Version 4.14 has just been released on win32 & linux.
There should be an improvement in speed with this build.

---

Great news, Tolu thank you.

Is 4.14 for slab?

So people with models that are less than 5% complete may want to consider suspending their model, get a new one and check the speed. If it is more than 5% faster, is it appropriate to abort the old slow one?

There is a possibility of converting a run from 4.13 to 4.14 but it sounds a bit complicated to me. Any thoughts on whether this should be considered or avoided?

I suspect AMD processors will see more of a speed up than intel processors.

Finally, I suggest that people that have had problems running BOINC CPDN may want to try again now. This might be worth having a news items.

---

Visit BOINC WIKI for help

And join BOINC Synergy for all the news in one place.

Great news, Tolu thank you.
Is 4.14 for slab?

Yes, great news.
It is a slab model - just got one \'cause didn\'t want to till weeked when my regular slab 4.13 finish.

Memory usage is about the same.
It is soon to judge speed improvment after 10 mins of running but BOINCView suggest 1.9 vs 1.7 sec/TS (20days 2 hours vs. 18 days 5 hours) on Pentium D.
I\'m interested in some figures for AMD as topic of AMD performance on CPDN is of interest to many CPDNers.

Executables are about the same size.
hadsm3se_4.14_windows_intelx86.zip now includes also globe.rgb and globe.tga which makes the initial download 4MB smaller.
Good timing since there are hundreds of new users (from SETI Classic).
http://www.boincstats.com/stats/project_graph3.php?pr=cpdn&table=users

---

<i>phpBB forum for CPDN, all are </i><a href="http://www.climateprediction.net/board">invited</a>

I\'m interested in some figures for AMD as topic of AMD performance on CPDN is of interest to many CPDNers.

On an AthlonXP 3200+, it went from about 2.27 s/TS to 2.02 s/TS, or 20.4 days to 18.2 days per model run. So about an 11% improvement in speed.

On an Athlon64 3400+ (2.2 GHz), it went from about 1.95 s/TS to 1.69 s/TS, or 17.6 days to 15.2 days per run. So about a 13% improvement in speed.

Thanks, geophi.
Just about the same as for Intel (so far). This leave the (im)balance of AMD vs. Intel on CPDN quite intacted.

When comparing your numbers, I just found that BOINCView gives ill numbers when calculating Completion times [for CPDN]; your calculations are correct.
We should rely on sec/TS (since the early days of CPDN) and figure days-to-complete from them.

Even a 5% gain would be worth the effort. With over 10%, a proposition suggested by Chris may be even more interesting.

It will be interesting to get some Linux numbers as well...

---

<i>phpBB forum for CPDN, all are </i><a href="http://www.climateprediction.net/board">invited</a>

Just about the same as for Intel (so far). This leave the (im)balance of AMD vs. Intel on CPDN quite intacted.

If you read the \"Benchmarks Beware\" section at
http://www.swallowtail.org/naughty-intel.html
you will see what is likely the problem why AMD didn\'t improve in performance more. While he is talking about Intel compiler version 8 in that section of the article, he basically said that with the -xW compiler switch used, some SSE2 is used on AMD64 chips (while none was used in 7), \"but calls to the vectorised single-precision math instrinsics will use SSE, not SSE2.\" So it appears that it is now using some optimization on AMD64 chips, but not as much as it could/should.

On an Intel P4 3.4 GHz running one model, in Linux, perhaps a 5% increase in performance. It\'s difficult to tell as that starts to fall within the variability of model speed just due to parameter differences.

At work, on an Intel Tualatin Celeron 1.4 GHz with 100 MHz SDRAM, the speed went from about 6.55 to 5.80 s/TS, about an 11% performance improvement.

Hmm.. only got downloading errors with 4.14.

Hopefully there will also be a new sulphur executable.
On an Linux AMD XP-M @2247MHz sulphur_4.21 shows 3.65 s/TS but in spinup the sulphur part runs at 3.00.

---

Is there any official changelog?

---

Linux Users Everywhere @ BOINC

Hmm.. only got downloading errors with 4.14.

Hopefully there will also be a new sulphur executable.
On an Linux AMD XP-M @2247MHz sulphur_4.21 shows 3.65 s/TS but in spinup the sulphur part runs at 3.00.

yep this should be resolved in the new version 4.22

---