Continuing from the php forum thread, I've been looking at the <a href="http://www.metoffice.com/research/hadleycentre/pubs/HCTN/">Hadley Centre technical papers</a> and it looks like HadGEM1 requires 12-15 times the processing power of HadCM3 (to be used for experiment2 and itself requiring about 1.5 times the processing power of the HadSM3 model we are currently running) so its probably not practical to run on current PCs. There is a slab model equivalent HadGSEM1.

HadGEM1 has beter skill than HadCM3, that is it is better able to model features of past climate, in a few areas it is worse, but most it is better. Unfortunately the two major areas it is worse are in global temperature and precipitation, the most important for the cpdn experiments.

Future PCs are likely to have multi-core CPUs, 2-core this year perhaps up to 8-core in couple of years. From what I've read HadGEM1 is designed to be parallisable on such CPUs and so I would expect a high end PC from 2007 to run it fast enough to be useful. Too late for the IPCC AR4, but it seems possible that cpdn experiments in support of IPCC AR5 will be using HadGEM.

---

____________________________<br>
<a href="http://www.boincforum.info/boinc/">boinc forum</a> and <a href="http://www.uk4cp.co.uk/">United Kindom</a> team, my climate change <a href="http://www.livejournal.com/users/mike_atkinson/">blog</a>.

Hi Mike,

Estimation and information you are providing seems accurate. One might wonder what amount of MBs (or even GBs) of RAM is needed in order to make HadGEM running; i assume that precission (number of cells and levels) is higher that CPDN's HadSM.

&gt; Hi Mike,
&gt;
&gt; Estimation and information you are providing seems accurate. One might wonder
&gt; what amount of MBs (or even GBs) of RAM is needed in order to make HadGEM
&gt; running; i assume that precission (number of cells and levels) is higher that
&gt; CPDN's HadSM.
&gt;

The HadGEM model can be run at the same lat/lon resolution as HadCM3 but twice the levels in ocean and atmosphere (N48L38), but it is really designed to run at twice the lat/lon number of grid points (N96L38), see <a href="http://www.metoffice.com/research/hadleycentre/pubs/HCTN/HCTN_55.pdf">HadGEM1 model description and analysis</a> page 4. The "new dynamics" library is designed to work at high resolutions. I would expect working memory to be about 10 times that of HadCM3, maybe about 500MB, not to be sneezed at, but much less than the probable capabilities of a year 2007 PC.

---

____________________________<br>
<a href="http://www.boincforum.info/boinc/">boinc forum</a> and <a href="http://www.uk4cp.co.uk/">United Kindom</a> team, my climate change <a href="http://www.livejournal.com/users/mike_atkinson/">blog</a>.

&gt; The HadGEM model can be run at the same lat/lon resolution as HadCM3 but twice
&gt; the levels in ocean and atmosphere (N48L38), but it is really designed to run
&gt; at twice the lat/lon number of grid points (N96L38), see <a href="http://www.metoffice.com/research/hadleycentre/pubs/HCTN/HCTN_55.pdf">HadGEM1
&gt; model description and analysis</a> page 4. The "new dynamics" library is
&gt; designed to work at high resolutions. I would expect working memory to be
&gt; about 10 times that of HadCM3, maybe about 500MB, not to be sneezed at, but
&gt; much less than the probable capabilities of a year 2007 PC.
&gt;
Thanks for the link, good reading...
HadCM3 became operational towards the end of 1997, so projecting a normal rate of
increase in computer power (doubling every 18 months) up until 2003/04 (the target
timetable for operational delivery of HadGEM1) suggested that a model of approximately
15-20 times the cost of HadCM3 would be affordable at that time. This was factored into the
design to allowing something like a 6-8 times cost increase from higher resolution, with a
further 2-3 times from additional physical complexity.