Racked GPU system
A racked GPU system includes a rack system defining a plurality of device housings. An interposer device is housed in the rack system adjacent the plurality of device housings. A plurality of compute devices that each include a plurality of Graphics Processing Units (GPU) devices are housed in a respective one of the plurality of the device housings and connected to the interposer device. At least one switch system including a plurality of networking processing devices is housed in the rack system opposite the interposer device from the plurality of compute devices and the plurality of device housings, and connected to the interposer device to communicatively couple each of the plurality of networking processing devices in that switch system to each of the plurality of GPU devices in each of the plurality of compute devices.
1 . A racked Graphics Processing Unit (GPU) system, comprising:
a rack system defining a plurality of device housings;
an interposer device that is configured to be housed in the rack system adjacent the plurality of device housings;
a plurality of compute devices that each include a plurality of Graphics Processing Units (GPU) devices, wherein each of the plurality of compute devices is configured to be housed in a respective one of the plurality of the device housings and connected to the interposer device; and
at least one switch system that includes a plurality of networking processing devices, wherein each at least one switch system is configured to be housed in the rack system opposite the interposer device from the plurality of compute devices and the plurality of device housings, and connected to the interposer device to communicatively couple each of the plurality of networking processing devices in that switch system to each of the plurality of GPU devices in each of the plurality of compute devices.
2 . The system of claim 1 , wherein all of the plurality of device housings that are defined by the rack system are configured to house a respective one of the plurality of compute devices.
3 . The system of claim 1 , wherein the interposer device includes:
a plurality of switch system connector groups that each include a plurality of switch system connectors that are configured to connect to a respective one of the at least one switch system; and
a plurality of compute device connector groups that each include a plurality of compute device connectors that are configured to connect to a respective one of the plurality of compute devices, and wherein the interposer device is configured to:
reconfigure the connectivity between the compute device connectors included in the plurality of compute device connector groups and the switch system connectors included in the plurality of switch system connector groups based on a subset of the plurality of switch system connector groups that are connected the at least one switch system.
4 . The system of claim 1 , wherein each of the plurality of compute devices includes four GPU devices, the at least one switch system includes at least two switch systems each including nine networking processing devices, and the interposer device is configured to connect to up to four switch systems.
5 . The system of claim 1 , wherein each of the plurality of compute devices includes four GPU devices, each of the at least one switch system includes six networking processing devices, and the interposer device is configured to connect to up to six switch systems.
6 . The system of claim 1 , wherein the interposer device includes an interposer device chassis, and wherein each of the plurality of networking processing devices is provided on a respective networking processing device board that is mounted to the interposer device chassis.
7 . An Information Handling System (IHS), comprising:
a rack system defining a plurality of device housings;
an interposer device housed in the rack system adjacent the plurality of device housings;
a plurality of compute devices that each include a plurality of Graphics Processing Units (GPU) devices, wherein each of the plurality of compute devices is housed in a respective one of the plurality of the device housings and connected to the interposer device; and
at least one switch system that includes a plurality of networking processing devices, wherein each at least one switch system is housed in the rack system opposite the interposer device from the plurality of compute devices and the plurality of device housings, and connected to the interposer device, wherein each of the plurality of GPU devices in the plurality of compute devices communicates via the interposer device and the at least one switch system with at least one of the others of the plurality of GPU devices in the plurality of compute devices.
8 . The IHS of claim 7 , wherein all of the plurality of device housings that are defined by the rack system house a respective one of the plurality of compute devices.
9 . The IHS of claim 7 , wherein the interposer device includes:
a plurality of switch system connector groups that each include a plurality of switch system connectors that are configured to connect to a respective one of the at least one switch system; and
a plurality of compute device connector groups that each include a plurality of compute device connectors connected to a respective one of the plurality of compute devices, and wherein the interposer device is configured to:
reconfigure the connectivity between the compute device connectors in the plurality of compute device connector groups and the switch system connectors in the plurality of switch system connector groups based on a subset of the plurality of switch system connector groups that are connected the at least one switch system.
10 . The IHS of claim 7 , wherein each of the plurality of compute devices includes four GPU devices, the at least one switch system includes at least two switch systems each including nine networking processing devices, and the interposer device is configured to connect to up to four switch systems.
11 . The IHS of claim 7 , wherein each of the plurality of compute devices includes four GPU devices, each of the at least one switch system includes six networking processing devices, and the interposer device is configured to connect to up to six switch systems.
12 . The IHS of claim 7 , wherein the interposer device includes an interposer device chassis, and wherein each of the plurality of networking processing devices is provided on a respective networking processing device board that is mounted to the interposer device chassis.
13 . The IHS of claim 7 , wherein each of the plurality of GPU devices included in the plurality of compute devices is connected to a respective one of the plurality of networking processing devices in the at least one switch system via a single bidirectional serial link.
14 . A method for providing a racked Graphics Processing Unit (GPU) system, comprising:
positioning, by an interposer device, in a rack system defining a plurality of device housings such that the interposer device is located adjacent the plurality of device housings;
positioning, by each a plurality of compute devices that each include a plurality of Graphics Processing Units (GPU) devices, in a respective one of the plurality of the device housings;
connecting, by each of the plurality of compute devices in response to being positioned in the respective one of the plurality of the device housings, to the interposer device;
positioning, by at least one switch system that includes a plurality of networking processing devices, in the rack system opposite the interposer device from the plurality of compute devices and the plurality of device housings;
connecting, by the at least one switch system in response to being positioned in the rack system, to the interposer device to communicatively couple each of the plurality of networking processing devices in that switch system to each of the plurality of GPU devices in each of the plurality of compute devices.
15 . The method of claim 14 , wherein respective one of the plurality of compute devices is positioned in all of the plurality of device housings that are defined by the rack system house.
16 . The method of claim 14 , further comprising:
connecting, by a plurality of switch system connector groups on the interposer device that each include a plurality of switch system connectors and in response to the at least one switch system being positioned in the rack system, to a respective one of the at least one switch system; and
connecting, by a plurality of compute device connector groups on the interposer device that each include a plurality of compute device connectors in response to each of the plurality of compute devices being positioned in the respective one of the plurality of device housings, to a respective one of the plurality of compute devices; and
reconfiguring, by the interposer device, the connectivity between the compute device connectors in the plurality of compute device connector groups and the switch system connectors in the plurality of switch system connector groups based on a subset of the plurality of switch system connector groups that are connected the at least one switch system.
17 . The method of claim 14 , wherein each of the plurality of compute devices includes four GPU devices, the at least one switch system includes at least two switch systems each including nine networking processing devices, and the interposer device is configured to connect to up to four switch systems.
18 . The method of claim 14 , wherein each of the plurality of compute devices includes four GPU devices, each of the at least one switch system includes six networking processing devices, and the interposer device is configured to connect to up to six switch systems.
19 . The method of claim 14 , further comprising:
mounting, by a respective networking processing device board included on each of the plurality of networking processing devices, to an interposer device chassis included on the interposer device.
20 . The method of claim 14 , further comprising:
connecting, by each of the plurality of GPU devices included in the plurality of compute devices, to a respective one of the plurality of networking processing devices in the at least one switch system via a single bidirectional serial link.