By: RichardC (tich.delete@this.pobox.com), February 1, 2017 7:05 am
Room: Moderated Discussions
Ireland (boh.delete@this.outlook.ie) on January 31, 2017 6:58 pm wrote:
> The best place for a supercomputer is out in the oceans surely? The best place for 'big data' is
> on dry land. And the best thing to connect one to the other, are the satellites for a means of communication.
There's no evidence at all that anyone wants to put this on a ship. *And* there's no evidence
at all that ARM-based architecture would help with the problem of putting a big computer on a
ship. You're stacking a big tower of implausible speculations on top of each other.
From where I stand, this appears to be the common pattern of people trying to use the technologies
developed for the highest-volume market, which are cheap and highly optimized, and grow them up
into smaller, but higher-margin market segments. The exact same way that the Pentium/Windows
dominance of the desktop was used to grow up into PentiumPro/Pentium-II/III and WindowsNT workstations, and Xeon servers.
And the arguments we're having about this are the same old ones - "mass-market architecture X
won't work for high-end application Y because it lacks essential feature Z". Maybe those obstacles
will be too big to overcome; or maybe there will be a few cracks in the wall which allow
low-end architecture X to creep through in a few niches; or maybe the higher-end architecture
will grow down far enough to fend off the threat (Xeon-D is certainly a good step towards a
lower-TCO x86 for throughput-optimized clusters).
So ISTM the effort to push ARM+GPGPU into supercomputing is not obviously stupid; and even it fails,
it might force - it might already have forced - Intel/AMD to improve their products for
parallel throughput, which would be a Good Thing (especially for those living on low-lying ground
and hoping for better weather forecasts - and I'm only about 15ft above sea level myself).
> The best place for a supercomputer is out in the oceans surely? The best place for 'big data' is
> on dry land. And the best thing to connect one to the other, are the satellites for a means of communication.
There's no evidence at all that anyone wants to put this on a ship. *And* there's no evidence
at all that ARM-based architecture would help with the problem of putting a big computer on a
ship. You're stacking a big tower of implausible speculations on top of each other.
From where I stand, this appears to be the common pattern of people trying to use the technologies
developed for the highest-volume market, which are cheap and highly optimized, and grow them up
into smaller, but higher-margin market segments. The exact same way that the Pentium/Windows
dominance of the desktop was used to grow up into PentiumPro/Pentium-II/III and WindowsNT workstations, and Xeon servers.
And the arguments we're having about this are the same old ones - "mass-market architecture X
won't work for high-end application Y because it lacks essential feature Z". Maybe those obstacles
will be too big to overcome; or maybe there will be a few cracks in the wall which allow
low-end architecture X to creep through in a few niches; or maybe the higher-end architecture
will grow down far enough to fend off the threat (Xeon-D is certainly a good step towards a
lower-TCO x86 for throughput-optimized clusters).
So ISTM the effort to push ARM+GPGPU into supercomputing is not obviously stupid; and even it fails,
it might force - it might already have forced - Intel/AMD to improve their products for
parallel throughput, which would be a Good Thing (especially for those living on low-lying ground
and hoping for better weather forecasts - and I'm only about 15ft above sea level myself).