PARAM Siddhi-AI

PARAM Siddhi-AI is a supercomputer for artificial intelligence and scientific computing. It belongs to the National Supercomputing Mission.

1. Who built it, and under which mission?

The National Param Supercomputing Facility (NPSF) at C-DAC in Pune commissioned the machine. C-DAC means the Centre for Development of Advanced Computing. The work is part of the National Supercomputing Mission (NSM).

The mission is steered jointly by the Ministry of Electronics and IT and the Department of Science and Technology. Atos worked with C-DAC on the system. The design follows the NVIDIA DGX SuperPOD reference plan.

2. What hardware is inside it?

The machine has 42 NVIDIA DGX A100 systems. Each system holds two AMD EPYC processors, eight A100 GPUs and 1 TB of memory. A GPU is a graphics processing unit. It does huge numbers of small sums at the same time, which suits AI.

That makes 336 A100 GPUs in all. The nodes talk to each other over Mellanox HDR InfiniBand, a very fast network link. C-DAC added its own HPC-AI engine, software frameworks and a cloud platform.

Speed figures for this machine come in three forms. Each one measures something different.

3. What does 210 AI petaflops mean?

A flop is one calculation with decimal numbers. A petaflop is one thousand million million of them each second. The NSM page gives the machine 210 petaflops for AI work.

The same page gives 6.745 petaflops for peak double-precision (DP) work. The 2020 announcement said 6.5 petaflops of peak DP. So the AI figure and the DP figure are not the same measure. Never compare them directly.

4. What did the TOP500 test measure?

The TOP500 list ranks the world’s fastest computers. It uses a test called Linpack. For this machine, TOP500 records 4.62 petaflops of sustained speed (Rmax). It records a theoretical peak (Rpeak) of 5.27 petaflops.

The entry shows 41,664 cores and the year 2020. The maker is listed as Bull. These numbers are lower than the 210 because the test uses double-precision maths.

Its rank changed with every new list. This is why old claims about it go stale.

5. Was it ranked 62nd or 63rd?

The Indian press and the IndiaAI portal said the machine ranked 63rd in the world in November 2020. The TOP500 entry lists the November 2020 rank as 62. The sources do not explain the gap.

Both sources agree on the main point. It entered the world’s top 100. At that time it was India’s largest and fastest supercomputer.

6. Did it stay in the world's top 100?

No. The TOP500 entry shows a steady slide as newer machines arrived. The ranks were 89 in June 2021 and 102 in November 2021. They were 111 in June 2022 and 120 in November 2022.

Then came 131 in June 2023, 162 in November 2023 and 185 in June 2024. So it left the top 100 in November 2021. The entry shows no later rank after June 2024.

The machine is meant as a shared national facility. Researchers and start-ups can use it.

7. Which problems is it meant to solve?

The NSM page says it serves as a cloud for AI and HPC work across India. Users include colleges, research institutes and start-ups. HPC means high-performance computing.

It lists health care, such as precision diagnostics. It lists farming, such as crop infestation and advice. Other uses are weather forecasting, security and surveillance, fraud detection and language tools.

8. What is the National Supercomputing Mission?

The NSM began in 2015. It places supercomputers in colleges and labs across India. Users reach them through the National Knowledge Network.

C-DAC says that by March 2025 the mission had set up 34 supercomputers. Together they gave 35 petaflops. PARAM Siddhi-AI is one of the systems listed on the mission’s website.

The machine did not stay as it was in 2020. Plans and numbers have moved since.

9. Was an H100 upgrade planned?

A news report dated 3 September 2024 said a C-DAC tender planned to upgrade the system. The plan would add eight NVIDIA H100 GPUs. The system then ran on NVIDIA A100 GPUs.

The report gives a plan, not a finished upgrade. No source seen here confirms a tender month. So remember only the year 2024.

10. Why do sources show both 42 and 82 nodes?

The NSM page describes 42 DGX A100 nodes and 336 GPUs in its first paragraph. Lower on the same page, a table lists 82 nodes and 656 GPUs under the name AIRAWAT-PSAI.

The pages do not explain the difference. The 2020 launch figure is 42 nodes. Treat 82 as a later or separate listing, and check the date before using it.

Questions test the maker, the mission, the place and the machine type. Learn the pairs below.

11. Which pairs and traps should you remember?

Match PARAM Siddhi-AI to C-DAC, Pune and the National Supercomputing Mission. Match its GPUs to NVIDIA and its design to the DGX SuperPOD plan. Remember 42 systems and 336 GPUs for the 2020 build.

It is not PARAM 8000. That was India’s first gigaflops-class parallel computer, finished in 1991 under Vijay Bhatkar. Also do not say it is still in the world’s top 100. It fell out of that group in 2021.