Supercomputers/Systems/Tianhe-2A
China · upgraded · hpc
Tianhe-2A
Also known as Tianhe-2, TH-2, Milky Way-2
Operated by National Supercomputer Center in Guangzhou at National Supercomputer Center in Guangzhou .
Researched with AI assistance, checked by scripts. These rows were researched by AI agents reading public operator, laboratory and vendor sources, appended by scripts and checked by automated validation; no subject-matter expert has reviewed them (how this site is made). A mechanical second-reader pass has since fetched every cited source: 1350 of 1581 systems have a readable citation that names them, and 64 are genuinely weakly sourced. Every claim carries its source and a confidence tier. Treat anything below verified as a lead, not a citation.
Measured performance
- Rmax
- 61.44 PFlop/s
- Rpeak
- 100.7 PFlop/s
- Rmax ÷ Rpeak
- 61.0%
- Cores
- 4,981,760
- Power
- 18.48 MW
- Per watt
- 3.3 GF/W
Figures are as last publicly reported for the configuration described below, not a live measurement. Where a system was upgraded in place, the post-upgrade configuration is shown and the earlier one appears in the timeline.
How it compares
Ranked against the other machines in this dataset that report the same figure: those that first came into service within two years of Tianhe-2A, and those in China. Computed from our tables; a metric is ranked only when at least eight peers report it.
| Metric | Value | Against its era | Against its country |
|---|---|---|---|
| Rmax | 61.44 PFlop/s | #1 of 227 · highest | #3 of 27 · top 8% |
| Energy efficiency | 3.3 GF/W | #4 of 186 · top 2% | #7 of 19 · top 33% |
| Rmax ÷ Rpeak | 61.0 % | #202 of 226 · bottom 12% | #17 of 27 · bottom 39% |
| Power draw | 18,482 kW | #1 of 188 · highest | #2 of 19 · top 6% |
Every metric is ranked largest-first, so a high power-draw rank means a big power draw, not a good one. Rmax ÷ Rpeak uses HPL-measured results only. Rankings move as machines are added; they describe this dataset, not the world.
What this machine is made of
One row per supplier relationship. “Supplier at build” is the company that shipped the part at the time; where that company has since been acquired, the parent it rolls up to today is shown beside it. That distinction is what makes ticker-level aggregation possible across a thirty-year dataset.
| Role | Part | Supplier at build | Quantity | Confidence | Source |
|---|---|---|---|---|---|
| Integrator | - NUDT | NUDT | - | Reported | nscc-gz.cnfederalregister.gov |
| Accelerator | Intel Xeon Phi 31S1P Accelerators x86-64 (Knights Corner) · 57C · 1.1 GHz · 22 nm · 270 W Xeon Phi 31S1P (2013-2017 configuration, superseded) | Intel | - | Reported | nscc-gz.cnfederalregister.gov |
| Accelerator | NUDT Matrix-2000 Accelerators NUDT custom · 128C · 1.2 GHz · 28 nm · 240 W Matrix-2000 (2018 Tianhe-2A configuration) | NUDT | - | Reported | nscc-gz.cnfederalregister.gov |
| CPU | Intel Xeon E5-2692 v2 CPUs x86-64 · 12C · 2.2 GHz · 22 nm · 115 W Intel Xeon E5-2692v2 12C 2.2GHz | Intel | - | Reported | nscc-gz.cnfederalregister.gov |
| Interconnect | NUDT TH Express-2 Interconnect Fat tree · 160 Gb/s per link · About 1.6 microseconds TH Express-2 | NUDT | - | Reported | nscc-gz.cnfederalregister.gov |
| Storage | H2FS Storage Hybrid virtual namespace over local + shared tiers, Lustre backend global shared parallel storage system containing 12.4 PB and uses the H2FS hybrid hierarchy file system | NUDT | - | Verified | netlib.orgjcst.ict.ac.cn |
| Operating system | Kylin Linux Operating systems Linux distribution · State-endorsed Chinese distribution Kylin Linux | KylinSoft | - | Reported | nscc-gz.cnfederalregister.gov |
Engineering, funding and access
What Tianhe-2A runs on, who paid for it and how people get time on it. Each line carries who said it and when.
