RB4082 – Accelerator Pooling Appliance
Fully Managed Disaggregated Compute Accelerator Appliance for AI/DL, HPC, Visualization and Data Science Applications
Highlights
The GigaIO™ Gen4 Accelerator Pooling Appliance is the industry’s highest performing PCIe accelerator appliance fully supporting PCIe Gen4 with up to 1Tb/sec bandwidth into and out of the box.
It delivers advanced provisioning and monitoring of compute accelerators and next-gen NVMe assets, supporting up to 8 double-width PCIe Gen 4.0 x16 accelerator cards with up to 300W delivered to every slot, and 2 PCIe Gen 4.0 x16 low-profile slots.
This flexible expansion platform enables users to add any PCIe Gen 4.0 application accelerators, including GPUs, FPGAs, IPUs, DPUs, thin-NVMe-servers and specialty AI chips. It also delivers a whopping 6400W of fully redundant power to feed even the most power-hungry accelerators.
Composable Disaggregated Infrastructure by dynamically provisioning and scaling PCIe devices
GigaIO FabreX fabric creates shared pools of PCIe devices in disaggregated GPU chassis connected by standard PCIe cables. Users can dynamically provision GPUs when needed and return them to the common pool when finished.
Intelligence for resource optimal utilization and failure prediction
GigaIO’s advanced Redfish® API interface provides complete centralized analytics such as GPU utilization and performance, as well as continuous monitoring for any faults for faster problem resolution and improved reliability.
Designed from the ground up for enterprise data center deployment
In addition to continuous monitoring and control of the accelerators, should a device fail, it is essential to provide fast, easy and intuitive serviceability. With service indicators internal and external, quick and sure identification of problem devices is assured. Coupled with independent card power control and redundant power supplies and fans, this appliance is the perfect answer for disaggregating essential accelerators but maintaining tight control and easy serviceability.