Disaggregation of System-On-Chip (SOC) architecture
Embodiments described herein provide techniques to disaggregate an architecture of a system on a chip integrated circuit into multiple distinct chiplets that can be packaged onto a common chassis. In one embodiment, a graphics processing unit or parallel processor is composed from diverse silicon chiplets that are separately manufactured. A chiplet is an at least partially packaged integrated circuit that includes distinct units of logic that can be assembled with other chiplets into a larger package. A diverse set of chiplets with different IP core logic can be assembled into a single device.
1. A general-purpose graphics processor comprising:
a 3-dimensional (3D) or 2.5 dimensional (2.5D) package assembly including:
an active base die including hardware logic and an interconnect fabric;
an array of multiple distinctly packaged chiplets coupled with the active base die and the interconnect fabric via an array of interconnect structures, the interconnect structures to enable electrical communication between the array of multiple distinctly packaged chiplets and the interconnect fabric, the array of multiple distinctly packaged chiplets comprising;
a first chiplet coupled with the active base die, the first chiplet including a first set of functional units configured to perform graphics processing operations;
a second chiplet coupled with the active base die, the second chiplet including a second set of functional units configured to perform graphics processing operations; and
a third chiplet coupled with the first chiplet and the second chiplet, the third chiplet including a cache memory that is shared by the first chiplet and the second chiplet.
2. The general-purpose graphics processor as in claim 1 , wherein the first chiplet includes an interconnect buffer to store data received via the interconnect fabric.
3. The general-purpose graphics processor as in claim 2 , wherein the first chiplet includes an interconnect cache coupled with a first fabric interface node, wherein the first fabric interface node couples the first chiplet with the interconnect fabric and the interconnect cache is to cache data transmitted and received via the first fabric interface node.
4. The general-purpose graphics processor as in claim 2 , wherein the first chiplet or the second chiplet additionally includes functional units configured to perform media encode or decode operations.
5. The general-purpose graphics processor as in claim 2 , wherein the array of chiplets additionally includes a fourth chiplet.
6. The general-purpose graphics processor as in claim 5 , wherein the fourth chiplet includes a network processor and a physical network interface to an external network.
7. The general-purpose graphics processor as in claim 6 , wherein the fourth chiplet includes memory cells associated with a memory device.
8. The general-purpose graphics processor as in claim 7 , wherein the memory device is a cache memory associated with the network processor.
9. The general-purpose graphics processor as in claim 7 , wherein the first chiplet, second chiplet, and fourth chiplet are independently power gated via power management hardware logic within the active base die.
10. The general-purpose graphics processor as in claim 1 , wherein the active base die includes a cache memory and the cache memory is connected to the interconnect fabric.
11. The general-purpose graphics processor as in claim 10 , wherein the cache memory is configured as a processor-wide cache.
12. A data processing system comprising:
a 3-dimensional (3D) or 2.5 dimensional (2.5D) package assembly including:
a graphics processor comprising an active base die including hardware logic, an interconnect fabric, and an array of multiple distinctly packaged chiplets coupled with the active base die and the interconnect fabric via multiple interconnect structures, the multiple interconnect structures to enable electrical communication between the array of multiple distinctly packaged chiplets and the interconnect fabric,
wherein the array of multiple distinctly packaged chiplets include a first chiplet including a first set of functional units configured to perform general-purpose graphics processing operations, a second chiplet including a second set of functional units configured to perform graphics processing operations, and a third chiplet coupled with the first chiplet and the second chiplet, the third chiplet including a cache memory that is shared by the first chiplet and the second chiplet, and
wherein the interconnect fabric is to receive a message or signal from a first fabric interface node associated with the first chiplet and relay the message or signal to a second fabric interface node associated with the second chiplet.
13. The data processing system as in claim 12 , wherein the interconnect fabric is to transmit messages or signals via multiple virtual channels over multiple physical links of the interconnect fabric.
14. The data processing system as in claim 13 , wherein the interconnect fabric is to transmit data for a single virtual channel across multiple physical links of the interconnect fabric.
15. The data processing system as in claim 13 , wherein the interconnect fabric is to transmit data for multiple virtual channels across a single physical link of the interconnect fabric.
16. The data processing system as in claim 13 , wherein a physical link of the multiple physical links is power gated when the physical link is idle via power management hardware logic within the active base die.
17. A method comprising:
via circuitry of a processor including a 3-dimensional (3D) or 2.5 dimensional (2.5) package assembly having multiple distinctly packaged chiplets coupled with an active base die including hardware logic:
generating data at a first functional unit within a first distinctly packaged chiplet of the multiple distinctly packaged chiplets, the first distinctly packaged chiplet coupled with the active base die of the processor;
caching the data in an interconnect cache within the chiplet, the interconnect cache coupled with a first fabric interface node;
transmitting the data to an interconnect fabric via the first fabric interface node;
routing the data within the interconnect fabric to a second fabric interface node;
receiving the data at the second fabric interface node;
transmitting the data to a second functional unit within a second distinctly packaged chiplet of the multiple distinctly packaged chiplets, the second distinctly packaged chiplet coupled with the active base die of the processor; and
performing an operation at the second functional unit based on received data.
18. The method as in claim 17 , additionally comprising associating the data with a virtual channel of the interconnect fabric and forwarding or switching the data based on the virtual channel.
19. The method as in claim 18 , additionally comprising:
diverging the virtual channel at the first fabric interface node;
transporting data of the virtual channel across multiple clock frequency domains within the processor using multiple physical links; and
converging the virtual channel at the second fabric interface node.
20. The method as in claim 18 , wherein the virtual channel is a first virtual channel and the method additionally comprises:
converging the first virtual channel at the first fabric interface node with a second virtual channel;
transporting the first virtual channel and the second virtual channel across multiple clock frequency domains within the processor using a single physical link; and
diverging the first virtual channel at the second virtual channel at the second fabric interface node.