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 and distinctly 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 graphics processor comprising:
a package assembly comprising a plurality of distinct chiplets in a 2.5-dimensional (2.5D) or 3-dimensional (3D) arrangement, the plurality of distinct chiplets including:
a first chiplet including a first processing cluster having a first plurality of graphics processing resources within a first clock domain, the first plurality of graphics processing resources configured for a first power and performance target; and
a second chiplet including a second processing cluster having a second plurality of graphics processing resources within a second clock domain, the second plurality of graphics processing resources configured for a second power and performance target, the second chiplet coupled with the first chiplet via a clock crossing interconnect network.
2. The graphics processor of claim 1 , the first chiplet to generate data at a graphics processing resource of the first plurality of graphics processing resources and transmit the data to a graphics processing resource of the second plurality of graphics processing resources via the clock crossing interconnect network.
3. The graphics processor of claim 2 , wherein to transmit the data to the graphics processing resource of the second plurality of graphics processing resources via the clock crossing interconnect network includes to cache the data in an interconnect cache associated with the clock crossing interconnect network.
4. The graphics processor of claim 2 , wherein to transmit the data to the graphics processing resource of the second plurality of graphics processing resources via the clock crossing interconnect network includes to transport the data from the first clock domain to the second clock domain.
5. The graphics processor of claim 4 , wherein to transport the data from the first clock domain to the second clock domain includes to transport the data from a first transport layer of the clock crossing interconnect network to a second transport layer of the clock crossing interconnect network.
6. The graphics processor of claim 4 , a processing resource of the second plurality of graphics processing resources configured to perform an operation based on received data.
7. The graphics processor of claim 1 , comprising a third chiplet coupled with the first chiplet and the second chiplet, the third chiplet including a cache memory accessible to the first plurality of graphics processing resources and the second plurality of graphics processing resources.
8. The graphics processor of claim 1 , the package assembly comprising a base chiplet coupled with the first chiplet and the second chiplet.
9. The graphics processor of claim 8 , wherein the base chiplet includes at least a portion of the clock crossing interconnect network.
10. The graphics processor of claim 1 , wherein the first plurality of graphics processing resources includes a matrix accelerator.
11. A graphics processing system comprising:
a graphics memory device; and
a graphics processor coupled with the graphics memory device, the graphics processor comprising a package assembly including a plurality of distinct chiplets in a 2.5-dimensional (2.5D) or 3-dimensional (3D) arrangement, the plurality of distinct chiplets including:
a first chiplet including a first processing cluster having a first plurality of graphics processing resources within a first clock domain, the first plurality of graphics processing resources configured for a first power and performance target; and
a second chiplet including a second processing cluster having a second plurality of graphics processing resources within a second clock domain, the second plurality of graphics processing resources configured for a second power and performance target, the second chiplet coupled with the first chiplet via a clock crossing interconnect network.
12. The graphics processing system of claim 11 , the first chiplet to generate data at a graphics processing resource of the first plurality of graphics processing resources and transmit the data to a graphics processing resource of the second plurality of graphics processing resources via the clock crossing interconnect network.
13. The graphics processing system of claim 12 , wherein to transmit the data to the graphics processing resource of the second plurality of graphics processing resources via the clock crossing interconnect network includes to cache the data in an interconnect cache associated with the clock crossing interconnect network.
14. The graphics processing system of claim 12 , wherein to transmit the data to the graphics processing resource of the second plurality of graphics processing resources via the clock crossing interconnect network includes to transport the data from the first clock domain to the second clock domain.
15. The graphics processing system of claim 14 , wherein to transport the data from the first clock domain to the second clock domain includes to transport the data from a first transport layer of the clock crossing interconnect network to a second transport layer of the clock crossing interconnect network.
16. The graphics processing system of claim 14 , a processing resource of the second plurality of graphics processing resources configured to perform an operation based on received data.
17. The graphics processing system of claim 11 , comprising a third chiplet coupled with the first chiplet and the second chiplet, the third chiplet including a cache memory accessible to the first plurality of graphics processing resources and the second plurality of graphics processing resources, the cache memory to cache accesses to the graphics memory device.
18. The graphics processing system of claim 11 , the package assembly comprising a base chiplet coupled with the first chiplet and the second chiplet.
19. The graphics processing system of claim 18 , wherein the base chiplet includes at least a portion of the clock crossing interconnect network.
20. The graphics processing system of claim 11 , wherein the first plurality of graphics processing resources includes a matrix accelerator.