CN112069121A - MCU control GPU server Switch board system and control method - Google Patents
MCU control GPU server Switch board system and control method Download PDFInfo
- Publication number
- CN112069121A CN112069121A CN202010852008.6A CN202010852008A CN112069121A CN 112069121 A CN112069121 A CN 112069121A CN 202010852008 A CN202010852008 A CN 202010852008A CN 112069121 A CN112069121 A CN 112069121A
- Authority
- CN
- China
- Prior art keywords
- unit
- mcu
- clock
- board
- switch
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Withdrawn
Links
- 238000000034 method Methods 0.000 title claims abstract description 17
- 230000003993 interaction Effects 0.000 claims abstract description 7
- 238000006243 chemical reaction Methods 0.000 claims description 3
- 230000005611 electricity Effects 0.000 claims description 2
- 238000012545 processing Methods 0.000 abstract description 5
- 230000002093 peripheral effect Effects 0.000 description 5
- 238000004891 communication Methods 0.000 description 2
- 238000013461 design Methods 0.000 description 2
- 238000003780 insertion Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000037431 insertion Effects 0.000 description 1
- 238000009434 installation Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000008054 signal transmission Effects 0.000 description 1
- 239000000758 substrate Substances 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F15/00—Digital computers in general; Data processing equipment in general
- G06F15/16—Combinations of two or more digital computers each having at least an arithmetic unit, a program unit and a register, e.g. for a simultaneous processing of several programs
- G06F15/163—Interprocessor communication
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F13/00—Interconnection of, or transfer of information or other signals between, memories, input/output devices or central processing units
- G06F13/38—Information transfer, e.g. on bus
- G06F13/42—Bus transfer protocol, e.g. handshake; Synchronisation
- G06F13/4282—Bus transfer protocol, e.g. handshake; Synchronisation on a serial bus, e.g. I2C bus, SPI bus
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Computer Hardware Design (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Software Systems (AREA)
- Information Transfer Systems (AREA)
Abstract
The invention discloses a system and a method for controlling a GPU server Switch Board by an MCU (microprogrammed control Unit), wherein the system comprises an MCU unit, and the MCU unit is respectively connected with a BMC unit, a PCIE Switch unit, a Clock Mux unit, a Clock Buffer unit and a Board VR unit; the MCU unit performs I2C data interaction with the BMC unit to acquire required configuration information on one hand, and configures the working modes of PCIE Switch, Clock Mux and Clock Buffer on the other hand through MCUGPIO; meanwhile, the MCU unit can also control the Power Enable signal of the Board VR and detect the Power Good signal so as to realize the main electrical sequence control function of the Switch Board. The working modes of the chips of the Switch board can be freely switched according to actual requirements so as to meet the performance requirements of different users/GPU (graphics processing unit) boards.
Description
Technical Field
The invention relates to the technical field of Switch board control, in particular to a system and a method for controlling a GPU server Switch board by an MCU (microprogrammed control unit).
Background
With the continuous development of internet technology and the continuous increase of parallel computing demand, the GPU can effectively release computing pressure with its excellent graphics processing capability and high-performance computing capability, and significantly improve the computing processing efficiency and competitiveness of products, so the GPU server gradually grows and shows a more and more intense trend.
In order to fully exert the calculation performance of the GPU, in the GPU server, the GPU installation mode has been gradually changed from the traditional GPU card insertion mode to the on-board GPU mode, and unlike the card-insertion GPU, the on-board GPU needs to provide multiple sets of PCIE signals for the GPU card to realize data interaction between the CPU and each GPU, which is limited to the server space structure and the number of PCIE ports of the CPU, and currently, PCIE Switch chips (PCIE Switch chips) are mostly used to realize data interaction between the CPU and the GPU. The existing card-inserting type GPU design cannot fully exert the powerful operation performance of the GPU.
Disclosure of Invention
The invention aims to solve the problems and provides a system and a method for controlling a GPU server Switch board by an MCU (microprogrammed control Unit), which can be used for not only NVIDIA HGX-2GPU boards but also PCIE Switch board control of other GPU boards; meanwhile, the control method can also freely Switch the working modes of the chips of the Switch board according to actual requirements so as to meet the performance requirements of different users/GPU boards.
