Abstract: A technique includes, in response to feedback received from ongoing link training with an endpoint device over a serial communication link, a processor identifying a first training set sequence to be communicated as part of the ongoing link training. The technique includes the processor selecting a first state machine of a plurality of state machines; and the processor programming the selected first state machine to communicate the first training set sequence to the serial communication link. The programming of the first state machine includes programming the first state machine with a data pattern that is associated with the first training set sequence and programming the first state machine with a condition to regulate a number of times that the state machine communicates the data pattern to the serial communication link.

Abstract: A technique includes, in response to feedback received from ongoing link training with an endpoint device over a serial communication link, a processor identifying a first training set sequence to be communicated as part of the ongoing link training. The technique includes the processor selecting a first state machine of a plurality of state machines; and the processor programming the selected first state machine to communicate the first training set sequence to the serial communication link. The programming of the first state machine includes programming the first state machine with a data pattern that is associated with the first training set sequence and programming the first state machine with a condition to regulate a number of times that the state machine communicates the data pattern to the serial communication link.

Abstract: In an embodiment, a memory scheduler is provided to process memory requests. The memory scheduler may comprise: a plurality of arbitrators that each select memory requests according to age of the memory requests and whether resources are available for the memory requests; and a second-level arbitrator that selects, for an arbitration round, a series of memory requests made available by the plurality of arbitrators, wherein the second-level arbitrator begins the arbitration round by selecting a memory request from a least recently used (LRU) arbitrator of the plurality of arbitrators.

Abstract: One embodiment is a method of dynamically adjusting a rate at which a dynamic random access memory (“DRAM”) module is refreshed in a computer system. The method comprises monitoring a plurality of system conditions; detecting a change in at least one of the monitored system conditions; responsive to the detection, determining an optimum refresh rate for a current state of the computer system; and setting the refresh rate to the determined optimum refresh rate.

Abstract: In an embodiment, a memory scheduler is provided to process memory requests. The memory scheduler may comprise: a plurality of arbitrators that each select memory requests according to age of the memory requests and whether resources are available for the memory requests; and a second-level arbitrator that selects, for an arbitration round, a series of memory requests made available by the plurality of arbitrators, wherein the second-level arbitrator begins the arbitration round by selecting a memory request from a least recently used (LRU) arbitrator of the plurality of arbitrators.

Abstract: A unified parallel processing architecture connects together an extendible number of clusters of multiple numbers of processors to create a high performance parallel processing computer system. Multiple processors are grouped together into four or more physically separable clusters, each cluster having a common cluster shared memory that is symmetrically accessible by all of the processors in that cluster; however, only some of the clusters are adjacently interconnected. Clusters are adjacently interconnected to form a floating shared memory if certain memory access conditions relating to relative memory latency and relative data locality can create an effective shared memory parallel programming environment. A shared memory model can be used with programs that can be executed in the cluster shared memory of a single cluster, or in the floating shared memory that is defined across an extended shared memory space comprised of the cluster shared memories of any set of adjacently interconnected clusters.

Abstract: A unified parallel processing architecture connects together an extendible number of clusters of multiple numbers of processors to create a high performance parallel processing computer system. Multiple processors are grouped together into four or more physically separable clusters, each cluster having a common cluster shared memory that is symmetrically accessible by all of the processors in that cluster; however, only some of the clusters are adjacently interconnected. Clusters are adjacently interconnected to form a floating shared memory if certain memory access conditions relating to relative memory latency and relative data locality can create an effective shared memory parallel programming environment. A shared memory model can be used with programs that can be executed in the cluster shared memory of a single cluster, or in the floating shared memory that is defined across an extended shared memory space comprised of the cluster shared memories of any set of adjacently interconnected clusters.

Abstract: A distributed architecture for the input/output system for a multiprocessor system provides for equal and democratic access to all shared hardware resources by both the processors and the external interface ports of the multiprocessor system. This allows one or more input/output concentrators attached to the external interface ports to have complete access to all of the shared hardware resources across the multiprocessor system without requiring processor intervention. The distributed input/output system provides for communication of data and control information between a set of common shared hardware resources and a set of external data sources. The result is a highly parallel multiprocessor system that has multiple parallel high performance input/output ports capable of operating in a distributed fashion.