The AMD Zen 2 CPU architecture is the next generation processor design set to land on our desktops next year. It will also be the first 7nm x86 CPU to hit the market… before Intel can get its own 10nm chips out the door. Zen 2 represents a brand new, unique approach to AMD’s CPUs and promises to really turn the screw on its Intel competition, both in terms of overall core-count and in its relative performance.
AMD has confirmed a 2019 release for the new Zen 2 CPU design, and has recently unveiled some of the architectural magic which promises to make the next-gen AMD processors the best the company has ever produced. And potentially the most advanced CPUs on the market when they launch next year.
Along with the new 7nm process node, AMD is taken a modular approach, still using some 14nm silicon in the design to keep costs down and yields high. This new approach will allow the company to double the core density of the architecture, which could potentially mean 16-core chips hitting our desktops. Quite what we’d do with those extra cores in our gaming rigs, I don’t really know, but it would comfortably increase both the core-count and process node lead over Intel.
AMD isn’t stopping there, with the Zen 3 design already in the pipe, and ‘on track’ for a release sometime in 2020 on the 7nm+ node. So, if Zen 2 only has a year to live what can the next-gen architecture offer to make it a must-have processor for our gaming rigs?
AMD Zen 2 release date
AMD confirmed at November’s Next Horizon event that the Zen 2 processors had begun sampling, as promised, this year and is set to hit the market in 2019. We’re expecting the mainstream desktop parts to land in March or April 2019.
AMD Zen 2 specs
Early engineering samples have gone out into the wild, sporting the same eight-core, 16-thread design as the current top chips, but with 4GHz base and 4.5GHz boost clocks. Though the architecture does allow for the doubling of core density…
AMD Zen 2 architecture
AMD is introducing a new chiplet design, mixing the 7nm CPU silicon from the 14nm I/O chip. TSMC’s 7nm architecture alone allows for either half the power at the same performance, or more than 25% extra performance within the same power envelope, and AMD is promising up to twice the throughput of its last gen CPU.
AMD Zen 2 performance
Early performance rumours have pegged the Zen 2 processors that are currently doing the rounds as delivering a 13% IPC performance boost over and above the existing Zen+ processors of the second-gen Ryzen range. That’s on a clock-for-clock basis, so with expected higher frequency chips that IPC hike could improve a good bit more.
At the Next Horizon event AMD confirmed that it was bringing the Zen 2 processors to the market in 2019, with sampling of the early chips already in process. As this was a data centre event it was ostensibly referencing the next-gen EPYC Rome server chip, but that 2019 release window will surely extend to the AM4-based desktop Zen 2 CPUs too.
If AMD follows previous the release schedule it used for both the first and second-gen Ryzen CPU launches that would mean we could see the Zen 2 chips hitting our desktops sometime around March or April 2019.
AMD’s CEO, Dr. Lisa Su, is delivering a keynote at CES in January, and the company has historically run tech days around the Las Vegas event, so we should hear some more concrete news about its desktop plans then.
As AMD has confirmed, it has already started sampling out processors using the Zen 2 CPU architecture. That means there are early engineering samples out in the wild, and the first noises we’ve heard about a desktop-based version has come from a September post on HardOCP’s forum claiming knowledge of a chip sent to AMD Radeon Technology Group.
This chip is reportedly an eight-core, 16-thread Zen 2 CPU with a base clock speed of 4GHz and a boost frequency of 4.5GHz. Considering the top Zen+ processor of the second-gen Ryzen range, the Ryzen 7 2700X, has base and boost clocks of 3.7GHz and 4.3GHz respectively that’s an impressive start for an early engineering sample.
The first Zen engineering samples ran at 2.8GHz and 3.2GHz for its base and boost frequencies, before hitting our desktops as the 3.6GHz / 4GHz Ryzen 7 1800X. The fact that AMD is getting early silicon out of the fab and into the hands of testers and it’s running at speeds already faster than today’s best Ryzen 7 processor bodes well for the first Zen 2 launch chips.
Though the forum entry does state that it was crashing. And crashing a lot.
Given the 7nm lithography used in the CPU chiplets of Zen 2 it’s not a stretch to believe that AMD will be looking to produce final silicon that’s capable of even higher clock speeds. If it can match Intel’s 5GHz boost then the tables really could turn, with AMD potentially offering both more cores and higher performance.
Just how many cores is still up for debate. The new Zen 2 architecture means that AMD can double the core density of its processors – the new chiplet design introduced around the upcoming EPYC chips seems to indicate that AMD is moved from a quad-core CCX to an 8-core setup for its modular Zen 2 layout. Looking at the 64-core Rome chip you can see eight distinct modules, with each chiplet made of a pair of eight-core modules.