In order to achieve the purpose, the invention adopts the following technical scheme:
an MCU control GPU server Switch Board system comprises an MCU unit, wherein the MCU unit is respectively connected with a BMC unit, a PCIE Switch unit, a Clock Mux unit, a Clock Buffer unit and a Board VR unit;
the MCU unit performs I2C data interaction with the BMC unit to acquire required configuration information on one hand, and configures the working modes of PCIE Switch, Clock Mux and Clock Buffer on the other hand through MCUGPIO;
meanwhile, the MCU unit can also control the Power Enable signal of the Board VR and detect the Power Good signal so as to realize the main electrical sequence control function of the Switch Board.
The BMC unit is connected with the MCU unit, reads and writes an MCU internal register through an I2C link to acquire a current Switch board configuration mode, and meanwhile, the remote control function of the Switch board can be realized by modifying corresponding I2C register parameters.
The PCIE Switch unit is connected with the MCU unit, receives configuration information from the MCU and realizes the working mode configuration function of the PCIE Switch chip; meanwhile, the PCIE Switch unit is connected with the CPU and the GPU board, and the PCIE signal expansion function is achieved.
The Clock Mux unit is connected with the MCU unit and receives the configuration information from the MCU to realize the working mode configuration function of the Clock Mux chip;
the Clock Mux unit is also connected with Clock sources inside and outside the board and a Clock Buffer, and different Clock sources are respectively selected according to different working modes.
The Clock Buffer unit is connected with the MCU unit and receives configuration information from the MCU to realize the working mode configuration function of the Clock Buffer chip; the Clock Buffer unit is also connected with a GPU board to realize the 100M Clock signal expansion function.
The Board VR unit is connected with the MCU unit, receives a Power Enable signal from the MCU and simultaneously sends Power Good to the MCU; and the Board VR unit is also connected with Power to realize the DC/DC voltage conversion function.
The control method for controlling the GPU server Switch board system by the MCU comprises the following steps:
reading a Board ID by the MCU, configuring the ID for a GPU Board needing to be configured with an ID signal according to the current Board ID, and entering a second step;
step two, the MCU unit respectively sets the working modes of the PCIE Switch unit, the Clock Mux unit and the Clock Buffer unit according to default configuration, and enters step three;
step three, judging whether the BMC unit modifies the working modes of the PCIE Switch unit, the Clock Mux unit and the Clock Buffer unit, if so, entering a step four, otherwise, entering a step five;
step four, the MCU unit resets the working modes of the PCIE Switch unit, the Clock Mux unit and the Clock Buffer unit according to the configuration of the BMC unit and enters step five;
step five, starting the Switch, electrifying the Switch board main electricity, and entering step six;
step six, the BMC unit judges whether the working modes of the PCIE Switch unit, the Clock Mux unit and the Clock Buffer unit are modified, if so, the step seven is carried out, and if not, the step eight is carried out;
step seven, waiting for shutdown, resetting the working modes of the PCIE Switch unit, the Clock Mux unit and the Clock Buffer unit by the MCU unit according to the configuration of the BMC unit after the Switch board is powered off, and entering the step nine;
step eight, waiting for shutdown, powering off the Switch board, and entering step nine;
step nine, judging whether the machine is shut down, if so, entering step ten; otherwise, returning to the step five;
step ten, stopping.
The step one needs to configure the HGX-2 included in the ID signal.
The invention has the beneficial effects that:
the invention provides a method for controlling a GPU server Switch board by an MCU (microprogrammed control unit) to solve the problem of data interaction of PCIE (peripheral component interface express) signals between a CPU (Central processing Unit) and a GPU (graphics processing Unit), so that the method can be used for not only NVIDIA HGX-2GPU boards but also PCIE Switch board control of other GPU boards; meanwhile, the control method can also freely Switch the working modes of the chips of the Switch board according to actual requirements so as to meet the performance requirements of different users/GPU boards.
Drawings
FIG. 1 is a block diagram of the system of the present invention;
FIG. 2 is a flow chart of a control method of the present invention.
