It’s finally happened. Over a decade after Intel’s Core architecture launched and began a period of market domination that few would have predicted, competition at the high end of the desktop market is back.
AMD Ryzen—a line of desktop CPUs that will soon range from four-core lightweights to eight-core monsters like the Ryzen 7 1800X—aren’t the fastest processors in terms of pure instructions per clock (IPC). Nor does every application take full advantage of their multicore prowess. And if you’re a gamer that demands that absolute highest frame rates, Ryzen in its current state is not the CPU to buy.
But the differences in certain tasks are thin enough, and the value proposition strong enough, that for anyone thinking of building or buying a high-performance workstation or home server that doubles up as a competent gaming PC, Intel is no longer the only option.
Ryzen puts an end Intel’s monopoly on high-performance computing, and—most importantly for consumers—the price gouging that went along with it. Yesterday, an eight-core, 16-thread (8C/16T) desktop chip cost $1000/£1000. Today, it’s just £500.
Table of Contents
What is Ryzen?
At the top of the Ryzen stack is the flagship Ryzen 7 1800X processor. For £500 you get a chip with 8C/16T, 3.6GHz base clock, and 4GHz boost clock. It’s joined by the £400 R7 1700X, which has the same core/thread configuration but lower base and boost clock (3.4GHz and 3.8GHz respectively), and the R7 1700, a $330/£320 part (the same price as Intel’s Kaby Lake i7-7700K) with clocks of 3.0GHz and 3.7GHz.
All of the top-end Ryzen 7 chips feature 24 PCIe lanes, with certain motherboards adding up to eight more, and all support modern features like dual-channel DDR4 memory, PCIe X4 storage, and USB 3.1 Gen 2.
|Specs at a glance||AMD Ryzen 7 1800X||AMD Ryzen 7 1700X||AMD Ryzen 7 1700||Intel Core i7-6590X||Intel Core i7-6900K||Intel Core i7-5960X||Intel Core i7-7700K|
|CPU PCIe Lanes||24||24||24||40||40||40||16|
|Chipset PCIe Lanes||Up to 8||Up to 8||Up to 8||8||8||8||Up to 24|
The 1800X is based on Zen, a CPU architecture first teased in 2015 by AMD CEO Lisa Su as featuring “competitive, high-performance cores,” with a vastly different direction and design (over and above the new 16nm FinFET manufacturing process) to the much lamented Bulldozer architecture that the company has been saddled with since 2011. In some ways Zen, and by extension Ryzen, has embraced much of what has made Intel’s processors since the Core 2 line so successful, including simultaneous multithreading (SMT, called Hyperthreading in Intel-land) across independent cores and a dedicated micro-op cache.
But with Ryzen, AMD has thrown in a few innovations of its own. By increasing the amount of low-level cache close to the CPU with 256KB L1 cache, 4MB L2 cache—doubling that of Intel’s Broadwell-E—and 16MB of L3 cache, AMD can feed the Zen cores more instructions more quickly for improved single-threaded performance. Then there’s “SenseMI” (Machine Intelligence), a cringe-worthy marketing term that brings together many of the engineering tricks AMD deployed in its attempts to rescue Bulldozer, which culminated in the surprisingly capable Bristol Ridge APUs of 2016.
My formerly hirsute colleague Peter Bright has a full deep-dive into the Zen architecture (and I highly suggest giving it a read if you’re at all curious about chip design), but at a high level, SenseMI is simply five different features of Ryzen that focus on power, clock speeds, and shuffling data around as quickly as possible. The first, “Pure Power,” is a set of temperature, clock speed, and voltage sensors (accurate to 1mA, 1mV, 1mW, and 1°C) that promise more efficient power delivery to the CPU. The second is “Precision Boost,” which takes the sensor information and uses it to adjust the clock speed on-the-fly in small 25MHz increments.
These 25MHz increments are smaller than what you typically see in a CPU, and are more akin to the aggressive clock gating and power management features of a GPU. The smaller clock speed increments allow for more accurate, and more frequent adjustments to be made, in theory saving power. Indeed, Ryzen’s 95W TDP looks impressive compared to the 140W TDP of Intel’s 6900K, a similar 8C/16T processor based on the Broadwell-E architecture. But in practice, Ryzen pulls just as much wattage from the wall as the 6900K, in some cases slightly more. Considering just how power-hungry Bulldozer was, that remains a significant achievement, even if it means that overclocking headroom isn’t as high as some might have hoped.
The multitude of sensors in Ryzen also allow for a feature called Extended Frequency Range (XFR). XFR is an extension of Ryzen’s 4GHz boost clock, giving users with a good cooling solution—including AMD’s own RGB-laden “Wraith Max” air cooler—an extra 100MHz boost without manual overclocking. The caveat is that XFR only works across two of the Ryzen cores at any one time, rather than all eight. With the exception of highly single-threaded tasks—which does admittedly include a good few games—you won’t see the difference most of the time. XFR is only fully enabled on Ryzen “X” parts like the 1800X. Non-X parts like the R7 1700 are limited to a smaller 50MHz boost.
