GigalO has launched FabreX 2.2, a native PCI Express Gen4 universal dynamic fabric that supports MPI, TCP/IP, GRD, and NVMe-oF. This brand-new product by GigalO introduces scalability across the PCIe fabric for AI workloads by allowing the creation of composable GigaClusters and GigaPods with cascaded interlink switches.
Along with that, FabreX 2.2 would offer performance improvements of up to 30% overall server-to-server communications via new and revamped DMA implementations.
GigalO’s latest technology allows the entire rack to be treated as a compute unit. It would break the barrier of the server box in a very smooth way. All I/O resources situated inside the server (Optane performance high storage, GPUs, FPGAs) can be pooled in storage appliances and accelerator where they are accessible for all the servers in the rack. Components like Intel Optane SSDs will continue to communicate across native PCIe and maybe CXL in the near future.
According to GigaIO, the services will be provided in such a way that it will feel like they are still connected to the server motherboard, for the highest performance and lowest possible latency. The basic compute unit, GigaCell, can be extended to several racks with FabreX 2.2. This would result in a faster time to result since workloads run as if they were using components inside a single server. Simultaneously, they will use the potential of various nodes, all communicating with one seamless universal fabric. Scale-out topologies like dragonfly, leaf, and spine are completely supported.
Data Center Managers
Senior Director of Product Management for Intel Optane, Kristie Mann, said that the GigalO FabreX environment with Intel Optane SSDs is allowing scalable performance with significantly lower latency than the other options available for NVMe.
Optane SSDs combination and native PCIe fabric are intended to boost the potential of both IT organizations and solutions can witness performance benefits and minimized storage bottlenecks over the network.
Data center managers might save some extra cost, maintenance burden, and the complexity of setting up of various networks including Ethernet and InfiniBand, running in a rack with a sea of NICs as now they are only supposed to deploy PCIe outside the server using the PCIe switches. FabreX’s dynamic disaggregation and composability capability allows for ‘pay as you grow’ using the provided resources independently.
“With our revolutionary technology, a true rack-scale system can be created with only PCIe as the network,” said Alan Benjamin, Chief Executive Officer (CEO) of GigaIO. “The implication for HPC and AI workloads, which consume large amounts of accelerators and high-speed storage like Intel Optane SSDs to minimize time to results, is much faster computation, and the ability to run workloads which simply would not have been possible in the past.”