Ampere
An NVIDIA GPU microarchitecture, such as the A100 and the RTX 30 series. It is the generation of architecture preceding Hopper.
Core Facts
Timeline (last 90 days)
The GA102 chip of the Ampere architecture has a large-capacity L2 cache, but the memory bandwidth of consumer-grade graphics cards is usually lower than that of professional cards.
Each SM in the Ampere architecture contains four sub-partitions, and each sub-partition has 32 FP32 cores.
Nvidia 3000 series graphics cards are based on the Ampere architecture.
Protein structure generation models often rely on custom CUDA extensions (such as efficient geometric operations in equivariant networks), and such extensions have requirements on compiler versions and GPU architectures (e.g., Ampere, Hopper).
Compute Capability is a version number NVIDIA defines for its GPU architectures, expressed as "major.minor" (e.g., 8.x for Ampere, 8.9 for Ada Lovelace, 12.0 for Blackwell), encoded directly in the GPU hardware identity and read by the driver via PCI configuration space and GPU registers; it does not depend on upper-layer software or platform labels and cannot be forged by editing product descriptions or the virtualization layer (MIG partitioning only affects the visibility of memory and compute units, not the architecture version number).
NVFP4 is a Blackwell tensor core format, and the FP4-Local model card explicitly states that Blackwell is the only supported microarchitecture, with Hopper and Ampere not supported.
Compared with the previous generation Ampere's mma instruction, WGMMA supports larger matrix tiles and higher throughput, and can overlap in a pipeline with asynchronous copy mechanisms such as TMA (Tensor Memory Accelerator).
The Blackwell architecture only went into mass production in 2025, and many mainstream tutorials and pre-compiled packages are still built for the Ampere or Ada Lovelace architectures.
The A100 uses the Ampere architecture, and the H100 uses the Hopper architecture.
Different GPU architectures (such as Ampere and Hopper) have different on-chip memory specifications, and tiling parameters often need to be re-tuned for specific hardware.
11 more timeline events
All Facts (20)
The Switch 2 uses a custom chip based on the NVIDIA Ampere architecture, supporting DLSS technology and ray tracing.
65%UnverifiedTensorRT的内核自动调优机制会针对目标GPU的具体架构（如Ampere、Hopper等）对候选内核进行性能基准测试
90%Unverified同一组编译选项在 Ampere、Hopper、Blackwell 等不同 GPU 架构上的表现可能截然不同
90%UnverifiedAmpere is a GPU architecture released by NVIDIA in 2020, and all RTX 30-series consumer graphics cards are based on this architecture.
60%UnverifiedNVIDIA GPUs introduced INT8 tensor cores starting with the Turing architecture, and the Ampere architecture further enhanced integer computation capabilities.
60%UnverifiedThe GA102 chip of the Ampere architecture has a large-capacity L2 cache, but the memory bandwidth of consumer-grade graphics cards is usually lower than that of professional cards.
50%UnverifiedNvidia 3000 series graphics cards are based on the Ampere architecture.
50%UnverifiedEach SM in the Ampere architecture contains four sub-partitions, and each sub-partition has 32 FP32 cores.
50%UnverifiedProtein structure generation models often rely on custom CUDA extensions (such as efficient geometric operations in equivariant networks), and such extensions have requirements on compiler versions and GPU architectures (e.g., Ampere, Hopper).
50%UnverifiedCompute Capability is a version number NVIDIA defines for its GPU architectures, expressed as "major.minor" (e.g., 8.x for Ampere, 8.9 for Ada Lovelace, 12.0 for Blackwell), encoded directly in the GPU hardware identity and read by the driver via PCI configuration space and GPU registers; it does not depend on upper-layer software or platform labels and cannot be forged by editing product descriptions or the virtualization layer (MIG partitioning only affects the visibility of memory and compute units, not the architecture version number).
50%UnverifiedNVFP4 is a Blackwell tensor core format, and the FP4-Local model card explicitly states that Blackwell is the only supported microarchitecture, with Hopper and Ampere not supported.
50%UnverifiedCompared with the previous generation Ampere's mma instruction, WGMMA supports larger matrix tiles and higher throughput, and can overlap in a pipeline with asynchronous copy mechanisms such as TMA (Tensor Memory Accelerator).
50%UnverifiedThe Blackwell architecture only went into mass production in 2025, and many mainstream tutorials and pre-compiled packages are still built for the Ampere or Ada Lovelace architectures.
50%UnverifiedThe A100 uses the Ampere architecture, and the H100 uses the Hopper architecture.
50%UnverifiedDifferent GPU architectures (such as Ampere and Hopper) have different on-chip memory specifications, and tiling parameters often need to be re-tuned for specific hardware.
50%UnverifiedThe project rendered a scene containing 5 million Gaussian splat points at 60 fps on an Ampere-architecture A6000 GPU.
50%UnverifiedNVIDIA CUDA's GPU architecture has evolved from Volta and Ampere to Hopper and Blackwell.
50%UnverifiedCUDA provides developers with a common computing platform that can be continuously upgraded and optimized throughout the entire lifecycle of three generations of architectures: Ampere, Hopper, and Blackwell.
50%UnverifiedThe original Switch uses a Tegra X1 based on the Maxwell architecture (2014), while the Switch 2 upgrades directly to Ampere (2020).
50%UnverifiedThe Ampere architecture was introduced by NVIDIA in 2020 and was first applied to the RTX 30 series desktop graphics cards.
50%