GIGABYTE has announced the G291-2G0, the latest addition to its Intel Xeon Scalable HPC server product family, featuring industry-leading GPU density due to GIGABYTE’s thermal and mechanical engineering expertise. The highlight of the 2U form factor G291-2G0 is the ability to accommodate up to 16 NVIDIA P4 single slot cards, the world’s fastest deep learning inference GPU to enable smart, responsive AI-based applications.
The G291-2G0 features dual Intel Xeon Scalable family processors with a TDP up to 205W, six channels of memory and 24 DIMM slots for up to 1.5 TB of memory per system, and is future ready for Intel’s upcoming Octane DC Persistent Memory (Apache Pass).
Storage wise, the G291-2G0 features a capacity for eight 2.5-inch hot swap SATA/SAS drives in the front, and an option for two additional 2.5-inch (7 mm) SATA/SAS drives to be installed internally, or one SATAIII port can be used instead for a SATA DOM module, both of which are ideal for applications such an OS boot disk.
The G291-2G0 also includes dual onboard 10GbE Base-T networking ports (via Intel X550-AT2 chipset), as well as a dedicated management port and Aspeed AST2500 controller for remote management via GIGABYTE’s GSM (GIGABYTE Server Management) software platform. Expansion wise, the G291-2G0 includes two PCIe x16 half-length low-profile slots in the rear for additional networking or storage card options.
Most importantly, the G291-2G0 has been designed specifically to support NVIDIA’s Tesla P4 accelerator, which has been labelled as the world’s fastest deep learning inference GPU for scale-out servers. The Tesla P4 is described by NVIDIA as being able to slash inference latency by up to 10X in any hyper-scale infrastructure while providing an incredible 40X better energy efficiency compared to CPUs.
The ability of GIGABYTE’s G291-2G0 to support up to 16 Tesla P4 GPU cards via four riser cards makes it ideal to be used as a high-density inference engine in the data center, and unlock a new wave of AI services such as speech recognition, image and video processing (such as an Intelligent Video Analytics Platform), or recommender systems used in online shopping and video streaming.
GIGABYTE said it continues to lead the HPC industry with products that are on-point to meet its customer’s performance needs in DNN training and inference, while also lowering OPEX through ease of management and maintenance as well as power and cooling efficiency.
– Supports up to 16 single slot GPU cards
– NVIDIA validated GPU platform
– Intel Xeon Processor Scalable Family
– Six-channel RDIMM/LRDIMM DDR4, 24 DIMMs
– Two 10 Gbps BASE-T LAN ports
– One Dedicated management port
– Eight 2.5-inch hot-swappable HDD/SSD bays
– 16 PCIe Gen3 expansion slots for GPUs
– Two PCIe x16 Half-length low-profile slots for add-on cards
– Aspeed AST2500 remote management controller
– Dual 2200W 80 PLUS Platinum redundant/hot-swap power supply
– Optimized performance with Mellanox Infiniband EDR 100G or Ethernet 100G products