NVIDIA A100 Tensor Core GPU: Performance and Innovation
The DOI matches its Crossref record; 3 of the 3 references it cites were found in a registry or an index.
Abstract
NVIDIA A100 Tensor Core GPU is NVIDIA's latest flagship GPU. It has been designed with many new innovative features to provide performance and capabilities for HPC, AI, and data analytics workloads. Feature enhancements include a Third-Generation Tensor Core, new asynchronous data movement and programming model, enhanced L2 cache, HBM2 DRAM, and third-generation NVIDIA NVLink I/O.
Overview
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
In the citing paper
“In addition to improved data locality, another advantage of batching is to turn the query from a vector into a matrix, allowing efficient matrix multiplications with tensor cores (Choquette et al., 2021).”
From ChunkAttention: Efficient Self-Attention with Prefix-Aware KV Cache and Two-Phase Partition (2024) · cites this paper as Choquette et al., 2021
In the citing paper
“All these formats discussed here offer several benefits over the traditional floating-point precision, enabling faster and more efficient inference with reduced memory bandwidth.”
From A Survey of Techniques for Optimizing Transformer Inference (2023) · cites this paper as [166]
In the citing paper
“The emerging numerical formats such as Bfloat16 [164], Microsoft Floating Point (MSFP) [165], Tensorfloat32 [166], and FP8 [167, 168] provide better computational efficiency and accuracy for several transformer workloads.”
From A Survey of Techniques for Optimizing Transformer Inference (2023) · cites this paper as [166]
In the citing paper
“arsitektur Ampere mampu menggandakan kinerja komputasi mentah dibandingkan generasi sebelumnya melalui peningkatan CUDA Cores [4]”
From Analisis Komparatif Kinerja Komputasi Brute Force pada Arsitektur GPU NVIDIA Turing, Ampere, dan Ada Lovelace (2026) · cites this paper as [4]
In the citing paper
“In this workload, these stable, highly reused chunks (119 GB for LLaMA-3-70B) fit comfortably within the 135 GB free GPU memory budget (corresponding to LLaMa-3-70B hosted on 4 A100-80GB GPUs [13], with Tensor-Parallelism).”
From Cache-Craft: Managing Chunk-Caches for Efficient Retrieval-Augmented Generation (2025) · cites this paper as [13]
Citations
Cites
Mattson, P., Cheng, C., Coleman, C., Diamos, G., Micikevicius, P., Patterson, D., Tang, H., Wei, G.-Y., Bailis, P., Bittorf, V., et al. (2019) · arXiv · doi:10.48550/arxiv.1910.01500 · Cited by 206
Volta: Performance and Programmability
Choquette, J., Giroux, O., & Foley, D. (2018) · IEEE Micro · doi:10.1109/mm.2018.022071134 · Cited by 126 · cited 7×
“The goal of CUDA is to compile intuitive C þþ sources into high-performance executable programs for the GPU.”
BackgroundInfluentialSelf-citationPhasers: a unified deadlock-free construct for collective and point-to-point synchronization
Shirako, J., Peixotto, D. M., Sarkar, V., & Scherer, W. N. (2008) · Proceedings of the 22nd annual international conference on Supercomputing · doi:10.1145/1375527.1375568 · Cited by 124
Research works printed but not linked: 1
- ishii, 2018, NVSwitch and DGX-2: NVLink-Switching chip and scale-up compute server, Hot Chips
Cited by
F1: A Fast and Programmable Accelerator for Fully Homomorphic Encryption
Samardzic, N., Feldmann, A., Krastev, A., Devadas, S., Dreslinski, R., Peikert, C., & Sanchez, D. (2021) · MICRO-54: 54th Annual IEEE/ACM International Symposium on Microarchitecture · doi:10.1145/3466752.3480070 · Cited by 294
“We assume 512GB/s bandwidth per PHY; this is similar to the NVIDIA A100 GPU [18], which has 2.”
MethodInfluentialA Survey on Model Compression for Large Language Models
Zhu, X., Li, J., Liu, Y., Ma, C., & Wang, W. (2024) · Transactions of the Association for Computational Linguistics · doi:10.1162/tacl_a_00704 · Cited by 273
CraterLake: a hardware accelerator for efficient unbounded computation on encrypted data
Samardzic, N., Feldmann, A., Krastev, A., Manohar, N., Genise, N., Devadas, S., Eldefrawy, K., Peikert, C., & Sanchez, D. (2022) · Proceedings of the 49th Annual International Symposium on Computer Architecture · doi:10.1145/3470496.3527393 · Cited by 207
“GPUs are inefficient on FHE: Prior work has studied the use of GPUs to accelerate FHE [3, 42, 65ś67].”
MethodInfluentialMixed precision algorithms in numerical linear algebra
Higham, N. J., & Mary, T. (2022) · Acta Numerica · doi:10.1017/s0962492922000022 · Cited by 278
“More recently, Oo and Vogel (2020) also used fp16 arithmetic on V100 GPUs.”
MethodInfluentialNVIDIA Hopper H100 GPU: Scaling Performance
Choquette, J. (2023) · IEEE Micro · doi:10.1109/mm.2023.3256796 · Cited by 192
“H100 features fi ve HBM3 sites with a total memory capacity of 80 GB and over 3 TB/s of memory bandwidth, twice the throughput of our previous generation A100 GPU.”
BackgroundInfluentialSelf-citation
Metadata
- Authors
- Choquette, J., Gandhi, W., Giroux, O., Stam, N., & Krashinsky, R.
- Affiliations
- NVIDIA, Singapore
- Venue
- IEEE Micro
- Published
- 2021-03-01
- Volume
- 41
- Issue
- 2
- Pages
- 29-35
- ISSN
- 0272-1732 (ISSN-L)
- 1937-4143
- OpenAlex
- W3130554079
- Cited by
- 410
- References
- 3
- Citations by year
- 2026: 68 · 2025: 104 · 2024: 102 · 2023: 65 · 2022: 45 · 2021: 24
- FWCI
- 38.70
- Type
- journal article
- Language
- en
- Open access
- closed
- Licence
- Locations
- IEEE Micro (publishedVersion)
- Topics
- Physical Sciences › Computer Science › Hardware and Architecture › Parallel Computing and Optimization Techniques
- Physical Sciences › Computer Science › Computer Graphics and Computer-Aided Design › Computer Graphics and Visualization Techniques
- Physical Sciences › Computer Science › Computer Networks and Communications › Interconnection Networks and Systems
- Keywords
- NVIDIA A100, tensor cores, GPU architecture, HPC workloads, AI workloads, data analytics, L2 cache, NVLink, GPU interconnects
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
How to cite
Cite CoChat for the checks on this page, and the paper for its own findings.
- This page
- CoChat. (2026). Citation check: NVIDIA A100 Tensor Core GPU: Performance and Innovation. https://app.cochat.ai/papers/10.1109/mm.2021.3061394
- The paper
- Choquette, J., Gandhi, W., Giroux, O., Stam, N., & Krashinsky, R. (2021). NVIDIA A100 Tensor Core GPU: Performance and Innovation. IEEE Micro. https://doi.org/10.1109/mm.2021.3061394
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.