Last year, IDC predicted that the HPC market would reach $15.2 billion by 2019. That represents a healthy 8.2 percent yearly growth rate. We knew it would be big, and we knew we would be ready for it. In fact, we’ve spent the past decade striving to take high performance computing (HPC) to new levels. We’re fully committed to adding new and higher performance functionality to our smart interconnect product lines, most recently with the launch of ConnectX-5 – the industry’s most advanced 10, 25, 40, 50, 56 and 100Gb/s InfiniBand and Ethernet intelligent adapter.
What is the new technology being delivered with ConnectX-5?
The ConnectX-5 delivers the highest available message rate of 200 million messages per second. And unlike competing products on the market, ConnectX-5 is leading in terms of performance, delivering the highest available throughput (aggregated throughput of 200Gb/s). Here again, Mellanox is leading the marketing with performance that is unmatched. It is also the first interconnect adapter to support PCI Express 3.0 and 4.0 connectivity options and to include an integrated PCIe switch.
How does ConnectX-5 fit into the Co-Design architecture that is revolutionizing the HPC industry?
ConnectX-5 is the latest crucial building block in the overall foundation of the Co-Design architecture. Its new features are designed to execute compute tasks on the network while improving overall system efficiency and speed. Co-Design enables all active components to serve as co-processors in the data center, sharing in data processing wherever the data is located in the network instead of waiting for the CPU bottleneck.
Google has announced its new Tensor processing core, which presents as much as a microcontroller as a CPU that will handle compute functions. This provides even more evidence that the future of the data center is one in which massive amounts of data processing and analysis will continue to occur throughout the network simultaneously. Data is so distributed and the requirements for analysis are so demanding that there is simply no way that the CPU can handle all the processing on its own, nor can data centers afford to wait for the data to get to the CPU. As such, the intelligent network is tasked with handling the processing and analysis on the fly.
ConnectX-5 is the perfect complement to Mellanox’s other intelligent building blocks within the Co-Design architecture. Between Switch-IB 2, the world’s first smart switch, and the NPS network processor family, Mellanox has produced a series of network devices that can analyze data as it moves throughout the network, enabling the utmost performance in the data center.
What markets will ConnectX-5 address?
Mellanox’s new intelligent adapter is designed to address numerous applications for performing data-related algorithms on the network to achieve the highest system performance with minimum CPU utilization. ConnectX-5 brings a series of enhancements and innovations, targeting various markets including:
ConnectX-5 is the first adapter that supports Co-Design and In-Network computing. It enhances HPC infrastructures by providing MPI and SHMEM/PGAS collective communications accelerations, MPI Tag Matching offload, hardware support for out-of-order RDMA operations, as well as additional Atomic operations support. ConnectX-5 also enhances the innovative transport service Dynamic Connected Transport (DCT), to ensure extreme scalability for compute and storage systems.
ConnectX-5 offers advanced NVMf target offloads by leveraging its RDMA capabilities, enabling very efficient NVMe storage access with no CPU intervention. Its innovative Host Chaining capabilities and the embedded PCIe switch enable customers to build standalone storage appliances.
The new Accelerated Switching and Packet Processing (ASAP2) technology enhances virtual switch and virtual router (e.g. Open V-Switch aka OVS) offloading, which results in significantly higher data transfer performance in virtualized environments with minimal overload of the CPU. Together, with native RoCE and RoCE for Overlay Protocols support, ConnectX-5 dramatically improves Cloud and NFV platform efficiency.
We’re as thrilled to announce ConnectX-5 as customers should be to soon receive the highest application performance and data center return on investment. For more information on ConnectX-5, visit: