Power gating a portion of a cache memory

1. A processor comprising:

a plurality of tiles, each tile including a core and a tile cache hierarchy, the tile cache hierarchy including a first level cache and a second level cache, wherein each of the first level cache and the second level cache is physically private to the tile; and

a controller coupled to the plurality of tiles, the controller including a cache power control logic to receive utilization information regarding the core and the tile cache hierarchy of a first tile and to cause the second level cache of the first tile to be independently power gated when a miss rate of the first level cache is less than a first threshold, based at least in part on the utilization information.

2. The processor of claim 1 , wherein the cache power control logic is to independently power gate the second level cache while the core is in an active state.

3. The processor of claim 1 , wherein the cache power control logic is to cause the second level cache to be power gated while the core is in a low power state and a core of at least another of the plurality of tiles is in an active state.

4. The processor of claim 1 , wherein the cache power control logic is to determine an energy break even value and to cause the second level cache to be power gated when an estimated idle duration for the core is greater than the energy break even value.

5. The processor of claim 4 , wherein the cache power control logic is to calculate the energy break even value using static values and dynamic values associated with the processor.

6. The processor of claim 4 , wherein the cache power control logic is to prevent the second level cache from being power gated when the estimated idle duration is greater than the energy break even value, based on quality of service (QoS) information associated with a workload being executed on the processor.

7. The processor of claim 1 , wherein when the second level cache is power gated, the first level cache is to issue a read request to a system memory responsive to a miss in the first level cache.

8. The processor of claim 1 , wherein at least a portion of information stored in the second level cache is to be flushed to a different second level cache of a different one of the plurality of tiles before the second level cache is power gated.

9. A machine-readable non-transitory medium having stored thereon instructions, which if performed by a machine cause the machine to perform a method comprising:

determining an estimated idle duration for a core of a multicore processor, the core associated with a private cache hierarchy including a first level cache, a mid-level cache (MLC) and a last level cache (LLC);

determining a time value for a break even condition for a first workload and storing the time value in a storage of the multicore processor, wherein the break even condition is a minimum amount of time for the LLC to be in a low power state; and

power gating the LLC when the estimated idle duration is greater than the time value, while at least one other core of the multicore processor associated with another private cache hierarchy remains in an active state.

10. The machine-readable non-transitory medium of claim 9 , wherein the method further comprises:

receiving performance metric information from the private cache hierarchy; and

comparing the performance metric information to a threshold value.

11. The machine-readable non-transitory medium of claim 10 , wherein the method further comprises power gating the LLC based on the comparison of the performance metric information to the threshold value.

12. The machine-readable non-transitory medium of claim 10 , wherein the performance metric information comprises a miss rate of the MLC and power gating the LLC when the miss rate is less than the threshold value.

13. The machine-readable non-transitory medium of claim 9 , wherein the method further comprises determining the estimated idle duration based at least in part on an idle duration report received from a device coupled to the multicore processor.

14. The machine-readable non-transitory medium of claim 9 , wherein the method further comprises determining a second time value for a break even condition associated with a second workload, and using the second time value to determine whether to power gate the LLC during execution of the second workload on the multicore processor.

15. A system comprising:

a multicore processor having a plurality of tiles, each tile including a core and a tile cache hierarchy, the tile cache hierarchy including a first level cache, a mid-level cache (MLC) and a last level cache (LLC), wherein each of the first level cache, the MLC and the LLC are private to the tile, and a cache power control logic to receive performance metric information regarding the tile cache hierarchy of a first tile of the plurality of tiles and to cause the LLC of the first tile to be power gated when a miss rate of the MLC is less than a first threshold based at least in part on the performance metric information, while the core, the first level cache and the MLC of the first tile remain in an active state; and

a dynamic random access memory (DRAM) coupled to the multicore processor.

16. The system of claim 15 , wherein the cache power control logic is to determine an energy break even value and to cause the LLC to be power gated when the energy break even value is less than an estimated idle duration for the core of the first tile.

17. The system of claim 16 , wherein the cache power control logic is to prevent the LLC from being power gated when the energy break even value is less than the estimated idle duration, based on quality of service (QoS) information associated with a workload being executed on the core of the first tile.