Jim Keller is a well-known figure in the world of microprocessors, having played a pivotal role in the development of processors for major companies such as Intel, Apple, and AMD. As the acting CEO of Tenstorrent, Keller has recently revealed some performance, frequency, and power estimates for the upcoming Zen 5 architecture, set to launch in 2024.
Jim Keller, a former AMD chip architect who played a crucial role in developing the Zen architecture, has unveiled the next-gen Zen 5 CPU, which promises improved performance, efficiency, and integrated AI and machine learning optimization. The re-pipelined front-end will also contribute to better performance. Keller’s comments about the Zen 5 architecture have generated excitement about the future of computing.
Zen 5: A New Architecture for the Future
Jim Keller, a legendary chip architect who previously designed the Zen architecture while at AMD, has shared details about the upcoming Zen 5 architecture. Zen 5 is being developed to improve performance and efficiency, featuring a re-pipelined front-end, wide issue, and integrated AI and machine learning optimization.
According to Keller, Zen 5 is projected to be 30% faster than Zen 4, which was 15% faster than Zen 3. These estimates are based on the commonly used SPEC2K17 INT workload metric for server performance evaluation. Additionally, Zen 5 is expected to provide an 18 to 20% performance boost over NVIDIA Grace and Intel Xeon Sapphire Rapids chips, indicating that AMD continues to compete in the server market.
Power and Frequency Estimates
Jim Keller has also provided information on the power and frequency estimates for Zen 5. Tenstorrent’s projections suggest that Zen 5 CPUs will be the first to breach the 4 GHz mark on the server platform. This is an important achievement for AMD, as hitting this milestone will put them ahead of their competitors. The EPYC Genoa and Intel Sapphire Rapids families have boost clocks close to this figure, with 3.70 and 3.80 GHz, respectively. However, it seems that the Zen 5 architecture will be the first to achieve this goal unless Intel manages to exceed this with its upcoming Emerald Rapids architecture.
Regarding power consumption, the Zen 5 architecture is expected to remain the same, but the projected 30% improvement in performance could lead to even higher efficiency. This would be thanks to using a 4nm/3nm process node, which is more advanced than the one used for the previous Zen architectures.
Three Variants of Zen 5
In addition to the standard Zen 5 variant, AMD has confirmed that two additional variants, Zen 5 V-Cache, and Zen 5C, will also be released. Zen 5 V-Cache is expected to significantly boost performance due to an additional cache layer. At the same time, Zen 5C will be designed for the consumer market with higher efficiency and lower clock speeds.
What’s particularly noteworthy about Zen 5 is its power and frequency charts. Tenstorrent expects that Zen 5 CPUs will be the first to reach or exceed the 4 GHz mark on the server platform.
While EPYC Genoa and Intel Sapphire Rapids CPUs have similar boost clocks to Zen 5, the latter is projected to be the first to achieve a boost clock of 4 GHz on the server platform unless Intel can surpass it with its upcoming Emerald Rapids architecture. Despite the performance gains, Zen 5’s power metrics are expected to remain unchanged, but the 30% boost in performance may result in even greater efficiency, thanks to the 4nm/3nm process node.
This is a remarkable achievement, and it shows that AMD is committed to pushing the limits of what is possible in the CPU market.
AMD has recently confirmed that it will release three versions of the Zen 5 architecture in 2024: Zen 5, Zen 5 V-Cache, and Zen 5C. The upcoming microarchitecture is focused on enhancing performance and efficiency, boasting a re-pipelined front-end, wide issue, and integrated AI and machine learning optimization.
The Zen 5 will also include V-Cache, which will significantly increase cache capacity without increasing power consumption or die size. This development is essential for achieving higher performance without sacrificing efficiency.