On-Prem

HPC

India flips the switch on three homebrew supercomputers

Rudra machines use local designs for server, interconnect, and cooling – but seemingly not a planned 96-core Arm CPU


Indian prime minister Narendra Modi yesterday dedicated three new supercomputers, and made the machines a symbol of his economic, social, and industry policies.

Those emphases mean Modi didn't detail the tech specs of the three "PARAM Rudra" machines – but the orgs that will run them have revealed some info.

India's National Centre for Radio Astrophysics has revealed that its machine features "several thousands of Intel CPUs and 90 state-of-the-art Nvidia A100 GPUs, 35 terabytes of memory, and two petabytes of storage."

The A100 is not state of the art – Nvidia's current champ is the H100 and the H200 will arrive any week now.

The S.N. Bose Centre for Basic Sciences mentioned an 838 TFLOPS system in a social media post about the PM's launch.

The Inter-University Accelerator Centre appears not to have disclosed the disposition of its machine, but has previously teased a three-petaflop machine built on Intel Xeon 2nd Gen scalable processors with up to 24 cores each, packing four petabytes of storage capacity and hooked up to a 240Gbit/sec interconnect.

That spec sheet is notable because it matches info in published documents [PDF] that describe [PDF] the spec for India's first-gen Rudra-class machines as employing:

Second-gen Rudra machines support fourth-gen Xeons and four GPUs.

A planned third generation of Rudra machines will use a 96-core processor designed by India's Centre for Development of Advanced Computing using the Arm 8.4 architecture and built by TSMC on its 5nm process.

The Register can't imagine Modi would not have trumpeted the arrival of that processor in these supers. We're guessing they're still in the works.

This launch is still a big deal for India – and Modi – as it adds to the national supercomputing fleet and means locally designed tech has been deployed. Given India's size and rapid pace of development, HPC vendors may wonder what they need to do to get into the market.

Modi also flipped the switch on an HPC system dedicated to weather and climate research. Details are again scarce, but the Indian Institute of Tropical Meteorology has stated the system runs in two locations: one with 11.77 PETAflops capacity and 33 petabytes of storage, and the other with 8.24 PetaFLOPS and 24 petabytes. Both run alongside a 1.9 PetaFLOP machine dedicated to AI and ML. ®

Send us news
1 Comment

China announces plan to label all AI-generated content with watermarks and metadata

PLUS: Foxconn wants 40 percent AI server market share; Atlassian CEO jets into controversy; Starlink reaches 800 million in India; And more!

ESA cuts the ribbon on 34,000-core Space HPC center tailored for space workloads

PUE of 1.09 and heats the building it lives in

Non-x86 servers boom even faster than the rest of the AI-infused and GPU-hungry market

Analyst finds 91 percent revenue growth with white box makers leading the way

India wants backdoors into clouds, email, SaaS, for tax inspectors

PLUS: Malaysia teams with Arm for local chip designs; NTT warns of possible breach; Samsung strikers settle; and more

Nvidia's Vera Rubin CPU, GPU roadmap charts course for hot-hot-hot 600 kW racks

Now that's what we call dense floating-point compute

India investigates whether Uber makes iPhone users pay more to ride

Rideshare companies said it isn’t happening. Lawmakers aren’t convinced

Do you DARE? Europe bets once again on RISC-V for supercomputing sovereignty

€240M found for three-year sprint to develop three chiplets for HPC, AI

India's top telco plans cloud PCs for its 475 million subscribers

PLUS: China bans AI leaders from visiting USA; Acer data leak suspect cuffed; and more

Three charged in Singapore with alleged link to illicit shipments of Nvidia GPUs to China

Accused face up to 20 years in prison

Ampere bets on Arm to muscle into Intel's telco territory

Chipmaker touts high-core, low-power Altra processors as the future of 5G and AI inferencing

Like a kid handing in homework at the last minute, Supermicro finally files its missing financial figures

SMCI had to come up with long-delayed report – or lose its slot on NASDAQ again

Apple promises to spend $500B, hire 20K over 4 years to swerve Trump import tariffs

Sorry, that should read: Boost US manufacturing and R&D, believe in the American people, etc etc