’s Hardware)
The arrival of AMD’s 1000-series Ryzen processors (like the flagship Ryzen 7 1800X) revitalized the processor market in ways we couldn’t have imagined just a few years ago, forcing Intel to become more competitive across the full breadth of its product stack. But that was just the beginning of AMD’s assault with its new Zen-based chips. Last year AMD moved from its original 14nm chips to the 2000-series 12nm Ryzen processors (like the Ryzen 7 2700X) that come wielding the Zen+ microarchitecture.
These newer chips were a solid evolutionary step, but now AMD is working feverishly on bringing its Zen 2 microarchitecture to market packing a 7nm manufacturing process. If AMD can pull off this feat in a timely manner, while Intel still struggles to roll out its 10nm CPUs, it will mark the first time in its history that AMD has taken the process node lead from Intel.
But we don’t know a lot about what is coming with AMD’s 3000-series processors yet, or when they’ll actually arrive. Let’s take a look at what we do know, and what we don’t.
AMD has promised AM4 socket compatibility until 2020 for all its Ryzen processors. That means you should be able to use any AMD Ryzen processor on any AM4 motherboard, providing AMD’s customers with a solid upgrade path in the future. That stands in stark contrast to Intel’s frequent socket changes that find enthusiasts having to migrate to new boards and chipsets. AMD’s long-lived support for the AM4 socket has earned plenty of cachet with enthusiasts, but it also restricts the company’s options for the new 3000-series processors.
Provided that AMD releases the 3000-series Ryzen processors in 2019, the AM4 socket gives us a few hints about the possible configurations. The AM4 socket supports two channels of DDR4 memory, and that isn’t likely to change with the 3000-series models because each memory channel requires its own dedicated pins for communication. Given the current alignment of the AM4 socket, that means the new chips will likely have a dual-channel memory controller.
Assuming the Ryzen 3000-series processors will have two channels of memory also means that huge increases in core counts, like the rumored doubling to 16 cores, is unlikely. A processors’ execution cores require access to enough memory throughput to ‘feed the beast’’ with data, and doubling the cores without access to more memory channels could present performance challenges. In fact, AMD’s highest-end Treadripper, the second-generation 2990WX has just that problem because some of the cores lack direct memory access. It’s somewhat acceptable in a workstation part, but starving cores in a mainstream CPU line of direct memory access would be a bad idea, and one we suspect AMD is smart enough to avoid.
We also don’t expect a move forward to DDR5 in the near future, as we haven’t seen enough signs in the broader motherboard and memory module ecosystem to indicate there will be a wholesale move to the new standard next year. While its market share is growing, AMD doesn’t likely have the pull to drive the move to DDR5. That will likely have to come from Intel.
AMD is leading the way with PCIe 4.0 support for its 7nm EPYC Rome chips, so it wouldn’t be surprising to find the standard enabled on the 3000-series processors. However, AMD has confirmed to us that while the new Rome processors are compatible with motherboards designed for the first-gen chips, it will require new motherboards to fully support the PCIe 4.0 standard. That means that while the new 3000-series Ryzen chips may support PCIe 4.0, you’ll likely need a new AM4 motherboard designed to accommodate the more stringent PCIe bus signaling requirements. Otherwise the processors will default to PCIe 3.0 signalling rates.
The 7nm Process
AMD hasn’t said much about the 7nm process that it will use for the Ryzen 3000-series chips, though the company claimed during its debut of the EPYC data center chips that the process will bring twice the density while halving the power consumption at the same performance level. The company claims 7nm also offers 1.25x the performance at the same power. These advantages will come to the end user in the form of faster and cheaper chips, but scaling difficulties and chip-level interconnect restrictions complicate matters as chips become smaller–see Intel’s ongoing issues with its own 10nm process.
AMD’s CTO Mark Papermaster tells us to expect a 25 percent increase in performance from the new process, but he reiterated to EETimes that the move to the new process is challenging:
“Moore’s Law is slowing down, semiconductor nodes are more expensive, and we’re not getting the frequency lift we used to get,” he said in a talk during the launch, calling the 7-nm migration “a rough lift that added masks, more resistance, and parasitics.”
Papermaster also said that the move to EUV (extreme ultraviolet) manufacturing, which will come with the 7nm+ node, will only provide “modest” device performance opportunities.
