- Supermicro (SMCI, Financial) introduces H14 GPU solutions with AMD Instinct MI350 Series, offering both liquid-cooled and air-cooled systems.
- The new systems deliver 40% more tokens-per-dollar compared to competitors and reduce power consumption by up to 40%.
- Key specifications include 2.304TB total HBM3e memory per 8-GPU server and 1.8x petaflops of FP16/FP8 performance.
Supermicro, Inc. (SMCI) has announced the launch of its new H14 GPU solutions, incorporating the latest AMD Instinct MI350 Series GPUs. These solutions are available in both liquid-cooled and air-cooled configurations, designed to optimize performance and efficiency for AI training, inference, and HPC workloads.
The systems feature dual AMD EPYC 9005 CPUs paired with AMD Instinct MI350 GPUs, offering a significant performance boost. Each 8-GPU server provides a total of 2.304TB of HBM3e memory, with 288GB allocated per GPU—a 1.5x increase compared to previous generations. Additionally, these solutions offer an impressive bandwidth of 8TB/s and achieve 1.8x petaflops of FP16/FP8 performance.
The liquid-cooled 4U system equipped with AMD Instinct MI355X GPUs stands out for its ability to cut power consumption by up to 40%, addressing the critical need for energy-efficient AI infrastructure. By providing both liquid-cooled and air-cooled options, Supermicro ensures flexibility for various deployment environments while enhancing computational density.
Leveraging AMD’s 4th Gen CDNA architecture, the new systems incorporate new FP6/FP4 data types, improving AI capabilities. This innovation supports larger AI model deployments effectively, helping to meet the evolving requirements of AI and HPC applications.
According to AMD CEO Dr. Lisa Su, the MI350 series GPUs offer 40% more tokens-per-dollar compared to competitors, making them a compelling option for companies deploying large-scale AI models.
