Solutions - GigaIO

Applications

Applications for Extreme Connectivity

GigaIO’s FabreXunlocks the next generation of capabilities in data center performance and throughput, especially where low latency response times and high bandwidth are crucial, or where massive amounts of data needs to be analyzed for real-time predictive analytics.

AI/ML/DL

The digitization of the real world implies massive data sets, and building accurate neural networks to make sense of the data requires high density processing via various types of accelerators. With increasingly complex machine learning pipelines, and fast evolving environment and enormous workflow variation, you need extreme flexibility and recomposability of your resources on the fly to feed these resources with the speed and the flexibility you need for ML/DL, a low-latency, high-bandwidth network fabric is a must.

Bioinformatics & Genomics

The world of Bioinformatics and Genomics is going through a transformation. The good news is that critical applications are now supporting GPUs and FPGAs, in addition to CPU-only computing, thus dramatically lowering the time per case. More cases can now be run with the same infrastructure. Now the question is how to deploy these expensive resources efficiently so that all users can gain access to these application accelerators, within your budget constraints. Only FabreX enables you to both share resources between groups, AND to combine the compute power of your servers and all your accelerators to gain even more processing throughput when necessary.

Edge Computing

Driven by the vast volumes of data generated by the rapidly growing number of sensors in IoT and IIoT, there is an increasing need to move compute where the data is being generated. To move processing to the edge your challenge is to make the most efficient use of resources to deal with space and power constraints. FabreX with half the power consumption of Infiniband or Ethernet enables you to deploy less equipment, gaining space, power and thermal efficiency.

High Performance Computing

HPC uses advanced heterogeneous computation architectures to enable faster execution of compute-intensive tasks such as entertainment & media, physical simulations, geophysical modeling, automotive & aerospace design. FabreX gives you a flexible architecture for your dynamically changing workloads, you gain sub-microsecond latency and high-bandwidth networking between all resources in your HPC systems.

FinTech

In a game where a nanosecond can be the difference between millions of dollars lost or made, trading platforms hunger for the lowest possible latency. FabreX gives you a performance advantage both at the network and the messaging levels. With the absolute lowest latency available of any computer network, FabreX is ideal for building your next generation order management system.

Cloud Computing

Whether in public or private cloud, optimizing resource utilization through virtualization is key. Your challenge is to move beyond virtualizing only storage, to also include accelerators, storage-class memory, and even CPUs and memory. FabreX gives you the sub-microsecond latency you need to build entire racks where every resource can be dynamically composed.

Choosing the Right Connectivity for the Job

Users and data center managers are faced with two options for connecting resources within multi-rack systems (aka POD, cells, or clusters):

Networks such as Ethernet, Fibre Channel or InfiniBand. Though these are good choices to share information between local and/or remote systems, they suffer from inherent overhead and high latency, creating connectivity performance bottlenecks impacting overall system performance and throughput.
A high-speed interconnect such as Intel’s Omni-Path, Cray’s Aries or InfiniBand can deliver superior performance because of built-in lower latency and higher bandwidth. However, they can be both proprietary and expensive, locking the user into also acquiring servers, storage and other critical components from one vendor.

What the industry needs is a high performance fabric based on a recognized industry standard.

GigaIO FabreX is the Right Network

This innovative architecture that delivers extreme connectivity solutions with industry leading performance.

Achieve breathtaking performance levels through end-to-end low latency and high bandwidth, typically found only with expensive proprietary interconnects. We connect it all: storage, IPC, memory, Intel, AMD, ARM, FPGA, GPU, SoC, etc.
An interconnect fabric 100% compliant with PCI SIG, that is both forward and backward compatible. PCI Express is the industry’s most prevalent standard and is supported by all vendors (servers, GPUs, storage, FPGA, etc.). The FabreX hyper-performance network protects your investment in all existing and future generation PCIe resources.

GigaIO FabreX also breaks the constraints of old architectures, opening up new configuration possibilities to maximize utilization of all elements within your advanced scale computing systems.

Heterogeneous compute
Heterogeneous storage
IPC – low message latency
No unnecessary added networking protocols