It is important to remember that node naming conventions have become more of a marketing exercise than a metric based on hard measurements. So TSMC’s 7nm is not denser than Intel’s pending, oft-delayed, 10nm node. In fact, Intel’s 10nm process is actually denser than TSMC’s 7nm process. But for now, it’s a race to see who can get their new process to market first. Intel plans to have 10nm processors in high volume manufacturing in late 2019 (presuming there aren’t any more delays), leaving AMD a big window of opportunity. AMD’s 7nm data center chips will debut next year, purportedly in Q3, but the company has not announced a timeline for its Ryzen 3000-series processors.
The high up-front costs associated with developing a new node pushed AMD’s primary manufacturing partner, Global Foundries, out of the 7nm race earlier this year. AMD remains committed to delivering the new node and using it as a vehicle to deliver the new Zen 2 microarchitecture to market, but it is partnering with TSMC for manufacturing. TSMC is the industry’s premiere third-party foundry, so AMD will have to compete for wafer output with big players such as Apple, Qualcomm, and Nvidia, that also use the fab’s chip production facilities. However, recent reports indicate that the 7nm node is expensive, thus leading several large players to scale back product development on leading nodes, thus leaving ~10% of TSMC’s 7nm production capacity underutilized. This is a double-edged sword for AMD: while the company shouldn’t have any problem sourcing wafers from TSMC, the progressively higher costs of each smaller node means we likely will not see big price drops with the second-gen Ryzen chips.
The EPYC Architecture – A Precursor?
AMD’s new EPYC processors for the data center will be the industry’s first x86 processor fabbed on the 7nm node. And while it likely won’t feature the exact design as the consumer chips, AMD built its first two generations of Ryzen desktop processors on the same underlying modular building blocks used for its EPYC chips. For the first- and second-gen processors, that meant a single four-core CCX (Core Complex) design was used to construct various die, like the eight-core Zeppelin die design that powers most the company’s Zen-based chips.
The new EPYC design is even more revolutionary, though. The 64-core behemoth comes armed with nine chips on a single package. AMD is now using a second-gen Infinity Fabric to connect the multi-chip design with a 14nm I/O die that serves as the linchpin of the design. That central chip ties together eight of the eight-core 7nm CPU “chiplets,” creating a massively scalable architecture. This design helps AMD keep the areas of the chip that don’t scale well, like the memory controllers and I/O, on a proven and mature node, while also leveraging the performance, density, and economic advantages of the 7nm node for the important compute functions. The innovative design is a sign of a broader trend in the industry to heterogeneous architectures, but we don’t expect such a radical design to be used for mainstream desktop processors like the Ryzen 3000 series.
EPYC’s large 14nm I/O chip features eight DDR4 controllers and eight Infinity Fabric links. For desktop processors, that would leave a lot of unused silicon to consume area and space, not to mention add to the cost of the final unit. It’s more likely that AMD will continue with its current approach of using multi-die designs for high-end chips, like the Threadripper and EPYC processors, while using a single die for the Ryzen processors destined for the mainstream desktop.
It’s also possible that AMD could use a pared-down I/O die to tie together two eight-core CPU chiplets, thus creating a 16-core mainstream desktop model. Or the company could use the I/O die to connect next-gen Navi GPU chiplets to the processor. That would make sense given the nature of the market: Most of AMD’s current Ryzen processors lack integrated graphics. This is a market-reducing disadvantage against Intel’s mainstream chips, all of which come armed with integrated graphics that, while the aren’t great for serious gaming, can at least handle 2D tasks and multi-monitor support without requiring a dedicated graphics card.
The AdoredTV Leak
A famous Youtuber recently reported that an anonymous tipster had sent a list of AMD’s new processors. The leak claims AMD will release the new Ryzen chips in Q1 2019, but several details of the leak are questionable.
The leaker contends that AMD will unveil the new processors and pricing at CES. This is a logical venue for big AMD announcements, especially given that AMD CEO Lisa Su is hosting the company’s first CES keynote. However, AMD has been clear that its 7nm EPYC chips will come to market before the Ryzen desktop models, and the company has only painted the data center chips in broad strokes. We don’t know specific product names, core counts, frequencies, or TDP’s of any of the EPYC chips, not to mention pricing.
The company also hasn’t announced a release date, though it is widely thought to be no earlier than Q3. We expect the lack of detail about the EPYC chips, as most of these final touches, especially clock speeds and pricing, to come later in the development process.
With any possible release of the Ryzen chips likely coming after Q2 at the earliest, it’s highly unlikely that AMD would release information this specific as soon as CES. We also haven’t seen any of the normal precursors from motherboard vendors or third-party ISV’s, meaning leaks or compatibility listings, that tend to foreshadow such large launches.
To read the full click here