Nvidia Corporation

James M Van Dyke

John Stephen Montrym

Brian D Hutsell

John H Edmondson

Sean Jeffrey Treichler

Steven E Molnar

Douglas A Voorhies

Michael B Nagy

Michael F Harris

Kun Fang

Jerome Francis Duluk, Jr

Brad W Simeral

James Leroy Deming

Michael B Diamond

Brian Matthew Fahs

Paolo Enrique Sabella

Mark Hairgrove

John M Danskin

Cameron Buschardt

Lucien Dunning

John Mashey

Nicholas J Foskett

Chenghuan Jia

James E Margeson, Iii

Edward Colton Greene

Prakash Bangalore Prabhakar

Charles Chew-Yuen Young

Stephen Donald Lew

William P Newhall, Jr

Bengt-Olaf Schneider

Anthony Michael Tamasi

Lingfeng Yuan

Bruce Cory

Stefan Scotzniovsky

Sean J Treicher

Jim E Margeson, Iii

A system in having M memory controllers between a first memory and a second memory having N operative memory slices, where N and M are not evenly divisible, includes logic to operate the M memory controllers to linearly distribute addresses of the second memory across the N operative memory slices. The system may be utilized in commercial applications such as data centers, autonomous vehicles, and machine learning.

Addressing cache slices in a last level cache

An addressing scheme in systems utilizing a number of operative memory slices in a last level cache that is not evenly divisible by a number of memory channels utilizes the operative slices exposes the full last level cache bandwidth and capacity to data processing logic in a high-performance graphics system.

Techniques are disclosed for allocating a global memory space defined within physical memory devices into strided memory space(s) (SMS) and partition memory space(s) (PMS). In an embodiment, a SMS is mapped across all of the devices, and a PMS is mapped to a subset of the devices to ensure resource isolation between separate PMSs. Typically, a memory space is allocated in unit sizes. When the locations mapped to most of the SMS align to an integer number of the unit size, a common boundary can be formed between the SMS and the one or more PMSs in each of the devices. Such a boundary can advantageously minimize a region of locations that are not available for allocation in the global memory spaces. In an embodiment, when a strided allocation is not an integer number of the unit size, a remainder is mapped to locations for one or more PMSs.

Techniques for efficiently partitioning memory

One embodiment of the present invention sets forth a computer-implemented method for migrating a memory page from a first memory to a second memory. The method includes determining a first page size supported by the first memory. The method also includes determining a second page size supported by the second memory. The method further includes determining a use history of the memory page based on an entry in a page state directory associated with the memory page. The method also includes migrating the memory page between the first memory and the second memory based on the first page size, the second page size, and the use history.

Migrating pages of different sizes between heterogeneous processors

One embodiment of the present invention includes a hash selector that facilitates performing effective hashing operations. In operation, the hash selector creates a transformation matrix that reflects specific optimization criteria. For each hash value, the hash selector generates a potential hash value and then computes the rank of a submatrix included in the transformation matrix. Based on this rank in conjunction with the optimization criteria, the hash selector either re-generates the potential hash value or accepts the potential hash value. Advantageously, the optimization criteria may be tailored to create desired correlations between input patterns and the results of performing hashing operations based on the transformation matrix. Notably, the hash selector may be configured to efficiently and reliably incrementally generate a transformation matrix that, when applied to certain strides of memory addresses, produces a more uniform distribution of accesses across caches lines than previous approaches to memory addressing.

Selecting hash values based on matrix rank

One embodiment of the present invention sets forth a technique for computing dynamic random access memory (DRAM) addresses from linear physical addresses for memory subsystems implementing integral power of two virtual page sizes, and an arbitrary number of available partitions. Each DRAM address comprises a row address, column address, bank address, and partition address. The linear physical address is used to generate to the DRAM address in units of a DRAM bank size. Address scrambling may be implemented to overcome transient access contention to specific DRAM pages by multiple client modules.

System and methods for distributing a power-of-two virtual memory page across a non-power-of two number of DRAM partitions

The present invention systems and methods enable configuration of functional components in integrated circuits. A present invention system and method can flexibly change the operational characteristics of functional components in an integrated circuit die based upon a variety of factors, including if the die has a defective component. An indication of the defective functional component identification is received. A determination is made if the defective functional component is one of a plurality of similar functional components that can provide the same functionality. The other similar components can be examined to determine if they are parallel components to the defective functional component. The defective functional component is disabled if it is one of the plurality of similar functional components and another component can handle the workflow that would otherwise be assigned to the defective component. Workflow is diverted from the disabled component to other similar functional components.

