GPU

GPU
ScotXW · CC0 · source
Name	Graphics Processing Unit
Developer	Nvidia, AMD, Intel
Type	Processor
Introduced	1999
Architecture	SIMD, MIMD
Application	Graphics, compute

GPU A graphics processing unit is a specialized electronic processor designed to accelerate image rendering, parallel computation, and data-parallel workloads for systems ranging from desktop Apple workstations to cloud servers operated by Amazon Web Services and Google. GPUs are central to products and platforms created by Nvidia, AMD, and Intel, and are used across industries including entertainment exemplified by Pixar, scientific research at CERN, and finance companies such as Goldman Sachs. They coexist with central processors from firms like Intel and AMD inside systems produced by Dell, HP, and Lenovo.

Overview

GPUs implement highly parallel pipelines derived from graphics hardware used in systems such as Sony PlayStation, Microsoft Xbox, and personal computers made by Acer. Early consumer adoption followed products from 3Dfx Interactive and ATI Technologies; modern devices integrate GPU dies into discrete cards by EVGA Corporation and integrated chipsets by Samsung Electronics. Key market trends have been shaped by compute workloads in data centers run by Microsoft and research clusters at NASA. Industry events like the Consumer Electronics Show and awards such as the Turing Award highlight advances and applications.

Architecture

GPU architecture combines shader cores, rasterizers, texture units, and memory controllers into tiled pipelines used in APIs developed by Khronos Group, with designs influenced by microarchitecture research at Massachusetts Institute of Technology and University of California, Berkeley. Architectures from Nvidia (e.g., architectures used in Tesla and GeForce families) and AMD (e.g., architectures behind Radeon) adopt thousands of parallel execution units organized into compute clusters and employ high-bandwidth memory interfaces like those championed by Micron Technology. Interconnects such as PCI Express and technologies like NVLink enable multi-GPU scaling found in supercomputers like Summit. Power and thermal designs are topics of collaboration with firms such as Cooler Master and Noctua.

Programming and APIs

GPU programming relies on APIs and frameworks including OpenGL, Vulkan, and platform-specific models such as CUDA from Nvidia and ROCm from AMD. Language bindings and compilers developed at institutions like Stanford University and companies like LLVM enable integration with languages such as C++ and Python for machine learning libraries like TensorFlow and PyTorch. Standards bodies such as the Khronos Group coordinate cross-vendor APIs while research on parallel algorithms appears in conferences like International Conference on Machine Learning and Conference on Neural Information Processing Systems.

Applications

GPUs drive visual computing in studios such as Industrial Light & Magic and real-time rendering in engines like Unreal Engine and Unity. In scientific computing, GPUs accelerate simulations used at Los Alamos National Laboratory and climate modeling at NOAA. In healthcare, image analysis projects at Mayo Clinic and genomics research at Broad Institute use GPU-accelerated pipelines. Financial institutions including JPMorgan Chase and Citigroup apply GPUs to risk modeling; autonomous vehicle projects by Waymo and Tesla rely on GPU inference for perception stacks.

Performance and Benchmarking

Benchmark suites from organizations such as SPEC and tools like 3DMark and Geekbench measure throughput, memory bandwidth, and energy efficiency; results influence procurement by hyperscalers including Google and Meta Platforms. Metric comparisons appear in peer-reviewed venues like IEEE journals and at symposiums including the International Symposium on Computer Architecture. Manufacturers tune drivers and firmware alongside partners such as ASUS and MSI to optimize scores on workloads used by studios like Weta Digital and research centers such as Argonne National Laboratory.

Historical Development

The evolution traces from fixed-function graphics accelerators by S3 Graphics and Matrox through programmable shading introduced in the early 2000s by ATI Technologies and Nvidia, to modern general-purpose compute enabled by initiatives from Stanford University researchers and commercial platforms like CUDA. Milestones include the consolidation of graphics APIs at Khronos Group and the expansion into machine learning workloads driven by breakthroughs in deep learning published by groups at University of Toronto and labs at Google Brain. The rise of GPU-accelerated supercomputers such as Summit and industry shifts toward heterogeneous computing have influenced roadmaps at companies including Intel and AMD.

Category:Computer hardware