AMD gives Ryzen 9000 full AVX-512 support and reveals other improvements

The release of Ryzen 9000 desktop processors is coming soon. More precisely, their sales start is scheduled for July 31st. On this occasion, AMD decided to talk in more detail about the new generation of chips.

Image source: AMD

The Ryzen 9000 processors use the new Zen 5 core microarchitecture, which AMD says is an evolution of the Zen 4 architecture found in the Ryzen 7000 processors. However, the new chips contain some internal changes that help explain the claimed performance gains of the Ryzen 9000.

The microarchitectural changes in the new Ryzen 9000 are best studied at the level of each processor core. Although there were some global changes in the new chips. For example, Ryzen 9000 switched to a more advanced 4nm process technology, which reduced the TDP for some models, as well as the overall thermal resistance of the new chips by 15%. In practice, this means that new processors are easier to cool. Their operating temperature is on average 7 degrees lower than their predecessors at the same TDP.

However, the Ryzen 9000 is in many ways similar to its predecessors, the Ryzen 7000 based on the Zen 4 architecture. The number of processing cores and the amount of cache memory have not changed. The frequencies of the new chips also remained at the same level. Still, the new generation of chips offers a noticeable increase in performance. Due to what?

According to AMD, Zen 5 uses an improved branch prediction system with dual decoding channels. This change provides higher throughput of the specified mechanism, and also helped reduce the delay in inter-core communication and improve its accuracy. The number of dispatch and execution mechanisms has been increased to 8, allowing the processor to execute more instructions per clock cycle.

To further bolster the performance gains, AMD has also increased the overall throughput of Zen 5 processors. The overall maximum L1 cache throughput has been doubled, as has the FPU throughput for each processor core. The L2 cache was moved from an 8-way to a 16-way associative design, which also effectively doubled its throughput.

AMD notes another important part of the architecture – the FPU unit, which supports the AVX-512 instruction set with a full 512-bit data path. While Intel dropped support for AVX-512 in its Raptor Lake and Meteor Lake processors, which was the subject of much heated debate between fans of both companies’ products, AVX-512 support appeared in Zen 4. In Zen 5, the company improved the FPU unit, equipping it has six new pipelines, as well as the ability to process fast addition instructions (FADD) in just two clock cycles.

Taken together, all these factors made it possible to increase the IPC (instructions executed per clock) of Zen 5 by 16%. In practice, AMD promises a percentage increase in workload performance for the Ryzen 9000 of up to 22% compared to the Ryzen 7000.

Some workloads will reportedly run even faster, especially those that can take advantage of the improved FPU. In this case, the increase in performance can be up to 35%.