Integrated circuit configuration system and method

The present invention systems and methods enable configuration of functional components in integrated circuits. A present invention system and method can flexibly change the operational characteristics of functional components in an integrated circuit die based upon a variety of factors. In one embodiment, manufacturing yields, compatibility characteristics, performance requirements, and system health (e.g., the number of components operating properly) are factored into changes to the operational characteristics of functional components. In one exemplary implementation, the changes to operational characteristics of a functional component are coordinated with changes to other functional components. Workflow scheduling and distribution is also adjusted based upon the changes to the operational characteristics of the functional components. For example, a functional component configuration controller changes the operational characteristics settings and provides an indication to a workflow distribution component. The workflow distribution component changes the workflow schedule based upon the operational characteristics settings (e.g., work flow is diverted to or away from functional components).

System and method for configuring semiconductor functional circuits

The present invention systems and methods facilitate increased die yields by flexibly changing the operational characteristics of functional components in an integrated circuit die. The present invention system and method enable integrated circuit chips with defective functional components to be salvaged. Defective functional components in the die are disabled in a manner that maintains the basic functionality of the chip. A chip is tested and a functional component configuration process is performed on the chip based upon results of the testing. If an indication of a defective functional component is received, the functional component is disabled. Workflow is diverted from disabled functional components to enabled functional components.

System and method for increasing die yield

The present invention systems and methods facilitate configuration of functional components included in a remotely located integrated circuit die. In one exemplary implementation, a die functional component reconfiguration request process is engaged in wherein a system requests a reconfiguration code from a remote centralized resource. A reconfiguration code production process is executed in which a request for a reconfiguration code and a permission indicator are received, validity of permission indicator is analyzed, and a reconfiguration code is provided if the permission indicator is valid. A die functional component configuration process is performed on the die when an appropriate reconfiguration code is received by the die. The functional component configuration process includes directing alteration of a functional component configuration. Workflow is diverted from disabled functional components to enabled functional components.

System and method for remotely configuring semiconductor functional circuits

A memory cell reconfiguration process is performed in accordance with the operational characteristic settings determined based upon the results of analysis and/or testing of memory cell operations. The memory circuit can include a plurality of memory cells and memory cell configuration controller. The memory cells store information associated with a variety of operations. The memory cell configuration controller coordinates selective enablement and disablement of each of the plurality of memory cells, which can be done on a subset or group basis (e.g., enables and/or disables memory cells on a word length or row by row basis). The address mapping can be adjusted so that the memory space appears continuous to external components. The memory cell configuration controller can also forward configuration information to upstream and/or downstream components that can adjust operations to compensate for the memory cell configuration (e.g., to prevent overflow).

Functional component compensation reconfiguration system and method

A shared resource management system and method are described. In one embodiment a shared resource management system includes a plurality of engines, a shared resource, and a shared resource management unit. In one exemplary implementation the shared resource is a memory and the shared resource management unit is a memory management unit (MMU). The plurality of engines perform processing. The shared resource supports the processing. For example, a memory stores information and instructions for the engines. The shared resource management unit manages memory operations and handles access requests associated with compressed data.

Compressed data access system and method

A mapping engine maps general processing clusters (GPCs) within a parallel processing subsystem to screen tiles on a display screen based on the number of enabled streaming multiprocessors (SMs) within each GPC. A given GPC then generates pixels for the screen tiles to which the GPC is mapped. One advantage of the disclosed technique is a given GPC performs a fraction of the processing tasks associated with the parallel processing subsystem that is roughly proportional to the fraction of SMs included within the GPC.

Programmable compute engine screen mapping

Systems and methods for determining a compression tag state prior to memory client arbitration may reduce the latency for memory accesses. A compression tag is associated with each portion of a surface stored in memory and indicates whether or not the data stored in each portion is compressed or not. A client uses the compression tags to construct memory access requests and the size of each request is based on whether or not the portion of the surface to be accessed is compressed or not. When multiple clients access the same surface the compression tag reads are interlocked with the pending memory access requests to ensure that the compression tags provided to each client are accurate. This mechanism allows for memory bandwidth optimizations including reordering memory access requests for efficient access.

Growing community of inventors

Austin, TX, United States of America

James M Van Dyke