COMPUTE ATLAS Supercomputer supply-chain graph
1345 systems 511 sites Sourced data

Parts/Accelerators/NVIDIA A100 SXM4 40GB

Accelerators · Ampere · 2020

NVIDIA A100 SXM4 40GB

The launch A100 configuration, before the 64 and 80 GB variants.

Why this part

The part that defined the four-GPU node. Ampere added multi-instance GPU, which partitions one physical device into as many as seven isolated instances, and that feature exists for cloud tenancy rather than for simulation: it lets an operator rent fractions of a device. A100 appears under Perlmutter at a national laboratory and under Selene at NVIDIA itself, running the same silicon for different businesses, which is roughly when the two markets stopped being distinguishable.

Systems using this part

Systems
43
Rmax underneath
459.0 PFlop/s
Units recorded
7,805
24 of 43 edges disclose a quantity
System Role Country Units System Rmax Supplied by Rank
HEMUS
2023-10
Accelerator Bulgaria - 2.53 PFlop/s NVIDIA #359 →
Derecho
2023-07
Accelerator USA 328 10.32 PFlop/s NVIDIA #59 →
ASPIRE 2A (GPU Partition)
2023-07
Accelerator Singapore 352 3.33 PFlop/s NVIDIA #167 →
Supek
2023
Accelerator Croatia - - NVIDIA -
Kempner AI Cluster
2023
Accelerator USA 144 16.29 PFlop/s NVIDIA #85 →
Komondor
2022-12
Accelerator Hungary - 3.10 PFlop/s NVIDIA -
Lucia
2022-11
Accelerator Belgium 200 2.72 PFlop/s NVIDIA #245 →
Delta
2022-11
Accelerator USA 106 3.81 PFlop/s NVIDIA #141 →
VSC-5
2022
Accelerator Austria 120 2.31 PFlop/s NVIDIA #301 →
LANTA
2022
Accelerator Thailand 704 8.15 PFlop/s NVIDIA #70 →
Discovery 5
2022
Accelerator USA - 26.13 PFlop/s NVIDIA #16 →
Narval
2021-12
Accelerator Canada 632 5.89 PFlop/s NVIDIA #83 →
Baskerville
2021-09
Accelerator United Kingdom 184 2.34 PFlop/s NVIDIA #445 →
Karolina
2021-08
Accelerator Czechia - 6.75 PFlop/s NVIDIA #69 →
Raven-GPU
2021-06
Accelerator Germany 768 8.62 PFlop/s NVIDIA #47 →
Perlmutter
2021-05
Accelerator USA - 79.23 PFlop/s NVIDIA #28 →
SQUID
2021-05
Accelerator Japan 336 6.11 PFlop/s NVIDIA #159 →
Vega
2021-04
Accelerator Slovenia - 6.92 PFlop/s NVIDIA #106 →
Earth Simulator 4
2021-03
Accelerator Japan 64 9.99 PFlop/s NVIDIA #39 →
Polaris
2021
Accelerator USA - 25.81 PFlop/s NVIDIA #83 →
Athena
2021
Accelerator Poland 384 5.05 PFlop/s NVIDIA #105 →
Berzelius
2021
Accelerator Sweden 480 5.25 PFlop/s NVIDIA #82 →
Tursa
2021
Accelerator United Kingdom - 5.23 PFlop/s NVIDIA #95 →
AlphaCentauri
2021
Accelerator Germany - 2.59 PFlop/s NVIDIA #181 →
mdx I
2021
Accelerator Japan 320 - NVIDIA -
Hortense
2021
Accelerator Belgium 80 - NVIDIA -
Pioneer-SCUS
2021
Accelerator USA - 16.59 PFlop/s NVIDIA #27 →
Pioneer-WEU
2021
Accelerator Netherlands - 16.59 PFlop/s NVIDIA #28 →
Pioneer-WUS2
2021
Accelerator USA - 16.59 PFlop/s NVIDIA #35 →
JUWELS Booster
2020-11
Accelerator Germany - 44.12 PFlop/s NVIDIA #31 →
Betzy
2020-11
Accelerator Norway 16 4.72 PFlop/s NVIDIA #55 →
PARAM Siddhi-AI
2020-11
Accelerator India 336 4.62 PFlop/s NVIDIA #62 →
Selene
2020-06
Accelerator USA - 63.46 PFlop/s NVIDIA #36 →
JURECA-DC
2020
Accelerator Germany 768 9.33 PFlop/s NVIDIA #45 →
Lyapunov
2020
Accelerator Russia 1,096 12.81 PFlop/s NVIDIA #40 →
Pioneer-EUS
2020
Accelerator USA - 16.59 PFlop/s NVIDIA #26 →
Euler
2014
Accelerator Switzerland 3 - NVIDIA -
Jetstream2
Accelerator USA 360 - NVIDIA -
Agate
Accelerator USA - 1.88 PFlop/s NVIDIA #379 →
Dream-AI
Accelerator South Korea - 3.18 PFlop/s NVIDIA #178 →
CSF3
Accelerator United Kingdom 8 - NVIDIA -
Vera (Chalmers)
Accelerator Sweden - - NVIDIA -
Niflheim
Accelerator Denmark 16 - NVIDIA -

“Units recorded” sums only the edges where a public source states a quantity. It is a floor, not a total, and should never be read as installed base.

Further reading