Detailed Description
The invention is further described with reference to the following figures and examples.
As shown in fig. 1, an MCU controlled GPU server Switch board system, as shown in fig. 1, includes: MCU unit, BMC unit, PCIE Switch unit, Clock Mux unit, Clock Buffer unit and Board VR unit.
The MCU unit is respectively connected with the BMC unit, the PCIE Switch unit, the Clock Mux unit, the Clock Buffer unit and the Board VR unit, on one hand, the MCU unit performs I2C data interaction with the BMC unit to acquire required configuration information, and on the other hand, the MCU unit configures the working modes of the PCIE Switch, the Clock Mux and the Clock Buffer through MCUGPIO; meanwhile, the MCU unit can also control the Power Enable signal of the Board VR and detect the Power Good signal so as to realize the main electrical sequence control function of the Switch Board.
(1) And the MCU is used for realizing all software control codes of the design method, including a Board ID reading and configuration code, an I2C data communication code, a PCIE Switch, a Clock MUX, a Clock Buffer working mode configuration code, a main power-on and power-off time sequence control code and the like.
(2) The BMC unit is connected with the MCU unit, the BMC unit can read and write the MCU internal register through an I2C link to acquire the current Switch board configuration mode, and meanwhile, the remote control function of the Switch board can be realized by modifying the corresponding I2C register parameters. The BMC unit carries relevant peripheral circuits and is used for realizing a server substrate control function, and in the patent scope, the BMC is used as a host in data communication with the MCU I2C and is used for acquiring a Switch board configuration mode and controlling the MCU to Switch PCIE Switch, Clock MUX and Clock Buffer working mode functions.
(3) The PCIE Switch unit is connected to the MCU unit, on one hand, receives configuration information from the MCU, and implements a function of operating mode configuration of the PCIE Switch chip, and on the other hand, is connected to the CPU and the GPU board, and implements a PCIE signal expansion function. The PCIE Switch unit is equipped with its peripheral circuits, and is mainly used to expand one or two sets of PCIE signals from the CPU into multiple sets of PCIE signals, and is used for PCIE signal transmission between the CPU and the GPU.
(4) The Clock Mux unit is connected with the MCU unit on one hand, receives the configuration information from the MCU, realizes the working mode configuration function of the Clock Mux chip, and is connected with the Clock source and the Clock Buffer inside and outside the Clock Mux board on the other hand, and different Clock sources are respectively selected according to different working modes. The Clock Mux unit is provided with peripheral circuits and is mainly used for selecting one group of Clock sources between an external 100M Clock and a Switch board local 100M Clock to be used as a GPU board 100M Clock source.
(5) The Clock Buffer unit is connected with the MCU unit on one hand, receives configuration information from the MCU and realizes the working mode configuration function of the Clock Buffer chip, and is connected with the GPU board on the other hand, and realizes the 100M Clock signal expansion function. The Clock Buffer unit is provided with peripheral circuits and is mainly used for expanding a group of 100M clocks from the Clock Mux unit into a plurality of groups of 100M clocks and providing 100M clocks for the GPU inside the GPU board.
(6) The Board VR (Voltage Regulator) unit is connected with the MCU unit on one hand, receives a Power Enable signal from the MCU, simultaneously sends Power Good to the MCU, and is connected with the Power on the other hand, so that the DC/DC Voltage conversion function of the Board VR unit is realized. The Board VR unit is used for providing a working Power supply for the chip in the Switch Board, and meanwhile, the Power Enable/Power Good signals of the Board VR unit are connected with the MCU unit and used for realizing the main Power-on and Power-off time sequence control function of the MCU.
