# US 10,311,013 B2 · High-speed inter-processor communications > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 10,311,013 B2 - Document ID: US10311013B2 - Status: Granted - Kind: B2 - Application number: 15/650,730 - Filed: 2017-07-14 - Published: 2019-01-17 - Granted: 2019-06-04 - Earliest claimed: 2017-07-14 - Assignees: Facebook, Inc. - Applicants: Facebook, Inc. - Inventors: Kevin Desmond Lee; Jia Ning; Narsing Vijayrao; Koh Yew Thoon - Web page: https://patentgrep.com/patent/US10311013B2 ## CPC - G06F 15/17: using an input/output type connection, e.g. channel, I/O port - G06F 15/161: {Computing infrastructure, e.g. computer clusters, blade chassis or hardware partitioning (casings, cabinets, racks or drawers for data centers H05K5/00)} - G06F 15/173: using an interconnection network, e.g. matrix, shuffle, pyramid, star, snowflake - G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N) ## Abstract A computing device has a motherboard, at least two daughter boards communicably connected to the motherboard, each of the at least two daughter boards having at least a specialized processor and a high-speed inter-processor communications port; and at least two high-speed inter-processor communication interconnects connecting at least two of the high-speed inter-processor communications ports. The configuration enables a flexible topology architecture, e.g., for different applications, and rapid reuse of system components even when new specialized processors become available. ## Claims ### Claim 1 A computing device, comprising: a motherboard having one or more central processing units; at least three daughter boards communicably connected to the motherboard, including a first daughter board and a second daughter board, each of the at least three daughter boards having at least a specialized processor and a high-speed inter-processor communications port; and at least two high-speed inter-processor communications interconnects connecting at least two of the high-speed inter-processor communications ports, including a high-speed inter-processor communications interconnect that connects a high-speed inter-processor communications port of the first daughter board with a second high-speed inter-processor communications port of the second daughter board, wherein: the high-speed inter-processor communications ports and the high-speed inter-processor communications interconnects conform to an industry standard; the industry standard is a QUAD SMALL FORM-FACTOR PLUGGABLE standard; the at least two high-speed inter-processor communications interconnects comprise a first QUAD SMALL FORM-FACTOR PLUGGABLE cable and a second QUAD SMALL FORM-FACTOR PLUGGABLE cable; and the specialized processor of the second daughter board is configured to directly access results of executing an application that are stored in a memory module of the first daughter board across the first QUAD SMALL FORM-FACTOR PLUGGABLE cable that directly connects the high-speed inter-processor communications port of the first daughter board with the second high-speed inter-processor communications port of the second daughter board such that a bottleneck of a data path through a network card connected to a standard interface is bypassed. ### Claim 2 (depends on 1) The computing device of claim 1, wherein the standard interface is a PERIPHERAL COMPUTING INTERCONNECT EXPRESS interface. ### Claim 3 (depends on 2) The computing device of claim 2, wherein the application comprises a machine learning processing application. ### Claim 4 (depends on 3) The computing device of claim 3, wherein the machine learning processing application comprises a deep learning application. ### Claim 5 (depends on 1) The computing device of claim 1, further comprising multiple additional daughter boards, each of the additional daughter boards having at least one specialized processor, wherein the additional daughter boards are interconnected in a first topology by interconnecting high-speed inter-processor communications cables in a first arrangement. ### Claim 6 (depends on 5) The computing device of claim 5, wherein the additional daughter boards and the three or more daughter boards are interconnected in a second topology by interconnecting high-speed inter-processor communications cables in a second arrangement. ### Claim 7 (depends on 1) The computing device of claim 1, wherein the daughter boards are adaptable for use with specialized processors from different manufacturers. ### Claim 8 (depends on 1) The computing device of claim 1, wherein the high-speed inter-processor communications port is adaptable for different pinout configurations. ### Claim 9 A daughter board, comprising: a specialized processor; a high-speed inter-processor communications port connected to an additional high-speed inter-processor communications port of an additional daughter board through a high-speed inter-processor communications interconnect; and a standard interface for communicably connecting with a motherboard, wherein: the high-speed inter-processor communications port and the high-speed inter-processor communications interconnect conform to an industry standard; the industry standard is a QUAD SMALL FORM-FACTOR PLUGGABLE standard such that the high-speed inter-processor communications interconnect comprises a QUAD SMALL FORM-FACTOR PLUGGABLE cable that directly connects the high-speed inter-processor communications port to the additional high-speed inter-processor communications port of the additional daughter board; and the specialized processor is configured to directly access results of executing an application that are stored in a memory module of the additional daughter board across the high-speed inter-processor communications interconnect that connects the high-speed inter-processor communications port with the additional high-speed inter-processor communications port of the additional daughter board such that a bottleneck of a data path through a network card connected to the standard interface is bypassed. ### Claim 10 (depends on 9) The daughter board of claim 9, wherein the standard interface is a PERIPHERAL COMPUTING INTERCONNECT EXPRESS interface. ### Claim 11 (depends on 9) The daughter board of claim 9, wherein the daughter board is disposed within a computing device. ### Claim 12 (depends on 9) The daughter board of claim 9, wherein the specialized processor is a processor designed to handle math computations more quickly than a general purpose central processing unit. ### Claim 13 (depends on 9) The daughter board of claim 9, wherein the specialized processor is configured to access a first memory associated with the daughter board via a bus on the daughter board. ### Claim 14 (depends on 9) The daughter board of claim 9, wherein the application comprises a machine learning processing application. ### Claim 15 (depends on 14) The daughter board of claim 14, wherein the machine learning processing application comprises a deep learning application. ### Claim 16 A method performed by a specialized processor, comprising: receiving a signal to access a memory module; determining whether the memory module to be accessed is associated with a second daughter board on which the specialized processor is attached or a first daughter board; and in an event the memory module to be accessed is associated with the first daughter board, directly accessing results of executing an application that are stored in the memory module of the first daughter board across a high-speed inter-processor communications interconnect comprising a QUAD SMALL FORM-FACTOR PLUGGABLE cable that directly connects a high-speed inter-processor communications port of the first daughter board with a second high-speed inter-processor communications port of the second daughter board such that a bottleneck of a data path through a network card connected to a standard interface is bypassed.