The AMD Ryzen CPU Preview; Zen Matures

[SKYMTL]

Like many of you, the current state of the x86 processor market has me in a state of frustration. On Intel’s side that meant successive launches of parts with moderate improvements over their predecessors and, if leaks are any indication, the new Kaby Lake desktop lineup will be much of the same. It’s obvious they’ve run face first into the limits of Moore’s Law. For their part AMD has been mired in a world of lackluster APU throughput due to the woefully slow uptake rate of GPU compute into mainstream applications. Meanwhile, other than a precious few successes on the Athlon side, their own x86 processors have severely underperformed in relation to Intel’s offerings. With the upcoming Zen series and their move away from AMD’s Bulldozer microarchitecture, there’s a lot of hope that things will change in a big way.

Late last week AMD hosted a select group of journalists for briefings about their upcoming initiatives, discussing updates for everything from compute-minded GPUs for deep learning to what’s expected from their Zen CPUs. Unfortunately we can’t discuss everything here since various NDAs are set to expire between now and CES but there are some major announcements about upcoming processors and their potential competitive positioning within the broader market.

Even though it was kept to the end of AMD’s presentations, perhaps the most broadly reaching announcement was the transition away from the Zen designation we’re all familiar with. While Zen will still be used as a reference name for the underlying core architecture, the actual products themselves will be given the Ryzen designation. Naturally the term Summit Ridge will continue to be applied to the platform as a whole.

The actual thought process behind this change wasn’t really discussed but there were enough subtle hints that a pattern came into focus. AMD wanted the name of their new processors to better represent a new path for the company as a whole. The Ryzen name (pronounced Rye-Zen, not “raisin” for those of you wondering) point subtly towards these processors rising like a phoenix from the ashes of past failures on their way to take leadership positions within their respective segments.

Personally I’m on neutral ground when it comes to this change but I can’t see the Ryzen name being all that meaningful past the first generation of parts. Once the usual architectural cadence reaches its second cycle, the whole “rising from the ashes” spiel may become a bit dated. For now it will hopefully herald in a list of expectations that will be exceeded when Ryzen is launched in Q1 2017.

Speaking of dates, the launch of Ryzen represents a truly remarkable upcoming year for AMD and one that couldn’t have come soon enough. The Summit Ridge platform will be spearheading the revamped lineup with broad scale availability slated for Q1 2017 (enough though actual dates weren’t discussed I’d expect this to hit sometime in late February to early March) and that will be followed in Q2 by enhanced Zen SoC’s in the Naples ecosystem for the server market.

One of the more noteworthy changes to this generation of AMD processors is the rollout of APUs AFTER the initial enthusiast-oriented Summit Ridge platform becomes broadly available. Next generation Raven Ridge APUs which combine the Zen x86 microarchitecture and Vega-based graphics subsystems will be rolled out closer to the end of 2017. In the interim, already-launched the Bristol Ridge platform will endeavor to maintain AMD’s foothold in non-enthusiast, more affordable and notebook markets.

In my conversations with them, AMD were the first to admit Intel has taken advantage of their lack of high performance parts which can effectively compete within the enthusiast and PC gaming segments. Buyers were hungry for an alternative to Intel processors but AMD’s CPUs and their respective platform couldn’t deliver. As a result many feel that Intel have taken advantage of their commanding position by steadily increasing the prices they charge for what turns out to be middling performance gains.

Despite this unfortunate situation PC gaming hardware sales are steadily rising and this –to the joy of many I’m sure- is exactly where AMD is hoping Ryzen will fit in. It is meant to offer a high performance solution for gamers with optimizations focused on delivering framerates that meet or exceed the best Intel has to offer. That may not be an idle boast either since its one that was made several times by a company that has (for the most part) taken a back seat on the hype train in the last year.

