Supercomputing in Taiwan

From Wikipedia, the free encyclopedia

Supercomputing in Taiwan has a history going back to the 1990s, with the establishment of the National Center for High-Performance Computing in 1993 by the Government of Taiwan.[1] In November 2018 the National Center for High-Performance Computing owned supercomputer Taiwania 2 debuted at number 20 on the TOP500 list of fastest supercomputers.[2] As of November 2025, the NCHC NANO 4 supercomputer is the fastest supercomputer in Taiwan, having been ranked 29th fastest in the world in the TOP500 supercomputer list.[3]

Pie Chart of Share of Supercomputers among countries from Top500 Supercomputers list as in November 2025, showing Taiwan as ranked joint eighth place with 10 supercomputers in the list.

The sector is primarily supported by government initiatives, academic institutions, and the technology industry, and plays a role in scientific research, industrial innovation, and the development of artificial intelligence (AI).

Taiwan's supercomputing capabilities are closely linked to its strengths in the semiconductor industry and information and communications technology (ICT), and have expanded significantly since the 2010s with the growth of data-intensive computing and AI applications.[4]

History

Supercomputer Taiwania 3 at NCHC in Taichung.

Early development of high-performance computing in Taiwan was driven by academic and scientific research needs, particularly in fields such as physics, meteorology, and engineering. In 1993, the Taiwanese government established the National Center for High-Performance Computing (NCHC). It is one of ten national-level research laboratories under National Institutes of Applied Research (NIAR), headquartered at Hsinchu Science Park, Hsinchu, Taiwan. It operates several of the nation's major supercomputers and provides computing resources to academia, government agencies, and industry. NCHC systems are integrated with cloud platforms such as the Taiwan Computing Cloud (TWCC), which provides distributed computing services and supports AI development, simulation, and data analysis.

In the early 2010s under the Ma Ying-jeou administration, according to the Institute of Electrical and Electronics Engineers, ineffective government budget allocation has slowed down the process of upgrading Taiwan's supercomputing infrastructure, causing Taiwan to have zero supercomputers in the TOP500 list in November 2015.[5]

In November 2018, under the Tsai Ing-wen administration, Taiwan's supercomputing capability made a major comeback to the TOP500 list with the introduction of the Taiwania 2 supercomputer, ranking at No. 20. At the time, it was the highest ranking ever achieved by a Taiwan-made supercomputer.[2]

In the 2020s, under the Tsai Ing-wen and Lai Ching-te administrations, government policies were strategically focused on transforming Taiwan into an "AI Island" and accelerating digital transformation by raising national computing power. The government is investing heavily in this sector, enforcing strict energy efficiency standards for data centers, and fostering sovereign research to build a complete AI ecosystem.[6]

Major systems

TAIWANIA series

Image showing the three Taiwania supercomputers.

The Taiwania series of supercomputers represents a major component of Taiwan's HPC infrastructure.

  • Taiwania 2 is a GPU-accelerated system designed for AI and deep learning applications. It consists of over 250 compute nodes with CPUs and GPUs, providing performance of up to approximately 9 petaflops (PFLOPS). The system emphasizes energy efficiency and supports a wide range of AI frameworks, including TensorFlow and PyTorch. In 2018, it was ranked number 20 in the TOP500 supercomputer list.[2]
  • Taiwania 3 is a CPU-based system focused on general-purpose computing. It features approximately 50,000 computing cores and is used in scientific research fields such as physics, chemistry, and atmospheric science.[7][8][9][10][11][12][13][14]

These systems are accessible to researchers and industry users through application-based access, supporting both academic research and industrial innovation.[15]

Forerunner 1

Forerunner 1 is a general-purpose HPC system comprising over 500 compute servers and tens of thousands of processing cores. It supports parallel computing applications across multiple scientific disciplines and includes both x86 and ARM-based architectures for heterogeneous computing research.[16][17][18]

Nano series

The Nano series represents newer generations of AI-oriented supercomputers developed by NCHC.[19]

  • Nano 5 (also known as 晶創25) is an accelerator-based clustered system incorporating NVIDIA H100 and H200 GPUs. It achieved a performance of approximately 13 petaflops and debuted in the TOP500 supercomputer rankings in 2025. The system is designed to support AI model training and scientific simulations.
  • NANO 4 (also known as 晶創26) is a large-scale AI supercomputer with a measured performance of approximately 81.55 petaflops. It incorporates NVIDIA H200-based architecture and GB200 NVL72 systems, combining general-purpose HPC with AI training capabilities. The system includes high-speed interconnects and large-scale storage infrastructure, and is used in research areas such as AI, telecommunications, and digital content.[20]

These systems form part of national infrastructure programs aimed at strengthening computing capacity and supporting technological development.

