AMD Instinct™ accelerator
of the MI300 series

Exponential data growth and the increasing need for real-time results are the main drivers behind advances in the industry, which is constantly reaching new frontiers of innovation to meet the diverse and complex demands of AI and HPC applications. Both AI and HPC require an architecture with high computing power, large memory and high bandwidth, but their utilization differs slightly.

For AI applications, high computational speed with less accuracy and large memory is critical to train models, make predictions and scale generative AI. In contrast, HPC applications require precise calculations and resources for processing large amounts of data and complex simulations, which are essential for scientific discovery.

The all-new AMD Instinct™ MI300 series with AMD CDNA™ 3 architecture offers customized solutions for the specific needs of AI and HPC. With the updated AMD ROCm™ 6 software platform, partners can quickly develop scalable solutions for a wide range of modern workloads, tackling some of the world's most pressing challenges in the most efficient way.

The AMD Instinct MI300 series accelerates based on the AMD CDNA™ 3 architecture and integrates matrix core technologies. This series supports various accuracies, from highly efficient INT8 and FP8 to demanding FP64 for HPC applications. The architecture enables heterogeneous integration and delivers outstanding performance through advanced packaging. A coherent programming model seamlessly connects CPUs and GPUs to provide customers with the density and energy efficiency to tackle complex problems.

The architecture is based on a heterogeneous package with up to eight vertically stacked compute chips and four I/O chips connected by the fourth-generation Infinity architecture. The integration of eight high-bandwidth stacks enables improved performance, efficiency and programmability.

The AMD Instinct MI300 series offers CPU and GPU chiplet variants to accommodate different workloads. The AMD Instinct MI300A APU combines three "Zen 4" x86 CPUs with six 3rd generation AMD CDNA architecture GPU accelerator compute chips. This APU utilizes a shared memory pool and AMD Infinity Cache™, making it the world's first high-performance hyper-scale APU.

The AMD Instinct™ MI300 series marks a significant advancement in delivering powerful solutions for the needs of AI and HPC applications.

The AMD Instinct™ MI300 series is equipped with the third generation Compute DNA architecture (AMD CDNA™ 3) and offers two different variants specifically tailored to the needs of AI and HPC applications. These accelerators deliver the compute power, memory capacity and memory bandwidth needed to train complex models, make accurate predictions and process large amounts of data for scientific discovery.

sysGen's expertise and commitment to innovation enable us to provide customized solutions that meet your needs and help you achieve your goals efficiently. Contact us today to learn more about the AMD Instinct™ MI300 series and how it can benefit your applications.