When it comes to high-performance computing, expectations are higher than ever. Commercial enterprises and academic research labs need to solve complex real-world problems and conduct large-scale simulations faster than ever. At the same time, they are facing new requirements to reduce power consumption and deliver outstanding TCO. That’s why the industry is increasingly turning to AMD for solutions.  

For the second year in a row, AMD is powering the fastest supercomputer in the world as well as eight of the top ten most energy efficient supercomputers in the world according to the latest Green500 list. In fact, AMD powers 140 systems on the latest Top500 list – including the Frontier supercomputer at the Oak Ridge National Laboratory, the fastest supercomputer in the world for the fourth list in a row. Frontier is the only supercomputer that has officially passed the exaflop barrier, at 1.194 exaflops. Amazingly, it did this with only 22.70 MW of energy consumed — making it #8 on the Green500 list.  

At SC23 last week, AMD and its partners showcased the many ways that innovation is powering a range of use cases, including supercomputers and powerful technical compute clusters. Engineers from Amazon showcased how customers are powering HPC applications on AWS with the AMD processor-powered HPC7A instances. These instances are ideal for compute-intensive, latency-sensitive workloads such as computational fluid dynamics (CFD), weather forecasting or multiphysics simulations.  

Meanwhile, Eviden VP Eric Eppe was on hand at SC23 to discuss the development of an AMD Instinct MI300A-powered blade for the BullSequana XH3000 full DLC SuperComputer line and the planned delivery of the first AMD Instinct MI300A-based SuperComputer in H1-24 at Max Planck Data Facility (MPCDF) in Germany. Additionally, Snowflake Technologies CEO and Chief Scientist Sheng Nan was there to speak about accelerating molecular dynamics and drug discovery with adaptive computing and HPC clustering, powered by AMD.  

There are a couple key components to note in AMD roadmap innovation. It begins with the AMD 4th Gen. EPYC CPU formerly codenamed “Genoa”. Launched in November 2022, the CPU leverages 5nm process node technology and is the fourth generation of chiplet-based processors. With this processor family, AMD has shown it can deliver performance with high energy efficiency. As of September 2023, the 4th Gen EPYC processors have set numerous world records for application performance and efficiencyi.  

Building on that innovation, the 4th Gen AMD EPYC platform with 3D V-Cache™ technology (formerly codenamed “Genoa-X”) includes the same powerful “Zen4” cores as “Genoa”, as well as AMD 3D V-Cache technology — a groundbreaking 3D stacking technology that puts SRAM memory dies on top of processor compute dies, providing more than 1 GB of L3 cache.  For many classes of high performance technical computing workloads, the 4th Gen EPYC with 3D V-Cache platform provides even higher levels of performance and scalability than general purpose 4th Gen EPYC processors. 

While improvements in performance are very much workload dependent, this amount of cache allows for the novel effect of “super linear” performance gains in some workloadsii. In other words, when more cores (and their large caches) are added, it actually allows more of the working set to reside on chip, improving performance beyond the normally best-expected linear gains. This is an example of the kind of advanced packaging innovations that will help power the future of computing.  

4th Gen EPYC with 3D V-Cache delivers knockout performance for HPC workloads such as computational fluid dynamics, electronic design automation and automotive crash simulations and pharma drug development. AMD customers are also finding strong performance and efficiency gains utilizing EPYC 9004 with 3D V-Cache-based servers for certain demanding AI models, financial analytics and in the oil and gas industry. EPYC 9004 with 3D V-Cache based servers are available from a number of OEM vendors today. 

The AMD multi-generational roadmap is also delivering unmatched innovation in acceleration. The AMD Instinct MI300A, coming soon, takes the groundbreaking approach of combining HBM memory, “Zen4” CPU chiplets and GPU chiplets. By packaging these technologies together, the MI300A will enable ultra-fast bandwidth connections.  

Already, AMD Instinct accelerators are powering some of the most advanced computations in the world. For instance, researchers using the Frontier supercomputer are leveraging AMD Instinct 250X accelerators to advance radiation therapy, high-energy physics, and photon science. 

For more of AMD’s latest EPYC performance numbers, hot off the press at SC23, see the blog from AMD’s Raghu Nambiar: link