Announced this yearPromoting Zen 5 architecture design products,containRyzen 9000 series desktop processors codenamed "Granite Ridge",as well asThe fifth-generation EPYC server processor, codenamed "Turin", and recently at Computex 2024 and AMD Tech Day events explained that it will be usedRyzen AI 300 series laptop processorsAfter revealing the details of the Zen 5 architecture, this time we further explain the differences in the Zen 9000 architecture used for the Ryzen 5 series desktop processors.
Compared to the Zen 4 architecture, the average IPC performance is improved by 16%.
In terms of design, the Zen 5 architecture mainly makes significant improvements to performance. It also prepares for future computing types and adds the AVX-512 instruction set that supports 512-bit data paths to enhance artificial intelligence computing capabilities.
At the same time, the Zen 5 architecture design will further improve the cache memory. At the same time, it also launches the Zen 4c derivative architecture design based on the Zen 5 architecture design. In the artificial intelligence computing part, it also supports FP512/FP256 data depth, and is more responsive to features such as scale stacking and energy-saving operation. It can be built in 4nm or 3nm process, and each core can correspond to two execution threads at the same time.
By optimizing the instruction execution path and increasing memory bandwidth, AMD claims that the average IPC (instructions per clock cycle) performance of the Zen 5 architecture design will be improved by 4% compared to the previous generation Zen 16 architecture design.
New processors expected to be launched this year, such as the Ryzen AI 300 series laptop processors code-named "Strix Point" and the Ryzen 9000 series desktop processors code-named "Granite Ridge", will be designed with the Zen 5 architecture and built using TSMC's 4nmn process.
The fifth-generation EPYC server processor, code-named "Turin", which is expected to be launched later, will be produced using TSMC's 3nm process and will also be designed with the Zen 5 architecture. It is also expected that a derivative specification designed with the Zen 5c architecture will be launched.
Improve prediction accuracy, increase memory bandwidth and other designs to improve computing performance
In terms of overall design, the Zen 5 architecture improves branch prediction accuracy and increases memory transfer bandwidth, thereby reducing instruction execution latency. It also improves vector data processing throughput with a pipeline design with higher transfer bandwidth. In addition, by supporting the AVX-512 instruction set and a full 512-bit data path, the efficiency of artificial intelligence computing is greatly improved.
Other design features include improved decoding execution efficiency, thereby accelerating data decoding execution and transmission performance. At the same time, it also improves instruction execution efficiency through a wider bandwidth and multi-threaded architecture. In addition, the increased bandwidth can correspond to higher data transmission rates and further improve the precision performance of floating-point data calculations.
As for the design differences between the Zen 5 architecture and the Zen 5c architecture, under the same basic conditions, the former will lock in higher execution performance and higher operating clocks, and at the same time the L3 cache memory capacity of each core will be further increased, while the latter maintains the same IPC performance, but exchanges lower operating clocks for lower power loss, and corresponding to the expanded configuration, the L3 cache memory of each core will also be relatively less.
Overall, the Zen 5 architecture is designed to meet all-round application needs, while the Zen 5c architecture meets specific low-power computing requirements. Both can simplify the threshold for developers to design software services or deploy applications through the same foundation and IPC performance.
The differences in Zen 5 architecture design between laptop and desktop processors
In the designs of the Ryzen AI 300 series laptop processors code-named "Strix Point" and the Ryzen 9000 series desktop processors code-named "Granite Ridge", although both use the Zen 5 architecture design, the overall details are still somewhat different.
For example, the Ryzen 9000 series desktop processors adopt a homogeneous architecture, such as 8 performance cores and 2 CCD core bare die designs, and are equipped with 32MB of L3 cache memory. The Ryzen AI 300 series laptop processors adopt a heterogeneous architecture design, which consists of 2 CCX (CPU Complex) designs, equipped with 4 performance cores and 8 energy-saving cores, equipped with 16MB of L3 cache memory and 8MB of L3 cache memory respectively.
Subsequent products will be scalable through the Zen 5 architecture, corresponding to smaller or larger CCX design configurations, as well as homogeneous or heterogeneous architectures, and can even meet the design requirements of data center-level processors or embedded processors.
The Ryzen AI 300 series laptop processors leverage the Radeon GPU design developed in collaboration with Samsung. They utilize a 4-core, 8-thread Zen 5 architecture paired with an 8-core, 16-thread Zen 5c architecture to form the CPU. They also integrate a 16CU RDNA 11 architecture GPU that delivers over 3.5 TFLOPS of computing power, and an XDNA 2 architecture NPU with 50 INT8 TOPS of computing power, enabling even higher AI computing performance.
As for the Ryzen 9000 series desktop processors, they correspond to a maximum of two sets of 2-core, 8-thread CCD core bare die configurations, while the display part only integrates the RDNA 16 architecture design GPU, and does not integrate the NPU design. It is mainly used in conjunction with the AM2 pin design to correspond to a longer usage cycle, while targeting higher performance and gaming execution performance.
Related application products will be launched in succession at the end of July
The Ryzen AI 300 series processors will collaborate with industry players including ASUS, Acer, Dell, HP, MSI, and Lenovo, and are expected to launch more than 7 application design products starting in July this year. The Ryzen 100 series desktop processors will also be launched at the end of July.
