July 15th, 2024

The AMD Zen 5 Microarcitecure

AMD introduced Zen 5 CPU microarchitecture at Computex 2024, launching Ryzen AI 300 for mobile and Ryzen 9000 for desktops. Zen 5 offers improved IPC, dual-pipe fetch, and advanced branch prediction. Ryzen AI 300 includes XDNA 2 NPU and RDNA 3.5 graphics, while Ryzen 9000 supports up to 16 cores and 5.7 GHz boost clock.

Read original articleLink Icon
The AMD Zen 5 Microarcitecure

AMD unveiled the Zen 5 CPU microarchitecture at Computex 2024, introducing the Ryzen AI 300 series for mobile and Ryzen 9000 series for desktops. The Ryzen AI 300 series features the XDNA 2 NPU promising 50 TOPS of AI performance and RDNA 3.5 graphics for enhanced visuals. The Zen 5 architecture brings improvements like a dual-pipe fetch and advanced branch prediction for reduced latency and increased accuracy. It boasts a 16% better IPC than Zen 4, with enhancements in data bandwidth, ALUs, and FPU. The Ryzen AI 300 series includes the Ryzen AI 9 HX 370 and Ryzen AI 9 365, offering a mix of Zen 5 and Zen 5c cores for high-performance notebooks. On the other hand, the Ryzen 9000 series for desktops features four models with Zen 5 cores, offering up to 16 cores and 5.7 GHz boost clock. These processors support overclocking and come with unlocked CPU multipliers. Both series leverage the latest Zen 5 microarchitecture, with the Ryzen 9000 series utilizing the TSMC N4 node and the AM5 platform.

Related

Testing AMD's Giant MI300X

Testing AMD's Giant MI300X

AMD introduces Radeon Instinct MI300X to challenge NVIDIA in GPU compute market. MI300X features chiplet setup, Infinity Cache, CDNA 3 architecture, competitive performance against NVIDIA's H100, and excels in local memory bandwidth tests.

AMD MI300X performance compared with Nvidia H100

AMD MI300X performance compared with Nvidia H100

The AMD MI300X AI GPU outperforms Nvidia's H100 in cache, latency, and inference benchmarks. It excels in caching performance, compute throughput, but AI inference performance varies. Real-world performance and ecosystem support are essential.

AMD Zen 4 vs. Intel Core Ultra 7 "Meteor Lake" in 400 Benchmarks on Linux 6.10

AMD Zen 4 vs. Intel Core Ultra 7 "Meteor Lake" in 400 Benchmarks on Linux 6.10

The article compares AMD Zen 4 laptops with Intel Core Ultra 7 "Meteor Lake" SoC in 400+ Linux 6.10 benchmarks. Testing involved Ryzen 7 7840HS/U vs. Core Ultra 7 155H. Insights on performance, power consumption, and upcoming releases were discussed.

AMD Is Becoming a Software Company

AMD Is Becoming a Software Company

AMD is shifting focus from hardware to software and AI experiences, tripling software engineering efforts, collaborating with major companies, and aiming to increase market share in AI PCs and data centers.

AMD Ryzen 7000/8000 Series vs. 14th Gen Intel Core CPU Performance on Linux

AMD Ryzen 7000/8000 Series vs. 14th Gen Intel Core CPU Performance on Linux

The article compares AMD Ryzen 7000/8000 Series processors with 14th Gen Intel Core CPUs on Linux 6.10 using 400+ benchmarks. Testing on Ubuntu 24.04 LTS with Linux 6.10 kernel includes various models like Ryzen 9 7950X and Core i9 14900K. Systems used DDR5-6000 memory, AMD Radeon RX 7900 GRE graphics, and NVMe SSDs. Results show performance differences and include performance-per-dollar metrics.

Link Icon 3 comments
By @adrian_b - 3 months
The original AMD slide deck can be found like this:

https://images.anandtech.com/galleries/9499/AMD%202024_Tech%...

from 01 to 23

https://images.anandtech.com/galleries/9500/AMD%202024_Tech%...

also from 01 to 23

https://images.anandtech.com/galleries/9501/AMD%202024_Tech%...

also from 01 to 23

https://images.anandtech.com/galleries/9502/AMD%202024_Tech%...

from 01 to 31

The last 2 of these 4 presentations (Papermaster & Boppan) are the most interesting, as they give details about the microarchitectures of the CPU/GPU and of the NPU.

The Anandtech article fails to link the fourth gallery and it links only downsized images in the first three galleries.

By @albinahlback - 3 months
Looks like Zen 5 will now have 3 integer multiplication ports? Hopefully that yields a reciprocal throughput of 2/3 of sequences of mulx instructions!

EDIT: With the 8-wide dispatch system, I suppose it can actually target Apple's silicon when it comes to asymptotics, no?