Manual overclocking (which sadly disables XFR) is supported across the entire Ryzen range, and is limited by motherboard rather than chip (there’s more on motherboards below, but tl;dr: you need an X-series or B-series board for overclocking). For traditional BIOS/EFI overclocking the process will vary depending on the motherboard maker, but for the most part there’s access to per-core clock speeds as well as an overall voltage control, just as with Intel chips.
There’s no support for XMP (eXtreme Memory Profile), since it’s an Intel standard, but Asus motherboards translate XMP over to a compatible compatible AMD (AMP) form shown as “DOCP” in the BIOS. While not supported by default in Ryzen processors, certain Asus and Gigabyte motherboards also enable baseclock (BCLK) overclocking thanks the addition of a third-party clock generator. According to Asus the OC limit tends to be around 106MHz, a 6MHz boost, before PCIe devices stop working correctly.
You also have the option to overclock using AMD’s new software-based overclocking tool called “AMD Ryzen Master.” It’s similar to the WattMan software for AMD graphics cards, with a set of sliders that allow for per-core clock speed, voltage, and memory timing adjustments, as well as the ability to disable cores entirely. Like BIOS overclocking, Ryzen Master also disables Precision Boost and XFR, but keeps low-power C-states (idle power saving states) active.
Due to Ryzen arriving rather late in the Ars Technica offices (thanks, AMD!), I haven’t fully tested Ryzen overclocking. AMD claims most 1800X chips will hit 4.2GHz at a voltage of 1.4V. That’s higher than the 1.35V typically needed to get a Broadwell-E CPU to 4.4GHz, but not dramatically so, and as always with CPU overclocking some chips will fare better than others. That said, 4.2GHz seems to be the typical overclock for the 1800X, as offered by retailers like Scan, which is selling pre-overclocked systems, and it’s unlikely to go much higher. Multiple publications I spoke to with earlier access to chips all report that 4.2GHz is the current limit with standard air or sealed liquid coolers.
Still, such speeds would make the £320 R7 1700 an enormous bargain, should it overclock to 1800X speeds.
But the many sensors AMD has included in Ryzen aren’t just for XFR and power saving: they’re for binning chips at the factory too. All 7-series Ryzen chips begin their lives as an 1800X, but only some of them reach the right clock speeds within the 95W TDP. I’m as curious as everyone else to see how good those cheaper chips are, but I wouldn’t get your hopes up just yet. The best result so far for the 3.4GHz 1700X (AMD is shipping other chips in the range to reviewers at a later date) is the 3.8GHz system sold by Scan.
The final two parts of Ryzen’s SenseMI—”Neural Net Prediction” and “Smart Prefetch”—help to shuffle data through the CPU more effectively. AMD claims there’s a “true artificial network” and “learning algorithms” inside Ryzen that are able to predict what an application might do next, pre-load instructions, and learn application data access patterns in order to improve performance. These improvements, coupled with others likes the “Infinity Fabric” design, have resulted in a 52 percent increase in IPC over Bulldozer-based designs—and, in CPU-focused tests at least, the benchmarks confirm it.
When the first details on Zen were dropped, AMD said it was aiming for a 40 percent uplift in IPC. Regardless of whether it was smart marketing or a genuinely unexpected result, that extra 12 percent increase in IPC means Ryzen performs better than Intel’s massively expensive multi-core Broadwell-E line in a range of workstation and productivity tasks, and better than Intel’s Kaby Lake 7700K, provided the apps you use can take full advantage of all eight cores.
That makes Ryzen a compelling processor, especially considering AMD’s aggressive pricing. However, one thing to note about the current Ryzen line-up is that none of the CPUs are equipped with a built-in GPU, so there’s no built-in support for video decoding and the like. Intel’s chips with more than four cores don’t feature a GPU either, but it does mean that if you opt for a Ryzen 7 1700 instead of an Intel i7-7700K, you may need to buy a graphics card as well. An APU version of Ryzen that incorporates a GPU is promised for later in the year.
It’s also worth noting that even with a suitably equipped motherboard, Ryzen’s maximum 32 PCIe lanes—16 for graphics, four for NVMe, four to communicate with the chipset, and an additional eight from the motherboard—is less than the 40 of Broadwell-E and Kaby Lake. It’s not a deal breaker right now, particularly as SLI and Crossfire aren’t as appealing these days, but the rapid proliferation of PCIe-based storage means PCIe lanes are quickly becoming a prized commodity.
Listing image by Sebastian Anthony