Introducing the Boston ANNA Pascal GPU server - with support for up to 4x NVIDIA Tesla P100 (codename Pascal) GPUs.
The revolutionary NVIDIA® Pascal™ architecture is purpose-built to be the engine of computers that learn, see, and simulate our world - making it the perfect GPU backbone for the 1028GQ-TXR and ensuring it is a contender for the title of 'world's fastest deep learning appliance'.
Powered by four ground-breaking technologies with discontinuous jumps in performance, Tesla P100 enables lightning fast nodes to deliver the highest absolute performance for HPC and deep learning workloads with infinite computing needs.
One of the most interesting features of Pascal is the new NVLink interconnect system. When using a standard multi-GPU configuration with a PCI-E switch, the limited PCI-E bandwidth to the CPU memory is shared between GPU's. The link will become even more saturated when peer-topeer GPU traffic is factored in. NVLink addresses this problem by providing a high-speed 80Gb/s interconnect path wholly devoted to peer GPU-to-GPU connections. Combining this with RDMA via Infiniband or Omnipath provides the most powerful possible parallel computing environment. As shown below with the increase in performance compared to the previous generation K80 GPUs.
Boston's industry leading server is a masterclass of server design and innovation, creating a streamlined architecture which eliminates complex cabling and GPU pre-heat for maximum airflow, cooling and performance-per-watt. Housing 3 GPUs in the front and 1 at the rear, no PCI-E extension cables or repeaters are used, instead GPU's are connected via the PCI-E Gen 3 bus providing reduced latency. In addition to the 4 GPUs, the 1U server provides 2 x PCI-E Gen 3 slots for InfiniBand or Omnipath allowing strong RDMA performance.
Key Features:
- NVIDIA Pascal Architecture
- NVLINK
- High Density
- Peer-to-Peer GPU communication