bullx B500 - Extreme Computing

> EXTREME COMPUTING > bullx systems >

Designed without compromise by the Bull R&D teams
to give you unlimited power to innovate

Download
white paper >

The bullx blade system was designed by Bull's Extreme Computing R&D team - Europe's largest team of experts in this field - with the following principles in mind:

Optimization and simplification of the compute node for HPC use
Integration of several compute nodes and first-level interconnect
Flexible structure for communication and I/O networks, for the closest fit with customer requirements

The bullx B510 compute blades - the second generation of bullx blades - were entirely re-designed to further optimize their compute performance and energy consumption, but they are fully compatible with the existing chassis. An existing B500-based system can therefore be upgraded on site, by simply replacing the previous blades by bullx B510 blades.

The bullx blade system leverages the latest technological advances such as:

New-generation Intel® Xeon® processors (E5-2600 Family, code-named Sandybridge-EP), delivering twice as much computing power
InfiniBand network connection through an integrated 36-ports QDR or FDR switch
Flexible storage: choice between local HDD or SSD
Availability of accelerated blades – the bullx chassis can indifferently house B510 compute blades or B515 accelerated blades, or a mix of both

Thanks to these innovations, the bullx blade system delivers:

Improved performance (enhanced compute node efficiency, reduced latency, higher communication and I/O throughput)
Reduced cost of ownership (fewer components, better fit of communication network, reduced installation time, improved upgradeability, power efficiency)
Enhanced reliability (less cabling)

More information on bullx blade system

bullx blade chassis: high density, uncompromised performance and optimized power consumption

The bullx blade chassis can host up to 9 double compute blades, i.e. 18 compute nodes in 7U. It also contains the first-level interconnect, a management unit and all components necessary to power and cool the blades, interconnect and management unit. It is the ideal foundation to build a medium to large-scale HPC cluster, with bullx compute blades combined with service nodes of the bullx R423 family.

The bullx blade chassis provides a peak performance of up to above 6 Tflops (with Sandybridge-EP CPUs) and offers a fully non-blocking InfiniBand QDR or FDR interconnect.

An ultracapacitor (option) protects the chassis from power outages up to 250ms. In areas with good quality power supplies, the ultracapacitor means a UPS is not needed – saving up to 15% on power consumption.

The chassis is equipped with high-efficiency (+90%) power supply units, and power consumption is further optimized by the smart fan control in each chassis.

More information on bullx chassis

bullx B510 compute blade

The bullx B510 is a double-width blade including two compute nodes, so as to have some components in common, such as fans, to optimize energy consumption. Each node features:

2 new-generation Intel® Xeon® processors E5-2600 Family
Up to 256 GB DDR3 memory per node
1 SATA HDD or 1 SSD drive per node

More information on bullx B510 blade

bullx B515 accelerator blade: embedded accelerators eliminating bottlenecks

Double-width blade

2 Intel® Xeon® E5 familly processors
2 NVIDIA® Tesla™ GPUs
OR
2 Intel® Xeon® Phi™ coprocessors
Dedicated PCI-Express x16 slot for each accelerator
Memory: up to 96 GB per blade
2 ConnectX adapters providing single QDR/FDR IB channel

The bullx B515 are the only blades on the market designed for full bandwidth between each accelerator and host CPU, and for double interconnect bandwidth between blades

More information on bullx B515 blade