Graphics processing integrated circuit package
An integrated circuit (IC) package apparatus is disclosed. The IC package includes one or more processing units and a bridge, mounted below the one or more processing unit, including one or more arithmetic logic units (ALUs) to perform atomic operations.
1. An integrated circuit (IC) package comprising:
a plurality of graphics processing unit (GPU) chips, each GPU chip of the plurality of GPU chips being coupled with memory for storage of data;
a plurality of channels between GPU chips of the plurality of GPU chips, the plurality of channels to facilitate exchange of data between the GPU chips;
a priority mechanism, the priority mechanism to establish a plurality of priority levels for the data exchanged between the plurality of GPU chips; and
an activation component to activate the plurality of channels based on a data traffic type for data to be exchanged via the plurality of channels and a priority level established by the priority mechanism.
2. The IC package of claim 1 , wherein the plurality of channels includes a plurality of separate physical connections between GPU chips of the plurality of GPU chips.
3. The IC package of claim 1 , wherein the plurality of channels includes a plurality of virtual channels between GPU chips of the plurality of GPU chips.
4. The IC package of claim 1 , wherein the activation component is to determine which memory surfaces of the memory can be attached to a particular priority level of the plurality of priority levels.
5. The IC package of claim 4 , wherein data traffic types and memory surfaces are mapped into particular priority levels of the plurality of priority levels to enable data flow with non-blocking conditions.
6. The IC package of claim 1 , wherein each of the plurality of GPU chips includes a plurality of processing elements.
7. The IC package of claim 6 , wherein the processing elements of one or more of the plurality of GPU chips include streaming multiprocessors (SMs).
8. The IC package of claim 1 , further comprising a central processing unit (CPU) chip.
9. A system comprising:
a processor integrated circuit (IC) including one or more processors;
a memory IC including memory for storage of data;
a plurality of graphics processing unit (GPU) ICs, each GPU IC of the plurality of GPU ICs being coupled with the memory;
a plurality of channels between GPU ICs of the plurality of GPU ICs, the plurality of channels to facilitate exchange of data between the GPU ICs;
a priority mechanism, the priority mechanism to establish a plurality of priority levels for the data exchanged between the plurality of GPU ICs; and
an activation component to activate the plurality of channels based on a data traffic type for data to be exchanged via the plurality of channels and a priority level established by the priority mechanism.
10. The system of claim 9 , the plurality of channels including a plurality of separate physical connections between GPU ICs of the plurality of GPU ICs.
11. The system of claim 9 , the plurality of channels including a plurality of virtual channels between GPU ICs of the plurality of GPU ICs.
12. The system of claim 9 , wherein the activation component is to determine which memory surfaces of the memory can be attached to a particular priority level.
13. The system of claim 12 , wherein data traffic types and memory surfaces are mapped into particular priority levels of the plurality of priority levels to enable data flow with non-blocking conditions.
14. The system of claim 9 , further comprising an accelerator IC to accelerate graphics operations.
15. One or more non-transitory computer-readable storage mediums having stored thereon executable computer program instructions that, when executed by one or more processors, cause the one or more processors to perform operations comprising:
establishing a plurality of priority levels for data exchanged between a plurality of GPUs in an integrated circuit (IC) package, the plurality of GPUs being connected by plurality of channels to facilitate exchange of data between the GPUs, each GPU of the plurality of GPUs being coupled with a memory for storage of data;
activating the plurality of channels based on a data traffic type for data to be exchanged between the plurality of GPUs and the established priority levels; and
transmitting data between the plurality of GPUs, wherein transmission of the data is prioritized according to the established priority levels.
16. The one or more non-transitory storage mediums of claim 15 , the plurality of channels including a plurality of virtual channels between GPUs of the plurality of GPUs.
17. The one or more non-transitory storage mediums of claim 15 , further comprising executable computer program instructions that, when executed by the one or more processors, cause the one or more processors to perform operations comprising:
determining which memory surfaces of the memory can be attached to a particular priority level of the plurality of priority levels.
18. The one or more non-transitory storage mediums of claim 17 , further comprising executable computer program instructions that, when executed by the one or more processors, cause the one or more processors to perform operations comprising:
mapping data traffic types and memory surfaces into particular priority levels of the plurality of priority levels to enable data flow with non-blocking conditions.