Ampere (microarchitecture)

Ampere (microarchitecture)
Name	Ampere
Designer	Nvidia
Launched	2020
Predecessor	Turing (microarchitecture)
Successor	Ada Lovelace (microarchitecture)
Process	Samsung 8 nm, TSMC 7 nm, TSMC 5 nm
Products	GeForce 30 series, Nvidia RTX A-series, Nvidia A100, Nvidia H100

Ampere (microarchitecture). Ampere is a GPU microarchitecture designed by Nvidia and officially announced in May 2020. It succeeded the Turing (microarchitecture) and is named after the French physicist André-Marie Ampère. The architecture brought significant improvements in ray tracing performance, artificial intelligence compute capabilities, and overall efficiency across consumer, professional, and data center markets.

Overview

The introduction of Ampere marked a major generational leap for Nvidia, targeting a broad range of computing segments from gaming to high-performance computing. Key technological focuses included enhanced CUDA cores, new RT Cores for accelerated ray tracing, and third-generation Tensor Cores for AI workloads. The architecture was implemented across multiple manufacturing processes from partners like Samsung (company) and TSMC, powering products such as the GeForce 30 series and the Nvidia A100 accelerator. This unified architectural approach allowed Nvidia to address markets including scientific research, cloud computing, and creative professionals.

Architecture

The Ampere architecture introduced several core innovations over its predecessor. It featured new streaming multiprocessors with dual FP32 units, significantly increasing traditional shader performance. The second-generation RT Cores delivered up to twice the ray-triangle intersection throughput of Turing (microarchitecture) cores, accelerating real-time rendering in titles like Cyberpunk 2077. Third-generation Tensor Cores supported new AI data formats including TF32 and FP64, crucial for workloads run on systems like the DGX A100. Memory subsystems were upgraded with GDDR6X technology developed with Micron Technology and support for HBM2e in data center products, improving bandwidth for applications in computational fluid dynamics and molecular dynamics.

Products

Ampere-based products spanned multiple Nvidia product lines. The consumer GeForce 30 series, led by the GeForce RTX 3090, launched in September 2020. For professional visualization, the Nvidia RTX A-series (such as the RTX A6000) replaced the previous Quadro brand. In the data center, the Nvidia A100 PCI Express and SXM modules became the foundation for servers from Dell Technologies, Hewlett Packard Enterprise, and Super Micro Computer. Later, the Nvidia H100, built on a refined TSMC 5 nm process, succeeded the A100 for advanced AI training and HPC workloads. The architecture also powered custom systems like Nintendo's Switch OLED model and Nvidia's own DGX Station A100.

Performance

Performance gains were substantial across benchmarks and applications. In gaming, the GeForce RTX 3080 often doubled the performance of the Turing (microarchitecture)-based GeForce RTX 2080 in traditional rasterization, while ray tracing performance saw even greater improvements. For AI and HPC, the Nvidia A100 demonstrated massive leaps in training models like BERT and GPT-3 compared to the previous Volta (microarchitecture)-based Tesla V100. The architecture's efficiency also improved, with Nvidia citing a near-doubling of performance per watt. These advancements were showcased in record-setting runs on benchmarks like MLPerf and in real-world deployments for COVID-19 research and autonomous vehicle development.

Software and ecosystem

Ampere's capabilities were unlocked through a mature software stack. Key enabling technologies included Nvidia DLSS, which used the Tensor Cores for AI-powered super sampling, and CUDA 11.0 with enhanced libraries for HPC. The architecture was fully supported by Nvidia OptiX for ray tracing, Nvidia Omniverse for collaborative design, and the Nvidia AI Enterprise suite. Major game engines like Unreal Engine and Unity (game engine) integrated support for its features, while cloud platforms such as Amazon Web Services, Microsoft Azure, and Google Cloud Platform offered instances powered by the A100 and H100.

Reception and impact

The Ampere architecture was widely praised by reviewers and industry analysts for its significant performance uplift. Publications like AnandTech, Tom's Hardware, and PC Gamer highlighted its dominance in the consumer GPU market upon release. Its impact extended beyond gaming, as the A100 became a standard accelerator in supercomputing centers, including the Cambridge-1 system in the United Kingdom and installations at the National Energy Research Scientific Computing Center. The architecture solidified Nvidia's leadership in AI and professional graphics, though it faced challenges during the global chip shortage and increased competition from AMD's RDNA 2 architecture. Its successor, Ada Lovelace (microarchitecture), built upon its foundations for the next generation of GeForce 40 series products.

Category:Nvidia microarchitectures Category:Graphics processing units Category:2020 in computing