AMD Ryzen 7 1800X Review

👤by Tony Le Bourne Comments 📅01-03-17
Features - Ryzen
Ryzen is AMD’s brand new desktop line of CPUs based on the Zen microarchitecture. Draw a line under that, it's important. The principles of Ryzen's design will also be applied to new APU and workstation-class CPU designs in the coming months and years, but for now the brand is being used solely for their Desktop CPU line which do not feature on-die graphics components. These CPUs, based on a brand new architecture differing substantially from the Bulldozer-derived parts of the past half decade, are manufactured with Globalfoundries’ 14nm FinFET process.

The Ryzen desktop line is split broadly into three categories: Ryzen 3, for entry-level and mainstream systems; Ryzen 5, for performance mainstream and gaming systems; and Ryzen 7, an enthusiast and prosumer category. This echoes the nomenclature of Intel’s Core i3/i5/i7 product range, a choice which is intentional on AMD’s part. The launch Ryzen CPUs are all only Ryzen 7 models; Ryzen 3 and 5 are expected in Q2 2017.



Ryzen 7’s launch comprises three SKUs, the Ryzen 7 1700, 1700X & 1800X. All are 8-core parts which support Simultaneous Multithreading (SMT) for a total of 16 threads.

Ryzen Fundamentals - The CPU Complex (CCX)


AMD Ryzen 7 CPU Die


AMD’s Bulldozer architecture was a modular design made up of Compute Units, one large core coupled to a smaller one with a shared cache, and that’s one of the fundamental aspects which has changed with Zen. The new architecture defines a CPU Complex, or CCX, which is four fully functional cores connected to a L3 Cache 8MB in size (16-way associative) which can be accessed at full speed (and with similar averaged latencies) by any core. Furthermore each core has 512KB L2 cache (8-way associative), double the size of Intel's Skylake and Broadwell-E L2 cache.

Ryzen L3 cache is a pure Victim Cache of the L2 rather than the write-back type present on Intel's desktop CPU products.

Clearly one of the advantages of the CCX structure is its scalability – quad-core Ryzen 3’s will include just the one CCX, 8-core Ryzen 7 CPUs feature two CCX modules, and the upcoming Naples server designs are set to be equipped with many more. However fine control is also possible as AMD include tools to disable sections of the CCX on a per-core basis, allowing not only 6-core configurations but also single-core modes for competitive overclocking.


AMD Ryzen CCX Block Diagram


IPC Improvements

An oft-criticised aspect of the Bulldozer architecture was poor Instructions Per Clock performance, requiring the part to operate at very high frequencies in order to remain competitive. Staying ahead of the field by using a more advanced lithographic process may have kept them in the game, but remaining at 32nm for too long and only transitioning to 28nm relatively late gave Bulldozer the reputation of a hot, slow architecture. Improving IPC became a core design goal early on for Zen.

AMD’s target was a minimum of 40% improved IPC vs. Piledriver (2nd Gen Bulldozer), an extremely ambitious target. Thanks to the revolutionary design of Zen, and a transition to the Samsung/Globalfoundaries 14nm process, AMD are claiming an average IPC improvement of over 51%. That’s huge whichever way you want to shake it, and goes to show how much of a misstep Bulldozer turned out to be.

Changes to the cache structure and implementation of Simultaneous Multithreading, as well as integration of new instructions, have been key to IPC improvements. These come at a price however: increased die size, potentially impacting yields and pushing up manufacturing costs.

Higher IPC has also allowed reduced operational frequencies compared to Bulldozer (whilst still hitting a very healthy 3.6GHz+ on high-end SKUs), and sub-100W TDP envelopes. It will also be interesting to see how well the chip can overclock, another core expectation of the enthusiast market.

Clock Speeds and XFR

In common with the vast majority of CPU models released in the last 10 years, AMD have explicitly defined base and boost clocks for their Ryzen 7 CPUs. These indicate the base level of performance whilst under heavily multi-threaded load, and peak clock acceleration whilst under a load suitable for two cores. Given that all Ryzen 7 SKUs feature eight cores with SMT, clock speed is the key differentiating factor between retail models.



AMD Precision Boost takes this a step further by continually adjusting clock speeds based on workloads and SenseMI sensor data. Utilising 25MHz steps the CPU can vary between base and boost frequencies, with key levels for fully loaded (8-core, 16-thread) and infrastructure (power draw) limited operation. Meanwhile, AMD Pure Power downclocks the CPU in low-load/idle operation; critically all this can be performed quickly thanks to improved sensor data collection.

Muddying the water somewhat is a new feature AMD are calling Extended Frequency Range (XFR). Sensor data (esp. temperatures and voltages) gathered by SenseMi is analysed to determine whether additional headroom is available above the Boost clock speeds. The CPU will then clock to a higher state beyond boost clock levels, providing just a little more performance.