Photoshop skills, courtesy AMD​

At this point in time AMD’s partners have already begun designing their motherboards for the Summit Ridge platform. Based on a new AM4 socket and boasting features like DDR4 compatibility, USB 3.1 Gen 2 and NVMe, there’s hope the entire Zen / Ryzen ecosystem can deliver a holistic experience. But will that really be enough? Let’s take a look at some additional features of this new architecture.

 

[SKYMTL]

Zen's Focus on Performance & Efficiency

Zen's Focus on Performance & Efficiency

While there is already quite a bit known about the Zen architecture and we will be discussing a whole lot more during our usual in-depth architectural deep dive at the launch, AMD has allowed a few additional details to escape about what we can expect from the inner workings of these new CPUs.

Let’s reiterate what’s already known: the Zen-based processors will use a 14nm FinFET manufacturing process and – on the desktop at least – will come with up to eight cores and 16 threads. Despite Intel using the technology for years, these will actually be AMD’s first products to use a symmetric multiprocessor layout. In essence, Zen is a completely new architecture from the ground up which builds upon the tough lessons AMD learned with Bulldozer and its derivatives.

Sitting behind all of the working components of Zen and many of AMD’s other upcoming designs is something they are now calling the Infinity Fabric. At its most basic, this is a structure which allows their engineers to impart a form of modularity and scalability into upcoming CPUs, APUs, SOC’s and GPUs by using a common backbone alongside an adaptable system management unit. This should allow AMD to assemble their chips in a way that defies Moore’s Law.

The Infinity Fabric was born of a need to move away from the older approach of simply shoving masses of transistors onto a chip. While we can’t get into specifics right now, think of this as a new DNA for microprocessor design that may significantly speed up the rollout of new architectures as many elements of the IP can move across product segments without requiring redesign.

The reason I bring this up now is that the ICF and new Infinity System Management Unit play key roles in overseeing Zen’s various power management features. Not only are they able to communicate in real time with the myriad number of integrated temperature, power and other sensors distributed throughout a common Zen processor but through those elements they can also control Precision Boost and Pure Power Control. Those two elements are part of what AMD is now calling SenseMI which is a suite of technologies that control on-chip behavioral algorithms.

Other than the antiquated manufacturing process being used for their production, current generation AMD processors haven’t benefitted in any way from the advancements being rolled into successive APU generations. Remember, today’s highest-end CPU the FX-9590 uses a Piledriver microarchitecture that was introduced nearly a half decade ago. Meanwhile both Steamroller and Excavator have been launched without any paralleling updates within the FX-series.

As you can imagine this has all led to Zen incorporating countless improvements in the power efficiency front. Actually I misspoke; AMD isn’t actually trying to make their next generation processors consume less power. Rather it’s all about allowing them to achieve maximum performance while minimizing power.

In order to achieve these goals the aforementioned adaptive power and clock gating technologies have been combined under two umbrella terms: Pure Power and Precision Boost. Very much like the older AMD initiatives like PowerTune and Enduro, Pure Power monitors core operations in real time with hundreds of sensors which log temperature, speed and voltage

While the Pure Power algorithm adaptively manages and logs various core functions it works hand in hand with Precision Boost to maximize operational frequencies. The “precision” part of this equation is due to the Boost algorithm’s ability to accomplish extremely fine-grain 25MHz clock speed adjustments in an effort to squeeze every last ounce of attainable performance out of the silicon.

By this point it should be obvious that every Zen-based processor will feature an effective frequency range denoted by a Base and Boost clock. However, AMD didn’t stop there and have implemented something of an ode to enthusiasts with XFR. This so-called Extended Frequency Range is supposed to scale clock speeds depending upon core temperatures and thus will reward cooler running systems with clock speed that could extend above the chip’s stated maximum Precision Boost limit.

There has been a lot of talk about machine intelligence and deep learning in the last year as scientists from all over the world over attempt to build computer networks that can think for themselves. For their part AMD has taken some of those highbrow concepts and have built an artificial network -albeit a simple one that isn’t Terminator-level smart- inside the Zen microarchitecture.

