Single instruction, multiple threads

Last updated

Single instruction, multiple thread (SIMT) is an execution model used in parallel computing where single instruction, multiple data (SIMD) is combined with multithreading.



The processors, say a number p of them, seem to execute many more than p tasks. This is achieved by each processor having multiple "threads" (or "work-items" or "Sequence of SIMD Lane operations"), which execute in lock-step, and are analogous to SIMD lanes. [1]

The SIMT execution model has been implemented on several GPUs and is relevant for general-purpose computing on graphics processing units (GPGPU), e.g. some supercomputers combine CPUs with GPUs.

SIMT was introduced by Nvidia: [2] [3]

Nvidia's Tesla GPU microarchitecture (first available November 8, 2006 as implemented in the "G80" GPU chip) introduced the single-instruction multiple-thread (SIMT) execution model where multiple independent threads execute concurrently using a single instruction.

ATI Technologies (now AMD) released a competing product slightly later on May 14, 2007, the TeraScale 1-based "R600" GPU chip.

As access time of all the widespread RAM types (e.g. DDR SDRAM, GDDR SDRAM, XDR DRAM, etc.) is still relatively high, engineers came up with the idea to hide the latency that inevitably comes with each memory access. Strictly, the latency-hiding is a feature of the zero-overhead scheduling implemented by modern GPUs. This might or might not be considered to be a property of 'SIMT' itself.

SIMT is intended to limit instruction fetching overhead, [4] i.e. the latency that comes with memory access, and is used in modern GPUs (such as those of Nvidia and AMD) in combination with 'latency hiding' to enable high-performance execution despite considerable latency in memory-access operations. This is where the processor is oversubscribed with computation tasks, and is able to quickly switch between tasks when it would otherwise have to wait on memory. This strategy is comparable to multithreading in CPUs (not to be confused with multi-core). [5] As with SIMD, another major benefit is the sharing of the control logic by many data lanes, leading to an increase in computational density. One block of control logic can manage N data lanes, instead of replicating the control logic N times.

A downside of SIMT execution is the fact that thread-specific control-flow is performed using "masking", leading to poor utilization where a processor's threads follow different control-flow paths. For instance, to handle an IF-ELSE block where various threads of a processor execute different paths, all threads must actually process both paths (as all threads of a processor always execute in lock-step), but masking is used to disable and enable the various threads as appropriate. Masking is avoided when control flow is coherent for the threads of a processor, i.e. they all follow the same path of execution. The masking strategy is what distinguishes SIMT from ordinary SIMD, and has the benefit of inexpensive synchronization between the threads of a processor. [6]

Nvidia CUDA OpenCL Hennessy & Patterson [7]
ThreadWork-itemSequence of SIMD Lane operations
Warp Wavefront Thread of SIMD Instructions
BlockWorkgroupBody of vectorized loop
GridNDRangeVectorized loop

See also

Related Research Articles

Superscalar processor CPU that implements instruction-level parallelism within a single processor

A superscalar processor is a CPU that implements a form of parallelism called instruction-level parallelism within a single processor. In contrast to a scalar processor that can execute at most one single instruction per clock cycle, a superscalar processor can execute more than one instruction during a clock cycle by simultaneously dispatching multiple instructions to different execution units on the processor. It therefore allows for more throughput than would otherwise be possible at a given clock rate. Each execution unit is not a separate processor, but an execution resource within a single CPU such as an arithmetic logic unit.

Parallel computing programming paradigm in which many calculations or the execution of processes are carried out simultaneously

Parallel computing is a type of computation in which many calculations or the execution of processes are carried out simultaneously. Large problems can often be divided into smaller ones, which can then be solved at the same time. There are several different forms of parallel computing: bit-level, instruction-level, data, and task parallelism. Parallelism has long been employed in high-performance computing, but it's gaining broader interest due to the physical constraints preventing frequency scaling. As power consumption by computers has become a concern in recent years, parallel computing has become the dominant paradigm in computer architecture, mainly in the form of multi-core processors.

Flynn's taxonomy is a classification of computer architectures, proposed by Michael J. Flynn in 1966. The classification system has stuck, and has been used as a tool in design of modern processors and their functionalities. Since the rise of multiprocessing central processing units (CPUs), a multiprogramming context has evolved as an extension of the classification system.

