High Performance Computing among primary use cases of GROVF RDMA
At the end of 2021, when we released GROVF’s low latency RDMA RoCE V2 FPGA IP Core for Smart NICs, we aimed to address several use cases like storage clustering and disaggregation offload, HPC application offload, algorithmic trading, database memory pooling, and more.
Although high-performance computing (HPC) was traditionally reserved for large organizations with the resources to construct these powerful computers, its ability to analyze massive amounts of unstructured data and derive meaningful business insights has made it appealing to a wide range of industries.
Companies are looking for ways to include HPC into their data operations as more computer processes shift to cloud platforms and software systems become more standardized.
Diving into the modern High-performance cluster architecture, we can see the following key features:
- Multi-core/many-core technologies
- Remote Direct Memory Access (RDMA) - enabled networking (InfiniBand and RoCE)
- Solid State Drives (SSDs), Non-Volatile Random-Access Memory (NVRAM), NVMe -SSD
- Accelerators (NVIDIA GPGPUs and Intel Xeon Phi)
One of the key futures of HPC is the data center networking technologies. Big data computing applications are built on top of MPI standard libraries which require highly parallel architecture and low latency communication
Grovf RoCE V2 networking card has extremely low latency < 2-microsecond roundtrip and high bandwidth (100Gbs) and has integration with standard Verbs API which is used in MPI.
As shown above, the Grovf RDMA supports standard Verbs API. On top of Verbs API, there can be initiated the MPI libs to build HPC AI and Big data processing applications.
Thus, GROVF RDMA over Converged Ethernet (RoCE v2) system can offer significant performance advantages over conventional send/receive network semantics and provide for high bandwidth and low latency computing for HPC application offload.