CUDA Programming Concepts Quiz

A type of memory accessible by multiple threads in a parallel computing architecture.

A technique for optimizing matrix multiplication by dividing the matrix into smaller tiles.

A part of a computing system that coordinates the execution of tasks on a device.

Factors that influence performance based on the size of computations and data.

Setting properties for CUDA kernels, such as the number of threads or blocks.

A computer architecture with multiple independent processing units (cores) on a single chip.

Single Instruction, Single Data. A computer architecture where a single instruction is executed on a single piece of data.

Single Instruction, Multiple Data. A computer architecture where a single instruction is executed on multiple pieces of data.

Multiple Instructions, Multiple Data. A computer architecture where multiple instructions are executed on multiple pieces of data.

A classification of computer architectures based on the number of instructions and data streams.

A multicore processor with identical cores.

A multicore processor with different types of cores.

A central processing unit (CPU) with multiple independent processing units (cores) on a single integrated circuit.

Faster processing speed and enhanced performance due to parallel execution of tasks.

Increased complexity in programming, needing specific programming techniques to take advantage of them.

A single-data operation. Each data element is processed one at a time.

A multicore architecture where all cores share the same main memory. Data is accessible by all cores.

A multicore architecture where each core has its own separate memory. Data must be moved between cores explicitly.

Writing programs that can take advantage of multiple cores in a processor. Doing calculations on multiple processors simultaneously.

An application programming interface (API) model that supports multi-platform shared memory parallelism in C, C++, and Fortran. Programming model that assists parallel programming.

A way to manage threads that aren't bound to a specific core, often used in parallel systems and is dependent upon availability.

Allocates tasks dynamically to available processors, efficiently utilizing processors.

A method where processors communicate by exchanging explicit messages to distribute a task.

Evaluating how well a parallel program performs by calculating factors that affect its performance.

A formula that quantifies the theoretical speedup in a system with certain aspects parallelizable.

Describes the size and complexity of tasks in parallel programs. Small tasks mean high granularity, large tasks mean low granularity.

The ability of multiple tasks to execute simultaneously to speed up a task.

A function in OpenCL that allows you to synchronize execution between different queues.

A feature in OpenCL that provides information about the performance of your OpenCL kernels and commands.

Specific data points that the OpenCL profiling interface provides, such as the start time and end time of a kernel execution.

A specific subset of the C programming language specifically designed for writing OpenCL kernels.

Key features of OpenCL C that make it suitable for parallel programming on GPUs.

Limitations imposed by OpenCL C to ensure efficient execution on different devices.

Additional functionality that OpenCL implementations can provide beyond the standard.

OpenCL's ability to interact with OpenGL, facilitating seamless data transfer between the two APIs.

The process of writing, compiling, and executing OpenCL programs.

The process of selecting appropriate hardware for your OpenCL program to run on.

The process of allocating memory on the GPU to store data used by OpenCL kernels.

Various memory allocation methods available in OpenCL, such as buffer memory or image memory.

The process of copying data between the host computer (CPU) and the GPU.

A representation of an OpenCL program that contains code for your kernels.

Settings for how your OpenCL kernels are executed on the GPU, such as the number of work-items or work-groups.

A method where events trigger actions in a coordinated manner, allowing tasks to be scheduled and executed based on specific occurrences.

A parallel programming pattern where a single program is executed on multiple data sets simultaneously.

A parallel programming pattern where multiple programs run on multiple data sets, allowing for independent processing and data exploration.

Divides the iterations of a loop among multiple processors to speed up execution, ideal for tasks that can be broken down into independent steps.

A structured parallel programming approach where a 'master' program distributes tasks to 'worker' programs, each processing data independently.

A pattern where tasks are divided into smaller units that can be run concurrently and later joined together to produce a final result.

A parallel programming paradigm that efficiently processes large datasets, breaking tasks into independent 'map' and 'reduce' steps.

Dividing data into smaller chunks for parallel processing, ensuring efficient data distribution among computing units.

The transfer of data between processing units in a parallel computing environment, essential for coordinating work.

Combining intermediate results from multiple processors into a final result, often used in parallel algorithms.

A sequence that can be divided into two monotonic subsequences (ascending and descending) that are concatenated in a specific order.

A sorting algorithm that efficiently sorts a bitonic sequence in parallel, taking advantage of bitonic sequence properties.

A C++ template library designed for efficient GPU programming, providing high-level functions to simplify parallel computations.

Instructions embedded in code that guide the compiler to optimize code for parallel execution, leveraging resources like GPUs or multicore CPUs.

Special instructions added to code to tell the compiler how to parallelize tasks for execution on GPUs or other accelerators.

OpenACC allows writing code that can run on different hardware platforms without significant changes, making it more versatile.

OpenACC uses a similar syntax and concepts to OpenMP, making it easier to learn for programmers familiar with that standard.

OpenACC offers a true open standard, meaning it's not tied to a specific vendor or hardware platform.

The kernels directive marks a section of code intended for execution on the accelerator.

A simple mathematical operation (y = a * x + y) used to demonstrate how to implement parallelization with OpenACC.

Similar to kernels, but allows splitting a loop across multiple cores to run concurrently.

Specifies that a loop should be parallelized by executing iterations in parallel, leveraging multiple cores.