Called Neural Net Prediction, it builds a model of the decisions driven by software code execution and anticipates future needs, can pre-loads instructions and then choose the best path through the CPU for workloads. In plain English this means a Zen processor could get faster over time as it “learns” your usage habits.

Every modern processor has some form of prefetch algorithm built into its design but AMD is hoping to take this to the next level with Smart Prefetch. This is an effort to boost execution stream performance so data can be fed through the core at a faster, more efficient pace. Smart Prefetch is supposed to anticipates the location of future data accesses by applications and then utilizes a high level algorithm to learn application data access patterns and model its responses in parallel. It will then prefetches vital data into local cache so it’s ready for immediate use.

 

[SKYMTL]

Performance Expectations & Closing Thoughts

Performance Expectations & Closing Thoughts

The last half decade has been rife with unfulfilled promises and bitter heartbreak for fans of AMD processors. Bulldozer and its derivatives were decidedly lackluster, APUs completely failed to gain any momentum whatsoever on the desktop and hoped-for IPC improvements never materialized. Zen on the other hand provides a shining ray of hope for everyone who is sick and tired of Intel running roughshod, doing as they please with everyone from consumers to partners to members of the media.

At this point I wish I could be yelling “ZEN IS HERE!” from the rooftops but that wasn’t meant to be. As AMD is wont to do, they’ve once again released tantalizing glimpses into the looking glass without allowing any real hands-on time or fully yanking the veil off their upcoming products. This leaves us in a state of perpetual anticipation, like that faithful dog waiting expectedly at the door until his master comes home. Unfortunately, even pets realize when the wait is futile and I’m hoping these little dribbles of information won’t turn some potential buyers towards bluer pastures.

This time the briefings given did give us some initial feeds and speeds for AMD’s flagship Zen part which is now officially called Ryzen. It will be an 8-core, 16 thread part operating at a minimum of 3.4GHz and once fully implemented Precision Boost and XFR should (hopefully) take that to even higher levels. Alongside those already-impressive specifications is a combined 20MB of L2 + L3 cache which operates alongside the full suite of technologies I’ve described here and in other Zen-related articles.

So what does that mean for performance? Unfortunately there wasn’t all that much disclosed other than a quick set of Handbrake and Blender scene renders but you may have noticed by specifications alone the higher-end Ryzen part is being set up to compete directly against Intel’s Broadwell-E i7-6900K. There will of course be derivatives of Ryzen with lower core counts and those may become go-to parts for more budget-friendly builds.

Now I wasn’t able to time the Handbrake benchmark but the Blender scene rendered out in extremely similar times on both systems. There is however something important to note about these results: the AMD processor was able to achieve its numbers while operating at a Base Clock of 3.4GHz while the Intel system had Turbo Boost enabled so it leveled out at 3.7GHz. Meanwhile Ryzen consumed about 5-6W less on average according to my back-of napkin calculations.

These results are obviously heavily weighted towards favorable conditions and the lack of Precision Boost on the engineering sample being shown points towards that feature needing some fine tuning for launch (or this is just a matter of AMD trying to keep something under wraps). Regardless of which side of the fence you fall into these very preliminary results are nothing if not encouraging.

Naturally one of the Zen architecture’s primary focuses is upon Instructions Per Clock improvements and in that vein it is supposed to offer a massive 40% uplift when compared to Excavator cores. It accomplishes this while within a very similar power envelope.

So is there finally hope for AMD’s processors in the enthusiast market? It certainly looks that way but we also have to take all of these results with a very fine grain of salt. Launching a successful product stack takes much more than a few benchmark wins.

To truly compete against Broadwell-E, Kaby Lake and whatever else Intel have up their sleeves will take a full court press of multiple processors targeting a multitude of price points from day one. Platform support and its maturity right out of the gate will be a key element of success as well. If Ryzen can achieve that and more, then the desktop CPU market could be heading for the shakeup we’ve been waiting on since Phenom launched.