On the shrunken desktop side, it’s possible to imagine that AMD will look to reintroduce its desktop core dominance over Intel by creating a two-chiplet, 16-core Zen 2 CPU to drop into the AM4 socket. The 14nm I/O die won’t need to be the same size as that used in the next-gen EPYC server chips so could easily be cut down to fit in an AM4-sized package. That will be serious overkill for gaming rigs, but will make monstrous video and content editing/creation rigs from relatively mainstream machines.
Though might upset a few Threadripper 2 owners in the process. Though given the limited PCIe lanes on offer the X399 chipset will still be best for power users… unless the X570 motherboards get a serious PCIe update.
AMD has said it will be sticking with the AM4 socket through to 2020 – which means until Zen 3 according to the latest roadmap – so the Zen 2 CPUs ought to be compatible with the current line of desktop motherboards. That’s certainly the way it’s going in the socket-compatible Zen 2 data centre, so it makes sense for that to follow down to our gaming rigs too.
The biggest change for AMD’s Zen 2 architecture is the extension of the modular design introduced with the first Zen architecture. That layout allowed it to essentially create a CPU design that could be shrunk down for wee mobile SKUs and expanded to the desktop, and all the way up to the beefy 32-core EPYC server chips that we now have.
Zen 2 takes that further, separating the actual CPU and I/O parts of the processor into discrete modules. This multi-chip design means AMD can use individual 7nm chiplets to house the actual Zen 2 cores and connect them to a larger 14nm die looking after the inputs and outputs.
Separating it out means that AMD doesn’t need to shrink down the physical connections for memory and can use the mature 14nm process to deliver great yields, focusing the 7nm node on the CPU cores themselves. And by keeping those 7nm cores separate it allows AMD to more easily sort functional silicon and help keep prices in check for the expensive new lithography and it’s biggest, high-performance CPUs.
It also means the 7nm chiplets can be smaller than the original CCX design of Zen – a 16-core pair of chiplets from Zen 2 looks a lot smaller than the eight-core silicon used in the first EPYC CPU.
Centralising the I/O also means that – for these monstrous, many-core parts at least – each core has the same level and speed of access to your PC’s memory. With Threadripper, for example, the different cores sometimes had physically longer connections to memory, which is what gave rise to the ‘Gaming Mode’ feature. That cut Threadripper’s core-count in half to only use the cores with closest access to the memory. That won’t be a thing with the Zen 2-based version of Threadripper, but that’s probably no consolation to Principled Technologies.
Though that’s not of huge interest to us PC gamers, what is more interesting is the promise of higher IPC from the new Zen 2 architecture. AMD is improving the front end of the Zen 2 CPUs, delivering improved branch prediction and pre-fetching, and boosting its floating point performance too.
With the higher instructions per clock and doubled throughput within the same power envelope, the Zen 2 architectural improvements could potentially make a big difference to the overall frame rate performance of the chip. Combine that with higher clocks enabled by the more efficient 7nm lithography, and a potential higher core-count, the gaming versions of Zen 2 will be hard to beat.
AMD hasn’t given any real predictions for the overall performance of its new Zen 2 processors beyond the rather ephemeral ‘double throughput’ claims and the standard TSMC 7nm mantra of around 25% extra performance at the same power. But there are still some early rumours based around the pace of the new Zen 2 engineering samples.
Bits and Chips recently tweeted out about the potential 13% average IPC performance boost being shown by Zen 2 in scientific tasks compared with the latest Zen+ processors. This was taken on a clock-for-clock basis, so with higher frequency chips – made possible by the 7nm lithography – that instructions per clock boost could be even higher gen-on-gen.
And that’s what could make the big difference this generation against the Intel competition. When Zen 2 launches on the desktop – if it follows the expected March/April release time – then Intel will still be running on its old 14nm Coffee Lake architecture. That does still offer an IPC lead over AMD’s current-gen Ryzen CPUs, which is largely what’s responsible for Intel’s traditionally better gaming performance, but when both Intel’s i9 9900K and AMD’s 2700X are run at the same clock speed there is less than a 13% delta between them.
In single-core performance metrics the Intel chip only has a 5% lead, and when using all 16 threads it’s actually the AMD chip which pulls ahead. If AMD can manage to run its Zen 2 desktop CPUs at the same frequency level as the latest 9th Gen gaming chips then there’s no longer going to be any gaming lead for Intel to crow about.