AMD strengthens its presence in the HPC (high performance computing) sector

Zen has been a key pillar in AMD’s recovery, that does not admit of discussion. The bet of the Sunnyvale giant for this architecture raised doubts, but these disappeared completely when the arrival of Zen 2 took place. This review confirmed that Lisa Su had put AMD back on track, and that by betting on an MCM design (module multichip) had created an architecture that was not only efficient on wafer, but also offered a great value in terms of scalability, and would allow a good level of performance to be achieved.

It is no coincidence that before the arrival of Zen, AMD’s presence in the HPC sector was very limited, and that since then, and until today, its participation in this sector has grown exponentially. To understand it, we only have to see the latest data that the Sunnyvale company has shared, taking advantage of the SC21 marking. As of today, 73 systems in the world’s top 500 supercomputers are based on AMD solutions, a figure that’s three times the number last year.

On the other hand, four of the ten most powerful supercomputers are powered by AMD CPUs, and after only eight months since the arrival of the third-generation EPYC series, this is already present in 17 of those 73 supercomputers using AMD solutions. As we had anticipated, they are very positive data that confirm, in short, that Lisa Su has mastered the helm of AMD since she took office. But this is not this, there are other details of interest that are worth taking into account, as they help us better understand the golden moment that the Sunnyvale firm is experiencing today.

Supercomputers as powerful and important as the Frontier, designed for the Oak Ridge National Laboratory, the HPE Adastra and Polaris, which enables scientists and developers at Argonne National Laboratory to tackle a number of major projects focused on science, engineering and artificial intelligence, use third-generation EPYC processors, and in the case of the Adastra we also find the new AMD Instinct MI250x graphics accelerators, the first to feature a GPU design based on two interconnected chips using the CDNA 2 architecture.

AMD’s short-term future is about memory utilization 3D stacked cache to improve performance, a novelty that, as we saw at the time, will be used in the new Milan-X processors. In the long term, we will see the debut of the Zen 4 and Zen 4c architecture, which will mean the jump to 5 nm and which will allow reaching a maximum count of 128 cores and 256 threads per processor, a figure that is currently only possible in dual socket configurations.

Related Articles

Leave a Reply

Your email address will not be published. Required fields are marked *