Rankings

Current TOP500

As of November 2025, there are 10 systems in Taiwan in the TOP500 supercomputer list.[3][21]

More information Rank, Name ...
RankNameRmax
(PFlop/s)		Rpeak
(PFlop/s)
29NANO 481.55117.92
46Ubilink45.8269.34
80Taipei-122.3034.53
135Nano 513.0619.01
153PRIMEHPC11.1612.98
158Honhai Super AI Computing Center 210.4612.03
183Taiwania 29.0015.21
242Honhai Super AI Computing Center 16.128.58
254PRIMEHPC FX10005.606.49
359Forerunner 13.534.00
Close

Taiwan's historical rank in TOP500

More information List, Number of systems in TOP500 ...
Rank of Taiwanese supercomputers in TOP500 list[21]
List Number of systems
in TOP500		 System Share (%) Total Rmax
(Gflops)		 Total Rpeak
(Gflops)		 Cores
2025 November 10 2 208,612,520 300,088,670 1,063,168
2025 June 8 1.6 116,600,520 170,134,420 753,216
2024 November 7 1.4 103,540,520 151,123,540 712,336
2024 June 6 1.2 53,895,080 77,560,100 600,688
2023 November 5 1 31,595,080 43,026,720 559,728
2023 June 2 0.4 11,297,520 19,562,790 220,752
2022 November 2 0.4 11,297,520 19,562,790 220,752
2022 June 2 0.4 11,297,520 19,562,790 220,752
2021 November 2 0.4 11,297,520 19,562,790 220,752
2021 June 2 0.4 11,297,520 19,562,790 220,752
2020 November 3 0.6 12,622,710 21,651,750 247,952
2020 June 2 0.4 10,325,150 17,297,190 197,552
2019 November 2 0.4 10,325,150 17,297,190 197,552
2019 June 2 0.4 10,325,150 17,297,190 197,552
2018 November 2 0.4 10,325,150 17,297,190 197,552
2018 June 1 0.2 1,325,150 2,088,960 27,200
2017 November 1 0.2 1,325,150 2,088,960 27,200
2017 June 0 0 0 0 0
2016 November 0 0 0 0 0
2016 June 0 0 0 0 0
2015 November 0 0 0 0 0
2015 June 1 0.2 177,100 231,859 26,244
2014 November 1 0.2 177,100 231,859 26,244
2014 June 1 0.2 177,100 231,859 26,244
2013 November 1 0.2 177,100 231,859 26,244
2013 June 1 0.2 177,100 231,859 26,244
2012 November 3 0.6 356,625 556,484 38,520
2012 June 3 0.6 340,955 526,138 37,212
2011 November 2 0.4 247,530 380,195 30,996
2011 June 2 0.4 220,504 313,570 32,148
2010 November 0 0 0 0 0
2010 June 0 0 0 0 0
2009 November 0 0 0 0 0
2009 June 1 0.2 19,910 24,960 2,080
2008 November 2 0.4 33,592 48,960 4,080
2008 June 3 0.6 45,677 69,792 5,816
2007 November 11 2.2 102,966 162,821 14,024
2007 June 10 2 66,680 97,861 8,868
2006 November 2 0.4 5,535 9,901 1,768
2006 June 3 0.6 7,562 12,205 2,152
2005 November 3 0.6 7,562 12,205 2,152
2005 June 2 0.4 4,089 6,784 1,184
2004 November 3 0.6 3,906 5,680 1,140
2004 June 3 0.6 2,381 3,779 684
2003 November 5 1 3,425 6,131 1,176
2003 June 2 0.4 1,155 2,003 448
2002 November 3 0.6 1,020 1,830 425
2002 June 3 0.6 993 1,823 449
2001 November 4 0.8 635 825 385
2001 June 4 0.8 533 729 375
2000 November 2 0.4 310 396 183
2000 June 2 0.4 310 396 183
1999 November 1 0.2 139 144 15
1999 June 0 0 0 0 0
1998 November 1 0.2 20 29 110
1998 June 1 0.2 20 29 110
1997 November 1 0.2 15 21 80
1997 June 1 0.2 15 21 80
1996 November 1 0.2 15 21 80
1996 June 2 0.4 18 17 100
1995 December 2 0.4 10 14 52
1995 June 2 0.4 7 9 34
1994 November 4 0.8 6 8 30
1994 June 2 0.4 3 4 22
1993 November 2 0.4 3 4 22
1993 June 3 0.6 2 3 14
Close

See also

References

Related Articles

Wikiwand AI