With the madness of CES week more or less out of the way, I take a look at what was announced about the 3rd generation of Ryzen CPUs!
Zen 2: Bring on the 7nm
First things first, I really have to hand it to AMD for a keynote that really came out swinging. You can tell that Ryzen has been doing really well for them so far in its 14nm and 12nm configurations, and (as they’ve cracked the 7nm build process) the team were hugely positive talking about what is coming later this year.
So the above image is one that AMD have been using since the announcement of Ryzen back in 2017, and back then (in what was my second article for Logical Increments!) I talked about how Intel seemed to be scrambling to cut off AMD. 2 years later and AMD is not only sticking to its word, but Intel are genuinely struggling to keep up. That being said, Intel did announce a few new designs as well, so they’re not willingly giving up ground by any means.
One of the key things for AMD is building on their server designs and (crucially) compatibility. Before I go into some important changes below, the main thing for AMD customers is that EPYC, Threadripper and Ryzen CPUs will use the same board slot size as their respective previous generations—yet, like what happened with other generation updates, there’s likely to be a semi-optional chipset change to provide additional improvements. This means that if you want to upgrade for more cores (and/or better cooling and lower power consumption with the same core count), you can take the old CPU out and directly replace it with a shiny new Zen 2 CPU, with minimal hassle.
Why is 7nm so important?
1. Server Design
The big bottom line for AMD is high-end server rollouts. You get big data centers on-board and you can be rolling in the mega-bucks not only for the setup, but then for the years of upkeep support they pay to AMD. The demo that they used for the Zen 2 EPYC was to throw one of them against 2 Intel Xeon 8180’s (so a combined 56-core/112-thread setup at a cost of around $20,000 for the CPUs alone) and the single EPYC came out 15% faster on a really complex atomic movement simulation. Like with most staged demos like this, do take them with a grain of salt—but that sort of performance difference is what makes people take notice, especially if AMD comes in at a super-competitive price point.
2. Chiplets and Hybrid CPUs
The big question is: what is going to be different with Zen 2? Well, a lot of this comes from a more hybrid approach to chip design, something that AMD has been a little quiet about, but you can see a lot of it just off the die difference.
For context, here’s a Ryzen 7 1700 with its heat sink removed:
So your main design is based around the 2 main sections in the middle which contain your processors and cache, then (simplifying) you have a lot of additional IO, memory controllers, and so on on the smaller surrounding chips. During the CES 2019 keynote we got to see the new design:
The larger chip on the left is a crucial part of the Zen redesign, building on Infinity fabric; they now are using this chip to help handle IO & CPU communication (increases speed and bandwidth). And part of that improvement will be coming with the new PCIe gen 4 support, which is a really nice jump up from gen 3 (which I talked about a while back in my explaining data rates post). The second smaller chip on the right is the processor itself, housing a 8-core/16-thread “chiplet.” This smaller design is what AMD are aiming to use in everything from mobile devices to your large server CPU configs. As others have pointed out, it does look like there’s more than enough space below that to add a second chip, adding to the rumours that there could be a 16-core/32-thread Ryzen on the horizon.
3. Gaming and Workstation Performance
One of the biggest limitations for AMD so far has been 1080p gaming. This is where most aren’t really GPU-bound (as they are in 4K), and it’s all down to how many frames your CPU can handle. This is where Intel have been king on the single-core gaming performance for so long. But with a short Forza Horizon 4 demo AMD were averaging ~130fps with an early Ryzen 3rd-gen at 1080p. Which isn’t a million miles off an i7-8700K (av. ~150fps) or even an i9-9900K (av. ~160fps), both of which are expected to be more expensive than the new Ryzen CPUs. Now, an official demonstration involving one game isn’t much to go on, but there’s certainly something to look out for when more benchmarks are released closer to launch!
For workstation performance, we only have a quick Cinebench R15 run to go off where (again) a pre-release Ryzen 3rd-gen (8-core/16-thread) went up against the i9-9900K at stock speeds, and AMD’s offering was not only a higher score (2040 for Intel vs. 2057 for AMD) but a significantly lower power draw (180W for Intel vs. 133W for AMD).
Early Conclusions
As I said above, it’s hard during a press event like this to really get an idea of exact performance. Most things are just quick one-offs to get the important points across; but AMD have been pretty much on the mark with demos shown for previous-generation Zen CPUs, so I suspect we’re getting a good idea here of non-final samples being used for Zen 2.
Gaming seems to be looking at a solid improvement on the new hardware, and we’re expecting an impressive uptick in workstation and server performance too—all while keeping the CPUs at far lower prices than what Intel is offering.
The only other thing we really know is that these are coming this year (mid-2019), so keep an eye out for when these release because we can’t wait to get some build guides out to you for these!
Yet what do you think? Is Intel in trouble with AMD’s Zen 2? Are you thinking about upgrading? Let us know your thoughts in the comments.