Wow, I am getting 0.5412 average sec/TS for several SM3MH models (in phase 3). Please note that my current model is NOT an ice-world.

I\'m curious, what is everyone\'s fastest sec/TS for the SM3 or SM3MH models that you\'re running? Oh, and what is your CPU?

Running MH models, I get 0.86 sec/TS consistently with my various Intel E8400\'s, each with 2GB RAM. If you would agree to \"unhide\" your computers on your preferences page, the rest of us could make comparisons!

John
GW3PRV

I\'m getting 2.3sec/ts on my AMD which usually runs HADCMs at about 70% of the speed of a core of the C2Duo 6600. As the AMD is an amalgam of recuperated parts with a broken on/off button, a lost restart button and a graphics card fan that\'s been broken for over two years, I\'m very pleased indeed with this performance.

---

Cpdn news

The best I\'ve gotten is .57 s/TS on an E8400 running at stock speed in Linux. That was running in tandem with a hadcm3. I haven\'t run one model at a time on that PC. Linux is definitely faster in the hadsm3 type models than Windows.

Edit...looks like DJ\'s is also an E8400 in Linux (overclocked?). Saw the link to a result in a previous post.

My times for HadSM3s:

2.5619 sec/TS, Win XP32, P4 @2.52GHz, 1GB DDR333@350MHz

1.3086-1.4199 sec/TS, Win XP32, Q6600 @2.4GHz 512MB DDR2 @800MHz (2 x VMWare 1 CPU Virtual Machine)

1.2106-1.2528 sec/TS, Same Q6600 VM\'s O/C @2.52GHz 512MB DDR2 @840MHz, Faster HDD

Hi,

I run the models on a pair of laptops. On my HP machine (AMD 1.7 GHz single coreprocessor and 1 GB of RAM) the HadSM3 that I am running now is running at 2.46s/TS. The other machine, an ACER, (an Intel
Core2duo 1.5 GHz with 2 GB of RAM) is running a HadSM3 on 1 core at 2.55s/TS. This might be a little faster, but, there is a HadCM3 80 year model running on the other core at 4.06s/TS.

Not to highjack this thread, but, I would be interested to know what other people are getting on HadCM3s per TS.

---

Hi,

I run the models on a pair of laptops. On my HP machine (AMD 1.7 GHz single coreprocessor and 1 GB of RAM) the HadSM3 that I am running now is running at 2.46s/TS. The other machine, an ACER, (an Intel
Core2duo 1.5 GHz with 2 GB of RAM) is running a HadSM3 on 1 core at 2.55s/TS. This might be a little faster, but, there is a HadCM3 80 year model running on the other core at 4.06s/TS.

Not to highjack this thread, but, I would be interested to know what other people are getting on HadCM3s per TS.

3.40 on my Opteron 1210 at 1.8 GHz running another project on the second core (Linux).
Tullio

---

I\'ve found HADSM speed to be much more variable, even on the same computer, than HADCM or HADAM. I assume this must be due to differences in HADSM parameter values.

Jim, what about starting a parallel thread to compare HADCM speeds?

---

Cpdn news

I\'ve found HADSM speed to be much more variable, even on the same computer, than HADCM or HADAM. I assume this must be due to differences in HADSM parameter values.

Jim, what about starting a parallel thread to compare HADCM speeds?

Will do.

---

My computer is here if you guys want to compare stats:
http://climateapps2.oucs.ox.ac.uk/cpdnboinc/show_host_detail.php?hostid=886140

Edit...looks like DJ\'s is also an E8400 in Linux (overclocked?). Saw the link to a result in a previous post.

No, I\'m running stock speeds. Fedora 8 linux. :)

OK, here is the fast host:
http://climateapps2.oucs.ox.ac.uk/cpdnboinc/show_host_detail.php?hostid=886140

And here is the 2.8GHz AMD Phenom. It\'s running at half the speed of the Intel Core2Duo. I didn\'t realize AMD chips suck so much.
http://climateapps2.oucs.ox.ac.uk/cpdnboinc/show_host_detail.php?hostid=955354

And here is the 2.8GHz AMD Phenom. It\'s running at half the speed of the Intel Core2Duo. I didn\'t realize AMD chips suck so much.
http://climateapps2.oucs.ox.ac.uk/cpdnboinc/show_host_detail.php?hostid=955354

They don\'t. The comparison isn\'t valid for a couple reasons. First, in hadsm3 type models, Linux is considerably faster than Windows. Second, the Intel compiler used to compile the executables does not allow full SSE2 or any SSE3 on AMD chips. I\'m not sure if SSE3 optimization is used in these, but SSE2 is heavily used. Check this link for more info on that.

