FYI - the supported languages are FORTRAN, C and C++.
Folk wanting in on the Arm-based goodness baked into Japan's upcoming "Post-K" exascale supercomputer are in luck – Fujitsu has finished the design and sales of commercial versions will begin some time between October 2019 and March 2020. The $910m Post-K will become the country's most powerful machine once it is switched on …
So its 910 Million for this computing power.
That could buy an awful lot of lesser computers
Obviously I know lots of things are massively computationally expensive (e.g. weather forecasts)
But was just wondering if this is (currently) good value (compared to other mega machines or even as a silly stat how pricey is it per flop compared to a bog standard PC or mobile), or if its just a pricey proof of concept and commercial models will be better value
It is good value. The cost per core is not really important, it's the interconnect (Tofu) that makes using those cores in big science applications fast.
With that, the peak:average performance ratio across a range of application types is close to 1:1. With just a large number of ordinary computers on an Ethernet, this ratio would be much, much lower for most tasks.
Tofu is completely mind bending (6 dimensional hypertoroid mixed up with a 4 dimensional hypercube), and they build it into the CPUs for the full effect of ultra - low latencies. A lot of other machines use infiniband, I think using PCIe NICs, which is fast but I don't think it's as good as Tofu.
The predecessor machine, K, was notable in this regard, meaning that it's benchmark figures were very good and very translatable into application performance.
Besides the new flagship, the company is considering a cheaper "entry-level model" – and said this could be built by a third party.
"Arm's successful collaboration with Fujitsu and Riken underscores our commitment to the HPC ecosystem," said Brent Gorda, senior director of HPC at Arm.
Is such Collaboration envisaged and expected to be Via AI Virtualised Realty ProMotions .... Exhibiting Addictively Engaging Programs for your Own Chosen Future Paths?
Which as you can appreciate is Presently a Holy Mess of Mass Conflicted Effort.
ITs Solution in a Mass Televising of the Search for ITs Immaculate Source by Engaged Nations in Novel JOINT AI Operations.
USCYBERCOM/APT10 Intellectual Property Territory, methinks .... and extremely lucrative and rewarding.
Will all models be able to Run Extreme Mega Meta Data Processing for Powers in Need of ITs Available Awesome Abilities in Utilities with Facilities Expert in Greater IntelAIgent Games Broad Band Casting.
And Here BetaTesting SMARTR AI Conceptual Play with Others for Signs of Engagement with Other IntelAIgent Phorms of Life.
Coming from a screaming streaming media screen heralding great news out of the blue and presenting it here in simple black on white text for TeleAudioVisualisation Programs to Project/Virtually Realise to Create Earthly Presents, now and drawing attention to it, so that one pause for a ponder and begin to wonder on the wanders of life in Live Operational Virtual Environments.
What's not to Like, when All can be Done Commanded and Controlled by So Few?
I imagine the Utility License Fees on that AIMother would need to be horrendous to ensure only the best and most needy are routinely emergency supplied with Bountiful Services in Especially Short Supplies before all others taste the results of ITs Advanced IntelAgent Program Deliveries.
There's a lot going on out there, El Reg. And yet, have you noticed the blissful silence from mass media moguls on, well, if IT must wear a name, Post Modern Virtualised Matters*.
Biting the hand that feeds IT © 1998–2020