• Colfax International produces novel servers with unprecedented amount of resources in a single node: up to 1 TB of RAM and four 10-core Intel Xeon CPUs
• In the paper, they benchmark and report the practical RAM to CPU bandwidth of these systems
• They also illustrate how to optimize the streaming of data from RAM to CPU for bandwidth in these NUMA systems by fixing the thread affinity

These systems can be used for problems that require a large amount of operating memory and cannot be effectively solved in distributed memory architectures. For problems requiring less resources, these machines can easily be partitioned in memory and CPU usage, and used as a cost-efficient substitute for several less powerful nodes. The following publications will present examples and cost analysis of such usage.

The White Paper is available for download here http://research.colfaxinternational.com/