Then, there\'s the capability of AMD quad cores to scale better in the high memory bandwidth models (hadam3s) compared to the Intel Q series quads. My Phenom X4 940 does not run one model as fast as an Intel Q9550, but runs four models faster.

Of course for single models, Wolfdale\'s rock. It would be interesting to see a Core i7 with hyperthreading turned off to see how fast one, two, three and four models would go. All Core i7\'s that I have viewed have hyperthreading on so the s/TS are only somewhat impressive.

They don\'t. The comparison isn\'t valid for a couple reasons. First, in hadsm3 type models, Linux is considerably faster than Windows.

Not true for me. 1.3 s/TS on Linux, slower than Windows.
http://climateapps2.oucs.ox.ac.uk/cpdnboinc/show_host_detail.php?hostid=955461

I normally run 1 climate model with other SETI or Rosetta tasks on the other core(s).

They don\'t. The comparison isn\'t valid for a couple reasons. First, in hadsm3 type models, Linux is considerably faster than Windows.

Not true for me. 1.3 s/TS on Linux, slower than Windows.
http://climateapps2.oucs.ox.ac.uk/cpdnboinc/show_host_detail.php?hostid=955461

I normally run 1 climate model with other SETI or Rosetta tasks on the other core(s).

Sorry, I should have been more specific. It\'s definitely faster on Intel in Linux than Windows. Patching the linux executable (the _um) with the perl script on the naughty-intel page results in full SSE2 and SSE3 utilized in Linux on AMD.

For example here. I\'m not sure the perl script on the archived naughty-intel page that I linked is truly downloadable anymore. If not, let me know if you want to go that way.

Sorry, I should have been more specific. It\'s definitely faster on Intel in Linux than Windows. Patching the linux executable (the _um) with the perl script on the naughty-intel page results in full SSE2 and SSE3 utilized in Linux on AMD.

For example here. I\'m not sure the perl script on the archived naughty-intel page that I linked is truly downloadable anymore. If not, let me know if you want to go that way.

Hmm... honestly, I think I\'m going to sell my AMD system. Thanks for the advice though.

On my current computer:
My 8 HADSM-MH models so far had a s/TS rate between 0.5187 and 0.6130, the current 9th is running at 0.6410.

My 7 HADSM varied between 0.6125 and 0.5274.

I\'ve registered that some other projects, and especially another CPDN-WU running in parallel are significantly slowing down the crunching.

You can see quite exactly when another CPDN, in this case a beta, ran in parallel.

---

Grüße vom Sänger

On an AMD Opteron 275 @2.2GHz with 4 GB RAM on Linux Fedora Core 3
HADSM3f currently getting 1.97 to 1.964 s/TS
HADSM3m currently getting 2.007 to 2.011 s/TS

These two often run at the same time as well as mixing with other projects.
As the SM3f is the more advanced it mostly runs solo so gets the lower times.

On an AMD Opteron 285 @2.6GHz with 2 GB RAM on Linux Fedora Core 3
HADSM3f last time I did one of these on this machine was 25/7/07 it was getting 1.735 to 1.765 s/TS
HADSM3m currently getting 1.947 to 1.983
When a second HADSM3m starts running it\'s speed drops to 2.05 to 2.07 s/TS.

Despite the faster speed of the second machine they run very similar times.

---

On my current computer:

That is quick! I wonder if the Core i7 is any faster?

Looks like I may have answered my own question. My Core i7 920 is up running a HadSM3MH model at 0.5134 s/TS. Note that this model was 2/3 done by the older computer. This is slightly faster than my old Core2Duo E8400 (that ran 333MHz faster).

In terms of floating point performance, the chips are about the same per core. Having hyperthreading enabled only decreases FP performance by 20% per thread (2312 vs 2907). Integer performance goes to 50% with HT enabled (4256 vs 8643).

Are CPDN models floating-point bound processes? Or would it make more sense to have HT disabled for best performance? By performance, I mean RAC.