High Performance Computing - Incubation

Project Lead

Allan Cantle

The HPC project has been established in the Open Compute Project to service the needs of the High Performance Computing, Supercomputing and High Performance Data Analytics of the computing industry and to service it with open hardware platforms delivering solutions in the market from system level down to silicon.

Scope

The project will focus on Modular, Flexible & Composable Heterogeneous multi-processor systems that can scale to hundreds or thousands of processors in an energy efficient way to service the needs of the target market.

Data Centricity will be a critical factor in this project, with a focus on supporting classic compute ratio metric of Flops : Bytes/s : Bytes Depth. True composability will be demonstrated when this ratio can be easily tuned to an HPC applications specific needs.

The single parameter that will be a clear measure of success of this project will be the Application Performance Density Metric. This metric acts as a proxy for many other metrics including Efficiency and Cost with the assumption that the proposed solutions deliver on the Modular, Flexible & Composable requirements. Other submetrics will be measured that provide a lower level measurement that drives toward the be Application Performance Metric.

This project will proceed as 3 phases, defining metrics, creating white paper concepts, and building demonstrator prototypes in chronological order. Radical ideas will be encouraged including potentially testing them out to prove their viability. These ideas will be open for all other OCP Projects to observe and pull from as they see fit.

This project will constrain its efforts to within the rack level of the HPC system and it will communicate with the Rack & Power Project on any potential needs that appear to be deviating from the Rack & Power Project's efforts.

The aspect of efficient interoperability between the different heterogeneous processors and memory/storage media will also be considered in a way that contributes to the Hardware Management Project.

While leveraging from existing projects within OCP is highly desired, the HPC project shall not be constrained by them, but instead inspired by them. A particular area of focus will be HPC in the cloud and shared cloud HPC infrastructures.

Consideration as to how we can successfully transfer the technology outputs from phase 1 and phase 2 efforts to other OCP Groups.

Workstreams

• Interconnect - Mailing List
• System Management - Mailing List
• Thermal Management - Mailing List
• Emulation FPGA Module Design - Mailing List

Regular Project Calls

Every two weeks on Tuesday at 8am Pacific Time.

Call Calendar

These meeting  are recorded via audio and video. By participating you consent that these recordings may be made publicly available. Any presentation materials, proposals and meeting minutes are published on th respective project's wiki page and are open to the public in accordance to OCP's Bylaws and IP Policy. This can be found at http://opencompute.org/about/ocp-policies/. If you have any questions please contact OCP. 

OCP Server - High Performance Computing Calendar

The calendar displayed here is updated nightly from the project's Groups.io Calendar