Rattledagger
Elite Member
For anyone that is interested in Climateprediction.net, but don't want to run a model for many weeks to not say months for the longest models, a new model-type is now released, the HADAM3P-model.
Like all CPDN-models, Trickle-messages will be sent to server during the progress of the model, so will get a fairly continuous stream of "credit". For a C2D it should trickle every 3-4 hours. But, please note, the Trickle-messages is basically only for credit-purposes, file-uploads only happens at the end of a wu, so please finish the wu if you've downloaded it.
For speed, my C2D-E4400-2GHz uses around 4.6 s/TS, this equals 92 hours or a little less than 4 days. My old Opteron242 1.6 GHz uses around 10 s/TS, this means 200 hours or a little over 8 days.
Not my computer, but a i7-920 seems to use around 3.6 s/TS running 6 models, this means 72 hours or 3 days per model.
So, for anyone that wants a fairly short CPDN-wu, now is your chance to try-it. 😎
System-requirements: 350 MB memory, uses 220 MB memory. Disk, 450 MB.
OS: Win2k or later, Linux, Intel/Mac.
Downloads: Application-download, close to 85 MB.
Per wu: AFAIK around 35 MB. (meaning initial download roughly 140 MB)
Upload: 5 files uploaded at end of run, total size AFAIK around 50 MB.
Due to the large download-size, the large upload-size and fast crunch-times, this model is not for dialup-users.
For anyone else that wants to try-it, choose to attach to http://climateprediction.net (if not already attached).
Choose the "HADAM3P" under http://climateapps2.oucs.ox.ac...efs.php?subset=project
And, good luck.
Note, it seems HADAM3P is very memory-dependent, if you run more than one HADAM3P-model at the same time on a dual/quad-core, all models will slow-down, ocassionally a huge slow-down. In my experience, my C2D jumps from 4.4 s/TS then running one model to 5.6 s/TS if runs two models, this is roughly 27%. The Opteron is around 15% variation
One post during beta-testing shows even larger variations on quad-computers:
Intel Q9550: 1 core: 2.45 s/TS, 49 hours or just over 2 days for a model. :Q
But, 4 core: 3.95 s/TS. This is a 61% slowdown.
Phenom X4: 940: 1 core: 2.65 s/TS; 4 cores: 3.65 s/TS. This is 38% slowdown.
Like all CPDN-models, Trickle-messages will be sent to server during the progress of the model, so will get a fairly continuous stream of "credit". For a C2D it should trickle every 3-4 hours. But, please note, the Trickle-messages is basically only for credit-purposes, file-uploads only happens at the end of a wu, so please finish the wu if you've downloaded it.
For speed, my C2D-E4400-2GHz uses around 4.6 s/TS, this equals 92 hours or a little less than 4 days. My old Opteron242 1.6 GHz uses around 10 s/TS, this means 200 hours or a little over 8 days.
Not my computer, but a i7-920 seems to use around 3.6 s/TS running 6 models, this means 72 hours or 3 days per model.
So, for anyone that wants a fairly short CPDN-wu, now is your chance to try-it. 😎
System-requirements: 350 MB memory, uses 220 MB memory. Disk, 450 MB.
OS: Win2k or later, Linux, Intel/Mac.
Downloads: Application-download, close to 85 MB.
Per wu: AFAIK around 35 MB. (meaning initial download roughly 140 MB)
Upload: 5 files uploaded at end of run, total size AFAIK around 50 MB.
Due to the large download-size, the large upload-size and fast crunch-times, this model is not for dialup-users.
For anyone else that wants to try-it, choose to attach to http://climateprediction.net (if not already attached).
Choose the "HADAM3P" under http://climateapps2.oucs.ox.ac...efs.php?subset=project
And, good luck.
Note, it seems HADAM3P is very memory-dependent, if you run more than one HADAM3P-model at the same time on a dual/quad-core, all models will slow-down, ocassionally a huge slow-down. In my experience, my C2D jumps from 4.4 s/TS then running one model to 5.6 s/TS if runs two models, this is roughly 27%. The Opteron is around 15% variation
One post during beta-testing shows even larger variations on quad-computers:
Intel Q9550: 1 core: 2.45 s/TS, 49 hours or just over 2 days for a model. :Q
But, 4 core: 3.95 s/TS. This is a 61% slowdown.
Phenom X4: 940: 1 core: 2.65 s/TS; 4 cores: 3.65 s/TS. This is 38% slowdown.
