HPE pulls sheets off largest Arm-based supercomputer Astra

HPE is building the world's largest Arm-based supercomputer, Astra – 2.5 petaFLOPS from 2,592 HPE Apollo 70s – for Sandia National Labs in the US, where it will run advanced modeling and simulation workloads in areas including national security and energy. Apollo 70s have four compute nodes in a 2U case. The CPUs are Cavium …

[rsondhi]

What is the connectivity between these Apollo compute nodes??

Is it Mellanox adapters? Is RDMA involved (if it's Ethernet and not IB)??

When discussing supercomputing and the nodes involved i would think that the inter-node connectivity is a critical element!

[HPCJohn]

Re: What is the connectivity between these Apollo compute nodes??

The HP product page says Mellanox ConnectX-5

https://www.hpe.com/us/en/product-catalog/servers/apollo-systems/pip.hpe-apollo-70-system.1010742472.html

[Anonymous Coward]

A significant article

I think it could be significant that at least one major system h/w manufacturer now thinks that they can make a worthwhile profit from an ARM based system.

[FrankAlphaXII]

Re: A significant article

Supercomputers aren't usually profitable, nor are they really supposed to be. They're machines built to show off potential capabilities than anything a normal customer is going to use for something in production, except when that customer is like TLA/TLA, a research university, or one of the National Labs.

Also, keep in mind that HPE is the least profitable in terms of earnings per share of the Hewlett-Packard spawn (HP Inc, HPE, and DXC). It was like 8 cents in July. HPI was at like 50 cents and DXC was around two bucks.

None of this really matters though, if SNL wants to try this then they're going to try it. Sandia really doesn't care about the IT market's fluctuations and what the beancounters say. They care about advancing the state of the art no matter what it costs. They're an incredibly cool place to work because of things like this, if of course you can put the whole weapons of mass destruction thing aside.

[trevorde]

Will it run Doom?

or DOS?

[StargateSg7]

Re: Will it run Doom?

You should see OUR supercomputer which runs CRYSIS at 65536 by 65536 64-bit RGBA pixels at its full 120 fps!

I should note though OUR supercomputer here in Vancouver, British Columbia, Canada heats up the company swimming pool to a nice toasty sub-tropical temperature via its opto-electronic, custom CPU/GPU design using GaAs on Sapphire to the tune of 119 ExaFLOPS (128-bit FP) SUSTAINED which is most definitely NUMBER ONE IN THE WORLD --- PERIOD !!!

That single computer barely the size of a large home swimming pool OUTPERFORMS EVERY SINGLE Top-500 listed supercomputer COMBINED !!!!

That New 200 PetaFLOPS not-so-supercomputer from the USA is a mere PIPSQUEAK ANT against our 10,000 Lbs Gorilla of a 119 ExaFLOP MONSTER MACHINE !!! That's 595 TIMES FASTER than the US machine !!! It is NOWHERE NEAR OUR LEAGUE !!!

We won the pssing contest HANDS DOWN !!!!!!

Top That!!!

P.S. It runs electro-chemical simulations of various human neural structures...and YES...it works VERY WELL as the world's first equal-to-100-IQ-functional A.I. -- Very Soon Now it will be Post-160-IQ firmly putting it into the TRUE Super-Intelligence category !!!