A control method for an MCU to control a GPU server Switch board system, as shown in fig. 2, includes:
step one, the MCU reads the Board ID, configures the ID for the GPU Board needing to configure the ID signal, such as HGX-2, and the like according to the current Board ID, and enters step two;
step two, the CU sets the working modes of the PCIE Switch, the Clock Mux and the Clock Buffer respectively according to default configuration, and enters step three;
step three, whether the BMC modifies the working modes of the PCIE Switch, the Clock Mux and the Clock Buffer or not is judged, if yes, the step four is carried out, and if not, the step five is carried out;
step four, the MCU resets the working modes of the PCIE Switch, the Clock Mux and the Clock Buffer according to the configuration of the BMC, and enters the step five;
step five, starting the system (comprising a mainboard, a GPU board and a Switch board), electrifying the Switch board mainly, and entering step six;
step six, in the working process of the system, whether the BMC modifies the working modes of the PCIE Switch, the Clock Mux and the Clock Buffer or not is judged, if yes, the step seven is carried out, and if not, the step eight is carried out;
step seven, waiting for the system to be shut down, and after the Switch board is powered off, resetting the working modes of the PCIE Switch, the Clock Mux and the Clock Buffer by the MCU according to the configuration of the BMC, and entering the step nine;
step eight, waiting for the system to be shut down, and turning on and off the Switch board, and entering the step nine;
step nine, whether the machine is shut down or not, if yes, entering step ten; otherwise, returning to the step five;
step ten, stopping.
Although the embodiments of the present invention have been described with reference to the accompanying drawings, it is not intended to limit the scope of the present invention, and it should be understood by those skilled in the art that various modifications and variations can be made without inventive efforts by those skilled in the art based on the technical solution of the present invention.
Claims (8)
1. An MCU control GPU server Switch Board system is characterized by comprising an MCU unit, wherein the MCU unit is respectively connected with a BMC unit, a PCIE Switch unit, a Clock Mux unit, a Clock Buffer unit and a Board VR unit;
the MCU unit performs I2C data interaction with the BMC unit to acquire required configuration information on one hand, and configures the working modes of PCIE Switch, Clock Mux and Clock Buffer on the other hand through MCUGPIO;
meanwhile, the MCU unit can also control the Power Enable signal of the Board VR and detect the Power Good signal so as to realize the main electrical sequence control function of the Switch Board.
2. The MCU-controlled GPU server Switch board system of claim 1, wherein the BMC unit is connected to the MCU unit, and reads and writes an MCU internal register through an I2C link to obtain a current Switch board configuration mode, and meanwhile, can implement a Switch board remote control function by modifying corresponding I2C register parameters.
3. The MCU-controlled GPU server Switch board system of claim 1, wherein the PCIE Switch unit is connected with the MCU unit, receives configuration information from the MCU, and realizes the working mode configuration function of the PCIE Switch chip; meanwhile, the PCIE Switch unit is connected with the CPU and the GPU board, and the PCIE signal expansion function is achieved.
4. The MCU-controlled GPU server Switch board system according to claim 1, wherein the Clock Mux unit is connected with the MCU unit, receives configuration information from the MCU, and realizes a working mode configuration function of the Clock Mux chip;
the Clock Mux unit is also connected with a Clock source and a Clock Buffer, and different Clock sources are respectively selected according to different working modes.
5. The MCU-controlled GPU server Switch board system according to claim 1, wherein the Clock Buffer unit is connected with the MCU unit, receives configuration information from the MCU, and realizes the working mode configuration function of the Clock Buffer chip; the Clock Buffer unit is also connected with a GPU board to realize the 100M Clock signal expansion function.
6. The MCU-controlled GPU server Switch Board system of claim 1, wherein said Board VR unit is connected to said MCU unit, receives Power Enable signal from said MCU, and transmits Power Good to said MCU; and the Board VR unit is also connected with Power to realize the DC/DC voltage conversion function.
