Retire in 40sSupermicro Unveils Advanced AI Data Center Solution with AMD Instinct MI350 GPU
Supermicro has launched a new data center solution featuring the AMD Instinct MI350 series GPU, supporting both liquid cooling and air cooling options.
This solution targets high-performance workloads such as AI, cloud, and high-performance computing (HPC), configured with two AMD EPYC 9005 CPUs to maximize performance and efficiency.
In fact, this product is perfect for companies aiming to achieve maximum performance in AI-driven data centers while lowering total cost of ownership.
Charles Liang, CEO of Supermicro, emphasized that the company is a leader in handling AI and HPC systems.
Notably, Supermicro's 'Data Center Building Block Solution' approach enables customers to rapidly build their desired data centers, and with the addition of the MI350 GPU, their AI solution capabilities have been further strengthened.
The H14 generation GPU servers are designed to be highly flexible and capable of handling AI, cloud, and enterprise workloads efficiently.
Simply put, these servers can be stacked like building blocks, allowing enterprises to gradually scale as needed.
AMD CEO Lisa Su revealed that the MI350 series offers around 40% better cost efficiency per token compared to competitors, where 'tokens' are units like pieces of text processed by AI models.
The form factor adheres to standards, ensuring compatibility with existing equipment.
Supermicro provides high-performance 8-GPU systems with both liquid and air cooling options to suit different environments.
The 4U liquid-cooled system features enhanced direct liquid cooling technology, designed to reduce power consumption by up to 40% while boosting performance.
To clarify terminology, 4U and 8U refer to server sizes, where 'U' stands for a server rack unit, and smaller numbers occupy less space.
These GPUs are equipped with the latest 288GB HBM3e memory, offering 1.5 times more memory than previous generations, significantly improving data processing speed and AI calculation capabilities.
FP16 and FP8 are numerical formats used in AI computations; employing these low-precision data types enables faster calculations while improving energy efficiency.
Industry analyst Paul Shell explained that as AI models grow larger and more complex, systems with such new technologies allow for fast and efficient processing.
He also highlighted that flexible cooling options make these systems appropriate choices for cloud service providers, emerging cloud companies, and large enterprises alike.
In other words, these systems deliver excellent performance not only for AI training (model learning) but also for inference (real-time processing).
The new FP6 and FP4 numerical formats are also supported, enabling effective handling of larger and more complex AI models.
Finally, these GPU server products are available for order now, offered as 4U liquid-cooled models and 8U air-cooled models.