Supermicro Unlocking Next-Gen Performance: AMD’s Instinct MI300 APU

Kena Johnson The rise of GenerativeAI, Large Language Models and HPC has increased the need for faster CPU and GPU processing. This has driven the demand for creating accelerated processing units such as the AMD Instinct MI300A APU which combines the CPU and the GPU together, resulting in reduced latencies and the sharing of memory. Daniel Lin from Supermicro discusses why partners are eager to adopt this technology that is being integrated into Supermicro’s server systems. Daniel Lin, Project/Product Manager, Supermicro “MI300A is not only a CPU, not a GPU. It's a combination of both, plus the unified memory altogether.” “So, with the integration of the CPU, GPU together on one die, plus the unified memory, you get to boost up the communications speed between, the , CPU and GPU and also minimize the latency. “So, the workload that MI300A that we're currently working is targeting at the HPC user, like the government national lab, those weather forecast modeling or health medical institution for DNA sequencing. And last but not least, the bank financial institution for the cryptocurrency or risk assessment. It's very important. All this workload require minimum latency.” “The good news is the MI300A will be installed into two different platforms and Supermicro systems. Both systems are 4 socket based, which means each system will carry four APUs. And the first one is our 2U high-density liquid cool system. With our 48U rack proprietary design, you can mount 21 systems…The second version is our 4U form factor of the air -cooled solution…. for those company that doesn't have the liquid cooled facility set up ready, they can always look for the air -cooled solution. And this air -cooled solution also give you... extra bandwidth on the storage and the PCI network.”