Graphics processing unit specialized electronic circuit; graphics accelerator

A graphics processing unit (GPU) is a specialized electronic circuit designed to rapidly manipulate and alter memory to accelerate the creation of images in a frame buffer intended for output to a display device. GPUs are used in embedded systems, mobile phones, personal computers, workstations, and game consoles. Modern GPUs are very efficient at manipulating computer graphics and image processing. Their highly parallel structure makes them more efficient than general-purpose central processing units (CPUs) for algorithms that process large blocks of data in parallel. In a personal computer, a GPU can be present on a video card or embedded on the motherboard. In certain CPUs, they are embedded on the CPU die.

General-purpose computing on graphics processing units is the use of a graphics processing unit (GPU), which typically handles computation only for computer graphics, to perform computation in applications traditionally handled by the central processing unit (CPU). The use of multiple video cards in one computer, or large numbers of graphics chips, further parallelizes the already parallel nature of graphics processing. In addition, even a single GPU-CPU framework provides advantages that multiple CPUs on their own do not offer due to the specialization in each chip.

Microarchitecture the way a given instruction set architecture (ISA) is implemented on a processor

In computer engineering, microarchitecture, also called computer organization and sometimes abbreviated as µarch or uarch, is the way a given instruction set architecture (ISA) is implemented in a particular processor. A given ISA may be implemented with different microarchitectures; implementations may vary due to different goals of a given design or due to shifts in technology.

A physics processing unit (PPU) is a dedicated microprocessor designed to handle the calculations of physics, especially in the physics engine of video games. It is an example of hardware acceleration.

In computing, hardware acceleration is the use of computer hardware specially made to perform some functions more efficiently than is possible in software running on a general-purpose CPU. Any transformation of data or routine that can be computed, can be calculated purely in software running on a generic CPU, purely in custom-made hardware, or in some mix of both. An operation can be computed faster in application-specific hardware designed or programmed to compute the operation than specified in software and performed on a general-purpose computer processor. Each approach has advantages and disadvantages. The implementation of computing tasks in hardware to decrease latency and increase throughput is known as hardware acceleration.

Stream processing is a computer programming paradigm, equivalent to dataflow programming, event stream processing, and reactive programming, that allows some applications to more easily exploit a limited form of parallel processing. Such applications can use multiple computational units, such as the floating point unit on a graphics processing unit or field-programmable gate arrays (FPGAs), without explicitly managing allocation, synchronization, or communication among those units.

CUDA parallel computing platform and programming model

CUDA is a parallel computing platform and application programming interface (API) model created by Nvidia. It allows software developers and software engineers to use a CUDA-enabled graphics processing unit (GPU) for general purpose processing – an approach termed GPGPU. The CUDA platform is a software layer that gives direct access to the GPU's virtual instruction set and parallel computational elements, for the execution of compute kernels.

In computing, Close To Metal is the name of a beta version of a low-level programming interface developed by ATI, now the AMD Graphics Product Group, aimed at enabling GPGPU computing. CTM was short-lived, and the first production version of AMD's GPGPU technology is now called AMD Stream SDK, or rather the current AMD APP SDK for Windows and Linux 32-bit and 64-bit. APP stands for "Accelerated Parallel Processing". and also targets Heterogeneous System Architecture.

Larrabee (microarchitecture) canceled Intel chip microarchitecture for GPGPU

Larrabee is the codename for a cancelled GPGPU chip that Intel was developing separately from its current line of integrated graphics accelerators. It is named after Larrabee State Park in Whatcom County, Washington, near the town of Bellingham. The chip was to be released in 2010 as the core of a consumer 3D graphics card, but these plans were cancelled due to delays and disappointing early performance figures. The project to produce a GPU retail product directly from the Larrabee research project was terminated in May 2010. The Intel MIC multiprocessor architecture announced in 2010 inherited many design elements from the Larrabee project, but does not function as a graphics processing unit; the product is intended as a co-processor for high performance computing.

Graphics Core Next codename fo a series of microarchitectures and an instruction set

Graphics Core Next (GCN) is the codename for both a series of microarchitectures as well as for an instruction set. GCN was developed by AMD for their GPUs as the successor to TeraScale microarchitecture/instruction set. The first product featuring GCN was launched in January 9, 2012.