Hardware
| Item | Detail | As of | Basis | Source |
|---|---|---|---|---|
| Compute nodes | 17,792 nodesUp from 16000 in the original Tianhe-2; each node has two Xeon CPUs and two Matrix-2000 accelerators | 2017-09 | Press report | top500.org |
| Total memory | 3,400 TB3.4 PB main memory, up from 1.4 PB; compute nodes have 192 GB each per The Next Platform | 2017-09 | Press report | top500.org |
| Storage capacity | 19 PBParallel file system at 1 TB/s, up from 12.4 PB at 512 GB/s; The Next Platform presumes Lustre | 2017-09 | Press report | nextplatform.com |
| Interconnect and topology | TH Express-2+ (Galaxy) with 14 Gb/s signalling, 112 Gb/s ports, 1 us port-hop latency and 24-port switchesTOP500 lists TH Express-2 | 2017-09 | Press report | nextplatform.comtop500.org |
Software
| Item | Detail | As of | Basis | Source |
|---|---|---|---|---|
| Programming and software stack | Kylin Linux with Intel icc, Intel MKL and MPICH2 with a customized GLEX channel | 2018 | Press report | top500.org |
Funding
| Item | Detail | As of | Basis | Source |
|---|---|---|---|---|
| Procurement | NUDT replaced the Xeon Phi accelerators with its own Matrix-2000 (128 cores, 2.4576 TFlop/s peak) after a US embargo blocked the planned Knights Landing upgrade | 2017-09 | Press report | top500.orgnextplatform.com |
The read
The clearest natural experiment in export control that this dataset contains. Tianhe-2 shipped in 2013 with Intel Xeon hosts and Intel Xeon Phi accelerators. In 2015 the US added NUDT and the operating centres to the Entity List, blocking the planned Xeon Phi upgrade. The 2018 Tianhe-2A rebuild replaced the American accelerators with NUDT's own Matrix-2000, and the machine got faster. Figures shown are the post-upgrade configuration; the component graph carries both eras.
Timeline
- 2010 Announced source
- 2013-06 List debut at #1 (June 2013) source
- 2013-06 First operational source
- 2015-02 US Entity List addition blocks the planned Intel Xeon Phi upgrade source
- 2017-09 Tianhe-2A upgrade unveiled at the International HPC Forum: Xeon Phi replaced by NUDT Matrix-2000, peak raised from 54.9 to 94.97 PFlop/s source
- 2018-06 Rebuilt as Tianhe-2A with domestic NUDT Matrix-2000 accelerators source
Public list appearances
Pointers only: rank and edition, linking to the canonical entry. We do not reproduce list tables. See sourcing policy.
Site & facility
National Supercomputer Center in Guangzhou
- Location
- Guangzhou, China
- Cooling
- Air and liquid hybrid
- Commissioned
- 2013
Change history
- 2026-09-22 Gained a Green500 rank pointer (#191 on the 2025-06 list)
Nearby in scale
The machines closest to Tianhe-2A by Rmax, smaller to larger.
- SuperPOD · United Arab Emirates, operational · 55.81 PFlop/s
- AI-03 · United Arab Emirates, operational · 54.12 PFlop/s
- Explorer-WUS3 · USA, operational · 53.96 PFlop/s
- Selene · USA, operational · 63.46 PFlop/s
- ASPIRE 2B (GPU Partition) · Singapore, operational · 66.4 PFlop/s
- Gefion · Denmark, operational · 66.59 PFlop/s
Related systems
Same site
- Tianhe-1A Guangzhou Solution · China, operational, 2012
Source check
We fetched this system's own citations and recorded whether each page actually mentions it. This is a corroboration signal, not a fact check, and it is published so you can see how well the sourcing holds up rather than take it on trust.
| Cited source | Result | Found on the page |
|---|---|---|
| federalregister.gov | names system and part | NUDT Matrix-2000, NUDT TH Express-2 |
| nscc-gz.cn | could not be fetched | - |
Checked 2026-10-18 by pnpm verify. Re-run it and the table changes with the web.
Related news
- NUDT and the Chinese national supercomputing centres added to the Entity List · Federal RegisterBlocked the planned Xeon Phi upgrade to Tianhe-2. The 2018 rebuild replaced the American accelerators with NUDT's own Matrix-2000 and the machine got faster.
Further reading
Inside a node
Why accelerators exist, what the memory hierarchy costs, and how the CPU and accelerator merged onto one package.
Operating a supercomputer
Acceptance testing, why "installed" and "in production" are six months apart, and the machine lifecycle.
The interconnect
Topologies, why latency and tail behaviour matter more than bandwidth, and how the fabric market consolidated.