7. The method for controlling the GPU server Switch board system by the MCU as claimed in claim 1, wherein the method comprises:
reading a Board ID by the MCU, configuring the ID for a GPU Board needing to be configured with an ID signal according to the current Board ID, and entering a second step;
step two, the MCU unit respectively sets the working modes of the PCIE Switch unit, the Clock Mux unit and the Clock Buffer unit according to default configuration, and enters step three;
step three, judging whether the BMC unit modifies the working modes of the PCIE Switch unit, the Clock Mux unit and the Clock Buffer unit, if so, entering a step four, otherwise, entering a step five;
step four, the MCU unit resets the working modes of the PCIE Switch unit, the Clock Mux unit and the Clock Buffer unit according to the configuration of the BMC unit and enters step five;
step five, starting the Switch, electrifying the Switch board main electricity, and entering step six;
step six, the BMC unit judges whether the working modes of the PCIE Switch unit, the Clock Mux unit and the Clock Buffer unit are modified, if so, the step seven is carried out, and if not, the step eight is carried out;
step seven, waiting for shutdown, resetting the working modes of the PCIE Switch unit, the Clock Mux unit and the Clock Buffer unit by the MCU unit according to the configuration of the BMC unit after the Switch board is powered off, and entering the step nine;
step eight, waiting for shutdown, powering off the Switch board, and entering step nine;
step nine, judging whether the machine is shut down, if so, entering step ten; otherwise, returning to the step five;
step ten, stopping.
8. The control method of claim 7, wherein the configuration ID signal required in step one includes HGX-2.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010852008.6A CN112069121A (en) | 2020-08-21 | 2020-08-21 | MCU control GPU server Switch board system and control method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010852008.6A CN112069121A (en) | 2020-08-21 | 2020-08-21 | MCU control GPU server Switch board system and control method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112069121A true CN112069121A (en) | 2020-12-11 |
Family
ID=73660631
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010852008.6A Withdrawn CN112069121A (en) | 2020-08-21 | 2020-08-21 | MCU control GPU server Switch board system and control method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112069121A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113177019A (en) * | 2021-04-25 | 2021-07-27 | 山东英信计算机技术有限公司 | Switch board and server |
| CN114461008A (en) * | 2021-12-21 | 2022-05-10 | 天津光电通信技术有限公司 | Method for realizing multi-path adjustable homologous clock |
-
2020
- 2020-08-21 CN CN202010852008.6A patent/CN112069121A/en not_active Withdrawn
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113177019A (en) * | 2021-04-25 | 2021-07-27 | 山东英信计算机技术有限公司 | Switch board and server |
| CN114461008A (en) * | 2021-12-21 | 2022-05-10 | 天津光电通信技术有限公司 | Method for realizing multi-path adjustable homologous clock |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN102193765B (en) | Display information updating device and method | |
| US20190303318A1 (en) | Configurable interface card | |
| CN101581964B (en) | Computer system and peripheral equipment drive method | |
| CN210954874U (en) | Domestic computer mainboard based on Shenwei SW421 treater | |
| CN112069121A (en) | MCU control GPU server Switch board system and control method | |
| CN110399328B (en) | Control method and device for board-mounted graphics processor | |
| CN102236631A (en) | Calculator system | |
| CN107463224A (en) | Display card expansion board and host and computing equipment applying same | |
| CN201383075Y (en) | PC104-plus controller based on PowerPC processor | |
| CN207020664U (en) | For extending the expansion card of PCI E interfaces | |
| CN104951268A (en) | Method for implementing extended high-performance graphics card based on CPCI | |
| CN112000545A (en) | Graphics processor board card and graphics processor management method | |
| CN103616934A (en) | FPGA core circuit board structure | |
| CN108628793B (en) | SPI communication circuit and method | |
| CN202617157U (en) | PCI express (PCIE) switched circuit | |
| CN213958044U (en) | High-reliability complex SoC full-function verification board | |
| CN112506851B (en) | SOC chip architecture construction method for solving multi-core access conflict | |
| CN101594719B (en) | Offline control device | |
| CN211786846U (en) | Operation display terminal based on MIMXRT1062 microcontroller | |
| CN113391570A (en) | Low-cost low-complexity multi-channel DA synchronous control method | |
| CN111143275A (en) | IP management and power consumption optimization system and method | |
| CN216286659U (en) | High-speed data processing equipment based on homemade VPX framework | |
| CN217880296U (en) | Data distributed processing device and stage lighting control system | |
| CN216352080U (en) | High-density industrial control mainboard based on domestic processor | |
| CN103577359B (en) | Support system and the method for managing power supply thereof of dual master control device access interface device |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| WW01 | Invention patent application withdrawn after publication |
Application publication date: 20201211 |
|
| WW01 | Invention patent application withdrawn after publication |