Indian Prime Minister Narendra Modi, according to The Register, announced the commissioning of three new high-performance computing systems PARAM Rudra. The launch of these supercomputers is said to be a “symbol of the economic, social and industrial policy” of the country.
Modi did not go into details about the technical characteristics of the cars during the presentation. However, some information was disclosed by organizations that will directly operate these LDC systems.
One of the supercomputers is located at India’s National Center for Radio Astrophysics (NCRA). This machine is equipped with “several thousand Intel processors” and 90 NVIDIA A100 accelerators, 35 TB of memory and 2 PB of storage. Another NRS complex is installed at the S. N. Bose Center for Basic Sciences (SNBNCBS): it is known that it has a speed of 838 Tflops.
The operator of the third system is the Inter-University Center for Accelerated Computing (IUAC): this supercomputer with a performance of 3 Pflops uses 24-core Intel Xeon Cascade Lake-SP chips. The storage capacity is 4 PB. An interconnect with a bandwidth of 240 Gbit/s is mentioned.
The Register notes that these characteristics generally correspond to the description of the first generation Rudra supercomputers. According to available documentation, such machines use:
- Half-width motherboard for 1U or 2U servers – up to 64 servers in a rack with a total power of 40 kW;
- Two Intel Xeon Cascade Lake-SP processors;
- Two unnamed GPU accelerators;
- Two NVMe SSD standard U.2;
- Two 10GbE ports and an additional network adapter;
- Trinetra interconnect – six full-duplex interfaces with a speed of 100 Gbit/s;
- Direct liquid cooling technology of our own design.
It is expected that the second generation Rudra machines will receive support for Xeon Sapphire Rapids processors and four GPU accelerators. The third generation supercomputers will use 96-core Arm AUM processors developed by the Indian Advanced Computing Development Center: these products will be manufactured using TSMC 5nm technology.
Image Source: Pixabay.com
Meanwhile, Eviden (a subsidiary of Atos) announced the delivery of two new supercomputers to India. One of them is installed at the Indian Institute of Tropical Meteorology (IITM) in Pune, the second at the National Center for Medium Range Weather Forecasting (NCMRWF) in Noida. These systems, built on the BullSequana XH2000 platform, are designed for weather and climate research. AMD, NVIDIA and DDN took part in the creation of the complexes.
The IITM system, called ARKA, has a speed of 11.77 Pflops: 3021 nodes with AMD EPYC 7643 (Milan), 26 nodes with NVIDIA A100, NVIDIA Quantum InfiniBand and 33 PB storage (previously 3 PB SSD + 29 PB HDD) . In turn, the NCMRWF supercomputer called Arunika has a performance of 8.24 Pflops: 2115 nodes with AMD EPYC 7643 (Milan), NVIDIA Quantum InfiniBand and DDN storage EXAScaler ES400NVX2 (2 PB SSD + 22 PB HDD). The system also includes a dedicated 1.9 Pflops AI and ML application engine (precision not specified) consisting of 18 nodes with NVIDIA A100.