Fermi is the codename for a graphics processing unit (GPU) microarchitecture developed by Nvidia, first released to retail in April 2010, as the successor to the Tesla microarchitecture. It was the primary microarchitecture used in the GeForce 400 series and GeForce 500 series. It was followed by Kepler, and used alongside Kepler in the GeForce 600 series, GeForce 700 series, and GeForce 800 series, in the latter two only in mobile GPUs. In the workstation market, Fermi found use in the Quadro x000 series, Quadro NVS models, as well as in Nvidia Tesla computing modules. All desktop Fermi GPUs were manufactured in 40 nm, mobile Fermi GPUs in 40 nm and 28 nm. Fermi is the oldest microarchitecture from NVIDIA that received support for the Microsoft's rendering API Direct3D 12 feature_level 11.

Pascal (microarchitecture) GPU microarchitecture developed by Nvidia

Pascal is the codename for a GPU microarchitecture developed by Nvidia, as the successor to the Maxwell architecture. The architecture was first introduced in April 2016 with the release of the Tesla P100 (GP100) on April 5, 2016, and is primarily used in the GeForce 10 series, starting with the GeForce GTX 1080 and GTX 1070, which were released on May 17, 2016 and June 10, 2016 respectively. Pascal was manufactured using TSMC's 16 nm FinFET process, and later Samsung's 14 nm FinFET process.

Digital signal processing (DSP) is a ubiquitous methodology in scientific and engineering computations. However, practically, DSP problems are often not only 1-D. For instance, image data are 2-D signals and radar signals are 3-D signals. While the number of dimension increases, the time and/or storage complexity of processing digital signals grow dramatically. Therefore, solving DSP problems in real-time is extremely difficult in reality.

In computer science, a 4D vector is a 4-component vector data type. Uses include homogeneous coordinates for 3-dimensional space in computer graphics, and red green blue alpha (RGBA) values for bitmap images with a color and alpha channel. They may also represent quaternions although the algebra they define is different.

A thread block is a programming abstraction that represents a group of threads that can be executed serially or in parallel. For better process and data mapping, threads are grouped into thread blocks. The number of threads varies with available shared memory. The number of threads in a thread block was formerly limited by the architecture to a total of 512 threads per block, but as of July 2019, with CUDA toolkit 10 and recent devices including Volta, blocks may contain up to 1024 threads. The threads in the same thread block run on the same stream processor. Threads in the same block can communicate with each other via shared memory, barrier synchronization or other synchronization primitives such as atomic operations.

Latency oriented processor architecture is the microarchitecture of a microprocessor designed to serve a serial computing thread with a low latency. This is typical of most Central Processing Units (CPU) being developed since the 1970s. These architectures, in general, aim to execute as many instructions as possible belonging to a single serial thread, in a given window of time; however, the time to execute a single instruction completely from fetch to retire stages may vary from a few cycles to even a few hundred cycles in some cases. Latency oriented processor architectures are the opposite of throughput-oriented processors which concern themselves more with the total throughput of the system, rather than the service latencies for all individual threads that they work on.

RDNA (microarchitecture) microarchitecture

RDNA is the codename for a GPU microarchitecture and accompanying instruction set developed by AMD. It is the successor to their Graphics Core Next (GCN) microarchitecture/instruction set. The first product lineup featuring RDNA was the Radeon RX 5000 series of video cards, launched on July 7, 2019. The architecture is also planned to be used in mobile products and the upcoming next generation of game consoles by Sony and Microsoft.


  1. Michael McCool; James Reinders; Arch Robison (2013). Structured Parallel Programming: Patterns for Efficient Computation. Elsevier. p. 52.
  2. "Nvidia Fermi Compute Architecture Whitepaper" (PDF). . NVIDIA Corporation. 2009. Retrieved 2014-07-17.External link in |website= (help)
  3. Lindholm, Erik; Nickolls, John; Oberman, Stuart; Montrym, John (2008). "NVIDIA Tesla: A Unified Graphics and Computing Architecture". IEEE Micro. 28 (2): 6 (Subscription required.). doi:10.1109/MM.2008.31.
  4. Rul, Sean; Vandierendonck, Hans; D’Haene, Joris; De Bosschere, Koen (2010). An experimental study on performance portability of OpenCL kernels. Symp. Application Accelerators in High Performance Computing (SAAHPC).
  5. "Advanced Topics in CUDA" (PDF). 2011. Retrieved 2014-08-28.
  6. Michael McCool; James Reinders; Arch Robison (2013). Structured Parallel Programming: Patterns for Efficient Computation. Elsevier. pp. 209 ff.
  7. John L. Hennessy; David A. Patterson. Computer Architecture: A Quantitative Approach (6 ed.). Morgan Kaufmann. pp. 314 ff.