Bull sequana GPU-accelerated blade

deep-learning-banner-sequana

The efficient accelerated solution for scale

Bull sequana X1115 is the sequana compute blade powered by four Nvidia® Pascal GPUs.

Don’t let your HPC be trapped in a proprietary environment

The Bull sequana approach is fully modular and GPU-centric to maintain full independence between the different functional units:

  1. no technological dependency versus CPU or interconnect,
  2. whatever the need, whatever the application, whatever the future, we offer the proper combination.

Bull sequana X1115 blade specifications

sequana_nvidia_1115Like all sequana compute blades, each 1U X1115 blade is based on a cold plate with active liquid flow, which cools all hot components by direct contact – the sequana blades contain no fan.

The Bull sequana X1115 blade incorporates two motherboards:

  1. a specific motherboard for the four GPUs, which supports the current NVIDIA® Pascal™ GPUs (P100), but is also ready for next generation Volta GPUs;
  2. a CPU board, which is currently equipped with two Intel® Xeon® E5-2600 v4 processors. More CPU options are envisaged for the future.

The Bull sequana X1115 blade offers all-to-all NVLink 2.0 connections between GPUs.

The GPUs are connected to the PCIe switches by pairs. To avoid technological dependencies, the GPUs are not directly attached to the CPUs or to the Network Interface Cards, they are connected through the PCIe switches.

Bull sequana X1115 currently supports two fast, state-of-the-art interconnects:

  1. Bull eXascale Interconnect (BXI),
  2. InfiniBand EDR.