Met Office to build £97m supercomputer
 

http://www.bbc.co.uk/news/science-environment-29789208

Should increase forecast accuracy.

Met Office to build £97m supercomputer
 

On 28/10/2014 15:01, Dawlish wrote:

http://www.bbc.co.uk/news/science-environment-29789208

Should increase forecast accuracy.

Shouldn't that be Cray to build supercomputer, or more correctly
to install it?

I can't help wondering why they didn't install a Cray rather than
an IBM six years ago. A cheaper option perhaps.

--

Brian W Lawrence
Wantage
Oxfordshire

Met Office to build £97m supercomputer
 

On 28/10/2014 17:06, Brian Lawrence wrote:
On 28/10/2014 15:01, Dawlish wrote:

http://www.bbc.co.uk/news/science-environment-29789208

Should increase forecast accuracy.

Shouldn't that be Cray to build supercomputer, or more correctly
to install it?

I can't help wondering why they didn't install a Cray rather than
an IBM six years ago. A cheaper option perhaps.

I presume competitive tender and IBM was cheaper then.
Also you tend to find bugs when code is ported from one machine
architecture to another so a change every now and then is a good thing
(although fans of one specific brand of machine may not think so).

The new one is an order of magnitude more powerful.

But in a full 3D simulation that only refines the computing grid by
12^(1/3) ~ 2.3x on linear scale.

--
Regards,
Martin Brown

Met Office to build £97m supercomputer
 

Brian Lawrence wrote:
On 28/10/2014 15:01, Dawlish wrote:

http://www.bbc.co.uk/news/science-environment-29789208

Should increase forecast accuracy.

Shouldn't that be Cray to build supercomputer, or more correctly
to install it?

I can't help wondering why they didn't install a Cray rather than
an IBM six years ago. A cheaper option perhaps.

--------------------------------------------------
Indeed - I don't think the Met Office are going to compete with the
likes of IBM in building computers. I wonder really if that amount of
money is getting into the realms of diminishing returns. I didn't think
it was long ago I read about a new super computer. Is this another one
or just the other one being delayed?
Dave

Met Office to build £97m supercomputer
 

On 29/10/2014 01:55, Dave Cornwell wrote:
Brian Lawrence wrote:
On 28/10/2014 15:01, Dawlish wrote:

http://www.bbc.co.uk/news/science-environment-29789208

Should increase forecast accuracy.

Shouldn't that be Cray to build supercomputer, or more correctly
to install it?

I can't help wondering why they didn't install a Cray rather than
an IBM six years ago. A cheaper option perhaps.

--------------------------------------------------
Indeed - I don't think the Met Office are going to compete with the
likes of IBM in building computers. I wonder really if that amount of
money is getting into the realms of diminishing returns. I didn't think
it was long ago I read about a new super computer. Is this another one
or just the other one being delayed?
Dave

Weather (and for that matter universe) simulation is always on bleeding
edge kit or very close to the limits. It goes with the territory.

There are a class of problems beyond our present computational ability
typically where the program runtime exceeds about 3 years where the
fastest route to a computational solution is work on something else for
two years and allow Moores law to do its stuff on the hardware.

http://en.wikipedia.org/wiki/Moore%27s_law

This regime may change shortly it is getting harder and harder to boost
CPU performance and the drive for ever faster domestic PCs has largely
gone. They are now fast enough for all practical purposes.

--
Regards,
Martin Brown

Met Office to build £97m supercomputer
 

On Wednesday, October 29, 2014 9:05:12 AM UTC, Martin Brown wrote:


This regime may change shortly it is getting harder and harder to boost
CPU performance and the drive for ever faster domestic PCs has largely
gone. They are now fast enough for all practical purposes.

--
Regards,
Martin Brown

------------------------------------------------------------------------------

Such a powerful computer we are told is needed to give hourly forecasts at
1.5 km resolution on an operational basis.

But we already have these postcode forecasts on an hourly basis 24 hrs ahead.

http://www.metoffice.gov.uk/public/w...ime=1414540800

Someone has been telling a porky pie.

13 times faster does not necessarily mean better accuracy.
Someone must have done a cherry picked forecast experiment at the said resolution to show the money is justified.

Or,am I just becoming an old cynic?

Len
Wembury


----------------------------------------------------------------------------

Met Office to build £97m supercomputer
 

On 29/10/2014 14:58, Len Wood wrote:


------------------------------------------------------------------------------

Such a powerful computer we are told is needed to give hourly forecasts at
1.5 km resolution on an operational basis.

But we already have these postcode forecasts on an hourly basis 24 hrs ahead.

http://www.metoffice.gov.uk/public/w...ime=1414540800

Someone has been telling a porky pie.

13 times faster does not necessarily mean better accuracy.
Someone must have done a cherry picked forecast experiment at the said resolution to show the money is justified.

Or,am I just becoming an old cynic?

Len
Wembury


----------------------------------------------------------------------------

The model being referred to is the global model. This is the basis on
which other models will run, which for the hourly forecasts are run with
a smaller grid length and shorter time-steps as well as being for
smaller areas (e.g. UK). See
http://www.metoffice.gov.uk/research...er-forecasting
for more information.
So no porky pies I'm afraid. The improvements will be from the reduced
grid-lengths, more layers and better science, and less parametrization.

Met Office to build £97m supercomputer
 

On Wednesday, October 29, 2014 2:58:50 PM UTC, Len Wood wrote:


Such a powerful computer we are told is needed to give hourly forecasts at
1.5 km resolution on an operational basis.

But we already have these postcode forecasts on an hourly basis 24 hrs ahead.

Someone has been telling a porky pie.



============

Not at all.

a) What Martin said.

b) The postcode forecasts are produced through model downscaling with reference to the closest grid points or perhaps the closest and most pertinent MOS sites, if the Met office uses MOS these days.

Stephen.

Met Office to build £97m supercomputer
 

"Stephen Davenport" wrote in message
...
On Wednesday, October 29, 2014 2:58:50 PM UTC, Len Wood wrote:


Such a powerful computer we are told is needed to give hourly forecasts
at
1.5 km resolution on an operational basis.

But we already have these postcode forecasts on an hourly basis 24 hrs
ahead.

Someone has been telling a porky pie.



============

Not at all.

a) What Martin said.

b) The postcode forecasts are produced through model downscaling with
reference to the closest grid points or perhaps the closest and most
pertinent MOS sites, if the Met office uses MOS these days.


The Met Office gave up MOS for postcode forecasting years ago!

The buzzwords now are "intelligent downscaling" . Cannot say any more for
fear of giving away state secrets, but the code is interesting :-)

Will
--
http://www.lyneside.demon.co.uk/Hayt...antage_Pro.htm
Will Hand (Haytor, Devon, 1017 feet asl)
---------------------------------------------

Met Office to build £97m supercomputer
 

On Wednesday, October 29, 2014 3:41:59 PM UTC, wrote:

The buzzwords now are "intelligent downscaling" . Cannot say any more for
fear of giving away state secrets, but the code is interesting :-)


=========

Aye, I suspect that we all use similar methods towards the same end.

Stephen.