The XFR boost levels are indicated to consumers by the CPU SKU. Those with feature an ‘X’ suffix are equipped with a 100MHz XFR, whereas those without have an XFR of half that. It’s worth noting that when user overclocking tools are enabled Boost and XFR frequencies are ignored, so they’re only applicable to stock clocked configurations.

Simultaneous Multithreading (SMT)

Intel’s proprietary Hyperthreading technology has been core to their CPU design for fifteen years now, parallelising computations such that multiple threads of instructions are processed in a more optimal manner on each x86 CPU core. It has become a key selling point for Intel’s premium SKUs, but arguably also resulted in a stagnation of core counts such that entry-level chips are often still limited to two physical cores. In fairness, only relatively recently have applications begun to exploit these capabilities; notably game engines tended to rely on strong single-core performance rather than evenly distributed multi-core processing.

Although proprietary, Hyperthreading is an implementation of a more general technique known as Simultaneous Multithreading, where two threads with broadly shared resources can be processed at the same time. Bulldozer featured partial SMT - the integer cores were single-threaded whilst other sections were multi-threaded – but Zen implements 2-way SMT (or two threads per CPU core).

The benefits of SMT are two-fold. Firstly threads can share resources, notably caches, improving performance compared with sequential processing. Secondly, the more optimal use of processor resources can significantly improve power efficiency, a key factor for enterprise markets in particular.

Thanks to the implementation of SMT across the Ryzen range, application developers can be reassured that hardware resources are available to take advantage of multi-threaded design even at an entry level. It will be interesting to see how, if at all, Intel respond to this move by AMD.

Memory Support



AMD Ryzen on the AM4 platform supports dual-channel DDR4 DIMMs up to a maximum frequency of 2667MHz according to JEDEC standards. AMD representatives indicated that while the maximum memory capacity possible is 128GB, the cost of 32GB DIMMs effectively limits support to 64GB or 32GB in two-DIMM configuration.

It should come as no surprise that Ryzen does not support XMP, Intel’s proprietary extension to the JEDEC SPD information resident on RAM DIMMs. Consequently additional memory timings will need to be input manually, but this will also allow support for frequencies beyond the stock 2667MHz defined by JEDEC.

Although Ryzen doesn’t support quad-channel memory configurations, this reduced bandwidth compared to Intel's X99 HEDT platform may not be quite as significant as it might seem. Applications which can exploit this bandwidth effectively are quite uncommon, and may not be a factor in typical desktop workloads.

Cooling

Ryzen 7 1800X and 1700X CPUs are 95W parts, significantly lower than Intel’s Core i7-6900K which sits at 140W TDP. Perhaps more impressively, the eight-core Ryzen 7 1700 has a listed TDP of only 65W, lower even than the quad-core Kaby Lake Core i7 7700K’s 91W. As each are on a similarly modern 14nm lithographic process it shows just how aggressively AMD have targeted improving TDP with the Zen architecture.

Both 1800X and 1700X are enthusiast designs for performance-oriented systems with bespoke cooling, and as a result are not equipped with an AMD stock cooler. It’s expected that users will make use of 3rd party coolers bought for the new systems, such as high-end air cooling or water cooling. If both can hold to their 95W TDP the options available are wide-ranging, limited only by AM4 support.

The Ryzen 7 1700 meanwhile is shipped with AMD’s Wraith Spire cooler, an evolution of the Wraith Stealth introduced last year to mainstream desktop systems. Among the changes in this generation are the spring-screw mounting system, a lower noise fan, and RGB lighting (which can be controlled via RGB headers present on many new motherboards).

PCI-Express

Ryzen marks AMD’s transition to PCI-Express 3.0, double the bandwidth of PCI-E 2.0. Ryzen 7 CPUs make available 24 PCI-E 3.0 lanes to the motherboard, 16 of which are reserved for the GPU and the rest distributed among other motherboard features and peripheral devices.

PCI-Express 3.0 support hasn’t come too soon for AMD. Although for years GPUs didn’t chafe under the limitations of x16 PCI-E 2.0, larger textures and 4K resolution gaming has stretched the 4GB/s per lane available under the spec. PCI-E 3.0 offers twice the bandwidth per lane, greatly increasing headroom as desktop computing transitions to a world were high resolution (4k and Virtual Reality) is ever more commonplace.

Also notable is (chipset-dependent) support for NVMe storage, typically via a single M.2 slot. Depending on system configuration this may entail up to x4 NVMe support (four PCI-Express lanes) or x2 with additional peripheral connectivity.

In Summary

Ryzen is, in all aspects, a substantial technological step forward for AMD which should set them up for competitiveness throughout the product stack. By all accounts they have addressed both IPC and TDP weaknesses, lining up competitive advantage over Intel even at the performance end of the market, and rolled in feature improvements necessary for modern gaming. Although the launch lineup consists only of eight-core Ryzen 7 CPUs, competing at the entry-level and mid-range will be critical to claw back market share lost over the last few years.


20 pages 1 2 3 4 > »

Comments