Back to results
US 11,698,853 B2Granted
Saturating local cache in memory-compute systems
AssigneeMicron Technology, Inc.
InventorsSkyler Arron Windh, Randall Meyer
Life of this patent
Estimating the term…
Abstract
Latency in a node-based compute-near-memory system can be problematic. A solution to the problem can include or use a dedicated software-based cache at each node. The cache can be configured to store information received from each of the other nodes in the system. In an example, the cache can be populated during a breadth first search algorithm to store frontier information from each of the other nodes.
Claims20
A method comprising:
- establishing an N-block cache at a first memory-compute node in a compute-near-memory system that includes N different hardware nodes, wherein the first memory-compute node is one of the N different hardware nodes and wherein the N different hardware nodes are coupled by a compute fabric;
- for each initial request from the first memory-compute node to read information from a particular one of the N different nodes, populating a corresponding one of the N blocks in the cache with frontier information received from the particular one of the N different nodes, wherein the frontier information indicates whether a parent of the particular one of the N different nodes was found during a prior search of some or all of the nodes coupled by the compute fabric;
- and
- for an initial request from the first memory-compute node to read target information from a first node of the N different nodes, retrieving the target information from the first node via the compute fabric and populating a first one of the N blocks in the cache with the target information retrieved from the first node, and for a subsequent request from the first memory-compute node to read the target information from the first node, reading the target information from the first one of the N blocks in the cache.
1 more element
An apparatus comprising:
- a first memory-compute node of multiple memory-compute nodes in a compute-near-memory system, wherein the first memory-compute node comprises a processor configured to perform operations comprising:
- establishing an N-block software cache, wherein each of the N blocks corresponds to a respective one of the memory-compute nodes in the compute-near-memory system;
- and
- populating the respective blocks in the cache using frontier information received from the respective memory-compute nodes in the compute-near-memory system, wherein populating the respective blocks comprises prioritizing a search algorithm according to a connectedness of the nodes in the system.
1 more element
A memory-compute system comprising:
- N different memory-compute nodes coupled by a scale fabric to provide a compute fabric, wherein each of the N nodes comprises a hybrid threading processor and a hybrid threading fabric, wherein a first memory-compute node of the N different memory-compute nodes comprises a processor configured to perform operations comprising:
- establishing an N-block software cache, wherein each of the N blocks corresponds to a respective one of the memory-compute nodes coupled by the scale fabric;
- populating the respective blocks in the cache using frontier information received from the respective nodes, wherein the frontier information indicates whether a parent of a corresponding one of the N nodes was found during a prior search of some or all of the nodes coupled by the hybrid threading fabric;
- and
- for an initial request from the first memory-compute node to read target information from a second node of the N different memory-compute nodes, retrieving the target information from the second node via the compute fabric and populating a first one of the N blocks in the cache with the target information retrieved from the second node, and for a subsequent request from the first memory-compute node to read the target information from the second node, reading the target information from the first one of the N blocks in the cache.
2 more elements
Dependencies are inferred from the claim text.