移至主內容

Supermicro’s Universal GPU System is the industry’s most advanced and flexible GPU server platform in the market today. Designed for large-scale AI deep learning and HPC workloads, this modular, standards-based platform supports the best available GPU technologies in a variety of form factors and combinations both today and into the future.

Supermicro 4U and 5U Universal GPU systems
  • Dual 3rd Gen Intel® Xeon® Scalable or AMD EPYC™ 7003 Series processors
  • Supports a range of industry-standard GPU form factors
  • Modular design for maximum flexibility
  • 4U chassis supports up to 10x 2.5” NVMe/SATA drives
  • Optional 1U expansion module for improved thermal capacity up to 700W GPUs and 2 additional AIOM/PCIe slots
Thumbnail for Universal GPU TECHTalk

TECHTalk: Universal GPU Systems

Check our latest TECHTalk video where Supermicro’s Josh Grossman, Principal Product Manager, shows various configurations of the Universal GPU System and the benefit of being able to standardize the platform regardless of GPU and CPU choices for AI and HPC clusters.

AS -4124GQ-TNMI

  • Dual AMD EPYC™ 7003 Series processors
  • 32 DIMMs; up to 8TB Registered ECC DDR4-3200MHz
  • 10x 2.5” U.2 NVMe or 2.5” SATA drive bays
  • 8x PCI-E 4.0 x16 slots (low-profile)
  • Optional 1U expansion module for up to 700W GPUs and 2 additional AIOM/PCIe slots
  • 4x 3000W Titanium Level redundant power supplies

SYS-420GU-TNXR

  • Dual 3rd Gen Intel® Xeon® Scalable processors
  • 32 DIMMs; up to 8TB Registered ECC DDR4-3200MHz or 12TB with Intel® Optane® Persistent Memory
  • 10x 2.5” U.2 NVMe or 2.5” SATA drive bays
  • 8x PCI-E 4.0 x16 slots (low-profile)
  • Optional 1U expansion module for up to 700W GPUs and 2 additional AIOM/PCIe slots
  • 4x 3000W Titanium Level redundant power supplies

Supports Industry-Standard GPU Form Factors

NVIDIA HGX™

NVIDIA HGX™

4 NVIDIA HGX A100 4-GPU (SXM4) with NVLink™ GPU-to-GPU interconnect

AMD Instinct™ and OAM

AMD Instinct™ and OAM

4 AMD Instinct MI250 OAM Accelerators with AMD Infinity Fabric™ GPU-to-GPU interconnect

Resources