AIC Shows 32-Drive PCIe Gen6 JBOF Server For AI Cache Workloads
ServeTheHome reported that AIC showed a 2U PCIe Gen6 JBOF server for AI cache workloads at Computex, with 32 E3 drive bays and Broadcom PCIe Gen6 switching but no public pricing or deployment list.
AIC Shows 32-Drive PCIe Gen6 JBOF Server
AIC used Computex to show the F2032-01-G6, a 2U just-a-bunch-of-flash server built around PCIe Gen6 storage for AI inference and high-performance data workloads, ServeTheHome reported.
The outlet said the appliance moves from the PCIe Gen5-era F2026-01-G5 design to 32 PCIe Gen6 E3 drives, up from 26 2.5-inch U.
2 bays in the prior generation.
The system is built for storage close to AI accelerators, not general compute expansion.
F2032-G6 Uses 32 E3 Drive Bays
The F2032-01-G6, also referred to as F2032-G6, can hold 32 PCIe Gen6 E3 drives in a 2U chassis, ServeTheHome reported.
It supports both E3.
L and E3.
S versions of 7.5mm-thick SSDs, with hot-swappable trays filling much of the front panel.
Capacity depends on drive configuration.
A fully equipped F2032-G6 can store up to 8PB of data when using 256TB SSDs, while TLC NAND configurations sit at about half that capacity.
Broadcom PCIe Gen6 Switch Sits At The Centre
The internal I/O fabric uses Broadcom's 144-lane PEX90144 PCIe Gen6 switch, ServeTheHome said.
The Computex system was not fully opened, so the expected lane routing is based on the previous-generation design rather than a published architecture diagram.
The network side is built for DPUs as much as NICs.
It uses two nodes, each with its own power supply and space for up to two network expansion devices, and the design can also house a GPU.
BlueField-4 DPUs Put The Appliance Near Vera Rubin Racks
The JBOF appliance is designed to be controlled by at least one Nvidia BlueField-4 DPU.
The Grace CPU and ConnectX-9 NIC-based DPUs support 800Gb Ethernet or InfiniBand networking and PCIe Gen6 I/O connections.
The server is pitched as a high-availability system.
Support for two DPUs and a second node allows redundant operation if a DPU or part of one fails.
The named use case is Nvidia's Context Memory Storage platform for backing Vera Rubin systems.
Other target uses include object storage, scale-out file systems and flash storage tiering.
Pricing, customer deployments, order volumes, a general-availability date, final architecture diagrams, or customer benchmark results for the F2032-01-G6 remain outside the public record.
