Modular GPU architecture for clients and servers
One embodiment provides a graphics processor including an active base die including a fabric interconnect and a chiplet including a switched fabric, wherein the chiplet couples with the active base die via an array of interconnect structures, the array of interconnect structures couple the fabric interconnect with the switched fabric, and the chiplet includes a first modular interconnect configured to couple a block of graphics processing resources to the switched fabric and a second modular interconnect configured to couple a memory subsystem with the switched fabric and the block of graphics processing resources, the memory interconnect including a set of memory controllers and a set of physical interfaces.
1 . A general-purpose graphics processor comprising:
an active base die including a fabric interconnect; and
a first chiplet including a switched fabric, wherein the first chiplet is configured to couple with the active base die via an array of interconnect structures, the array of interconnect structures additionally to couple the fabric interconnect with the switched fabric, the first chiplet including:
a first modular interconnect configured to couple a block of graphics processing resources to the switched fabric, wherein the first modular interconnect is configured to selectively couple, to the switched fabric, one of a first block of graphics processing resources having a first number of graphics core clusters and a second block of graphics resources having a second number of graphics core clusters;
a second modular interconnect configured to couple a memory interconnect of a memory subsystem with the switched fabric and the block of graphics processing resources, the memory interconnect including a set of memory controllers and a set of physical interfaces; and
a tile-to-socket interface to couple a connectivity die to the first chiplet, the connectivity die to enable a multi-board graphics processor via a multi-board connection with an external general-purpose graphics processor, the multi-board connection established via the first chiplet.
2 . The general-purpose graphics processor as in claim 1 , the first chiplet additionally including a third modular interconnect configured to couple a host interconnect physical interface and upstream port to the switched fabric, the third modular interconnect configured to couple with the switched fabric via a fabric bridge.
3 . The general-purpose graphics processor as in claim 1 , wherein the memory subsystem is associated with a graphics double data rate (GDDR) memory device.
4 . The general-purpose graphics processor as in claim 1 , wherein the memory subsystem is associated with a high bandwidth memory (HBM) device.
5 . The general-purpose graphics processor as in claim 1 , wherein the connectivity die is configured to couple with a mezzanine connector.
6 . The general-purpose graphics processor as in claim 1 , wherein the first block of graphics processing resources includes a first set of graphics cores, each graphics core of the first set of graphics cores including a matrix accelerator and a vector engine.
7 . The general-purpose graphics processor as in claim 1 , wherein the second block of graphics processing resources includes a first set of graphics cores, each graphics core of the first set of graphics cores including a vector engine.
8 . The general-purpose graphics processor as in claim 7 , wherein the vector engine includes hardware logic configurable to perform variable vector width operations.
9 . The general-purpose graphics processor as in claim 1 , additionally including a second chiplet coupled with the first chiplet, the second chiplet including a cache memory, wherein the cache memory is a last level cache.
10 . The general-purpose graphics processor as in claim 1 , wherein the memory subsystem has a dynamically configurable voltage and frequency.
11 . A graphics processing apparatus including:
a circuit board including a host interconnect; and
a system on a chip integrated circuit (SoC) including an active base die including a fabric interconnect and a first chiplet including a switched fabric, wherein the first chiplet is configured to couple with the active base die via an array of interconnect structures, the array of interconnect structures additionally to couple the fabric interconnect with the switched fabric, the first chiplet including:
a first modular interconnect configured to couple a block of graphics processing resources to the switched fabric, wherein the first modular interconnect is configured to selectively couple, to the switched fabric, one of a first block of graphics processing resources having a first number of graphics core clusters and a second block of graphics resources having a second number of graphics core clusters;
a second modular interconnect configured to couple a memory interconnect of a memory subsystem with the switched fabric and the block of graphics processing resources, the memory interconnect including a set of memory controllers and a set of physical interfaces; and
a tile-to-socket interface to couple a connectivity die to the first chiplet, the connectivity die to enable a multi-board graphics processor via a multi-board connection with an external general-purpose graphics processor, the multi-board connection established via the first chiplet.
12 . The graphics processing apparatus as in claim 11 , the first chiplet additionally including a third modular interconnect configured to couple a host interconnect physical interface and upstream port to the switched fabric, the third modular interconnect configured to couple with the switched fabric via a fabric bridge.
13 . The graphics processing apparatus as in claim 11 , wherein the memory subsystem is associated with a graphics double data rate (GDDR) memory device.
14 . The graphics processing apparatus as in claim 11 , wherein the memory subsystem is associated with a high bandwidth memory (HBM) device.
15 . The graphics processing apparatus as in claim 11 , wherein the connectivity die is configured to couple with a mezzanine connector.
16 . The graphics processing apparatus as in claim 11 , wherein the first block of graphics processing resources includes a first set of graphics cores, each graphics core of the first set of graphics cores including a matrix accelerator and a vector engine.
17 . The graphics processing apparatus as in claim 11 , wherein the second block of graphics processing resources includes a first set of graphics cores, each graphics core of the first set of graphics cores including a vector engine.
18 . The graphics processing apparatus as in claim 17 , wherein the vector engine includes hardware logic configurable to perform variable vector width operations.
19 . The graphics processing apparatus as in claim 11 , additionally including a second chiplet coupled with the first chiplet, the second chiplet including a cache memory, wherein the cache memory is a last level cache.
20 . The graphics processing apparatus as in claim 11 , wherein the memory subsystem has a dynamically configurable voltage and frequency.