For the last 5 years computer hardware enthusiasts have been eagerly awaiting the arrival of AMD’s next generation CPUs. In a few week AMD will announce the launch of their Ryzen family of processors. In this article we will provide an in-depth review of AMD’s high performance Zen CPU microarchitecture, its features, specs, and performance.
Ryzen: The most significant products by AMD in more than a Decade
A Decade in Development
The last time that AMD was competitive with Intel in Chipset Architecture was with the development of their Athlon XP and Athlon64 processors. Since then Intel has vastly improved and been at the forefront of CPU development. In 2012 AMD started development on the architecture for the Ryzen family AMD decided to start from a new clean slate design. The design team was let by CPU architect Jim Keller (responsible for the Athlon Architecture) who developed AMD’s first simultaneous mulithreading CPU in four years time. For the first time since the 2000’s AMD’s CPUs will not be disadvantaged by Intel’s design lead. AMD’s long-term strategy does not only rely on high performance desktop CPUs. They are also simultaneously developing 32 core Zen server CPU, a 16 core Zen HPC APU, and a Quadcore Zen consumer APU called Raven Ridge.
The Zen Mircoarchitecture
The Zen processor core consists of one integer engine and one floating point unit. Each integer cluster contains 6 pipes, 4 ALUs (Algorithmic Logic Units) and 2 AGUs (Address Generation Units). This is a huge step up from the Bulldozer architecture which featured 2 integer engines and 1 floating point per core. These AGUs perform two 16-byte loads and one 16-byte store per cycle via a 32 KB 8-way set associative write-back L1 data cache. This move from write-through to write-back cache has reduced lags in several code paths. Lots of work has gone into increasing the power efficiency and area footprint as compared to the bulldozer architecture. Below is a visual representation of the Zen core on silicon.
The L2 and L3 chaches have been cleverly regrouped to reduce access times and a change from write-through to write-back cache has improved the power and size efficiency of the Zen Cores. Another major difference from the Bulldozer architecture is that the Zen chips have access to twice as much L3 cache and the floating point performs FMAC operations or a single 256-bit AVX operation per cycle.
The First Simultaneous Multithreading Microarchetecture by AMD
The Bulldozer module used CMT (clustered multithreading) and executed two identical threads through the utilization of two seperate integer clusters with a single front end. For the Zen architecture AMD has gone back to the more traditional SMT (simultaneous multithreading). The advantage here is that AMD can build a larger integer cluster with higher single thread performance. A secondary smaller thread is also available for use as needed. This SMT design allows for clever space and power saving options.
AMD’s Ryzen 8 core processors are loaded by sensors that are monitoring voltage, temperature, frequency and power hundreds of times a second. This technology allows the processors to be auto overclocked using the XFR feature. A major design flaw that has the Bulldozer series suffered from is latency caused by the deep pipelines in the architecture. To combat this Zen front end of each core is capable of handling 4 instructions per cycle. And a micro-op cache has a throughput of 6 instructions per cycle.
|CPU Microarchitecture||AMD Phenom II / K10||AMD BD/PD||AMD SR/XV||AMD Zen||Intel Skylake|
|Instruction Decode Width||3-wide||4-wide||8-wide||4-wide||4-wide|
|Single Core Peak Decode Rate||3 instructions||4 instructions||8 instructions||4 instructions||4 instructions|
|Dual Core Peak Decode Rate||6 instructions||4 instructions||8 instructions||8 instructions||8 instructions|
|AMD Ryzen CPU||Cores/Threads||L3||TDP||Base||Turbo||XFR||Price|
|AMD Ryzen 7 1800X||8/16||16MB||95W||3.6GHz||4.0GHz||4.0GHz+||$499|
|AMD Ryzen 7 1800 Pro||8/16||16MB||65W||TBA||TBA||N/A||TBA|
|AMD Ryzen 7 1700X||8/16||16MB||95W||3.4GHz||3.8GHz||3.8GHz+||$389|
|AMD Ryzen 7 1700||8/16||16MB||65W||3.0GHz||3.7GHz||N/A||$319|
|AMD Ryzen 5 1600X||6/12||16MB||95W||3.3GHz||3.7GHz||3.7GHz+||$259|
|AMD Ryzen 5 1600||6/12||16MB||65W||TBA||TBA||N/A||TBA|
|AMD Ryzen 5 1500||6/12||16MB||65W||3.2GHz||3.5GHz||N/A||$229|
|AMD Ryzen 5 1400X||4/8||8MB||65W||3.5GHz||3.9GHz||3.9GHz+||$199|
|AMD Ryzen 5 1400||4/8||8MB||65W||TBA||TBA||N/A||TBA|
|AMD Ryzen 5 1300||4/8||8MB||65W||3.2GHz||3.5GHz||N/A||$175|
|AMD Ryzen 3 1200X||4/4||8MB||65W||TBA||3.4GHz||3.8GHz||$149|
|AMD Ryzen 3 1200||4/4||8MB||65W||TBA||TBA||N/A||TBA|
|AMD Ryzen 3 1100||4/4||8MB||65W||3.2GHz||3.5GHz||N/A||$129|
|PCIe 3 Lanes||24||24||TBA||TBA|
|PCIe 2 Lanes||8||6||TBA||TBA|
|USB 3.1 Gen2||2||2||TBA||TBA|
|USB 3.1 Gen1||8||4||TBA||TBA|
|Form Factor||ATX||ATX, M-ATX||M-ATX, Mini-ITX||Mini-ITX|
At the CES AMD made a big deal that all Ryzen CPUs have unlocked frequency multipliers to allow overclocking easily. This means that customers can pair any of the 17 Ryzen CPUs with either a B350 or X370 AM4 motherboards. Intel, meanwhile, only makes 3 CPU models that are unlocked for overclocking. AMD’s amazing XFR (extended Frequency Range) feature automatically increases the default clock speed whenever the thermal environment allows. This mean that if you get an aftermarket cooling system then the processor will automatically overclock itself.
AMD has clearly been working hard over the past few years to improve their processors while at the same time keeping a long-term vision in place. Soon consumers will be able to opt for AMD processing for their desktops,notebooks, and servers. Finally since the mid 2000’s AMD stands a good chance to compete in performance with Intel. After the announcement in March we will finally be able to determine whether AMD will once again be the benchmark for enthusiasts.