LINUX Beta Test App 4.17 available
Bernd Machenschalk
Moderator
Administrator
Joined: 15 Oct 04
Posts: 4330
Credit: 251607438
RAC: 36662
29 Jul 2006 13:06:10 UTC
Topic 191630
(moderation:
A new Beta Test App is available on our usual Beta Test Page.
This has some changes to the code used in the 4.16 App that apparently further speed it up, though I'm still puzzled why. These changes are only suitable for the SSE code, the generic code (on non-SSE CPUs) probably won't show an improvement over the 4.16 App.
I've seen very large variations regarding the speed compared to the 4.01 App depending on the CPU type, so if you report speed values here, please also report the CPU.
Happy tessting!
BM
BM
LINUX Beta Test App 4.17 available
)
What do we do with 4.16's files ? Delete or keep them in the ../einstein... directory ?
"Entia non sunt multiplicandam praeter necessitatem"
(OKHAM)
Leave them. Overwrite the
)
Leave them. Overwrite the app_info.xml with the new one. The tasks already assigned to 4.16 will be finished with that, new work should be assigned to the 4.17.
BM
BM
RE: Leave them. Overwrite
)
Yup got the permissions right this time and the next batch are designated for 4.17
Thanks - Gray
BTW did another WU using 4.16 and got this time: 28,137.84
Here are some timings for the
)
Here are some timings for the new version (4.17), in seconds:
9998.97
9881.78
9971.07
avg: 9950.61
Compared with my timings for the previous beta (4.16, timings here), whose avg=10225.20, there is an increase of about 2.7%, or compared with the 4.01 data I cited (avg=11596.44), there's an overall increase of 14.2%
Curiously, the claimed credit is now 19.56, as compared with 19.55 claimed for all my previously cited wus.
Bernd, you said the speed increase was mysterious. Does the (albeit small) change in credit claim shed light on that?
As I said before, I'm using a P3 coppermine, 600MHz, with sse and 256 KB cache. It's nice to get below 10000s on this machine.
Pete
Speaking of trusty old
)
Speaking of trusty old Coppermines, here's how mine has fared, comparing the last 6 units crunched with 4.01 and the first 6 crunched with 4.17.
[pre]
CPU: Genuine Intel Celeron 1GHZ (Coppermine)
OS: Debian Linux 3.1
All Workunits h1_028x range.
v4.01 v4.17
------- -------
7063.02 5991.15
6732.55 5934.10
6761.93 5910.45
6832.57 6397.05
6909.30 5933.99
7123.99 6162.89
--------- ---------
6903.89 6054.93 (Average)
14.15% increase over 4.01
[/pre]
Based on the above, this old slowboat should be able to average a couple more workunits per day (from 12 to 14).
Nice work!
Hello Using 4.17 I just
)
Hello
Using 4.17 I just got 28,092.47 on a big WU compared to a best of 33,290.42 for same size WU using 4.10 - nice, very nice ! Thankyou for the hard work.
Gray
Comparing 4.01 - 4.16 - 4.17
)
Comparing 4.01 - 4.16 - 4.17 with WUs h1_1278.5_S5R1 (credit176,31)
avg. time (sec): 29108 - 25444 - 25546
No. of results: 4 - 5 - 1
What CPUs are these? I've
)
What CPUs are these?
I've seen a bit more speedup, but I mostly tested it on Xeon and Pentium-M ...
BM
BM
I've installed 4.17 but first
)
I've installed 4.17 but first there are the results vom 4.16 to compare.
First there are no validation errors.
I've 4 Linux boxes running:
Dual Xeon 2,2GHz (SSE2 capable)
4.01 4.16
4153 3726
4156 3719
Speedup ~11.5 %
Pentium M 725 @ 2,13GHz
4.01 4.16
27325 22533
27300 22563
Speedup ~ 21 %
Celeron M 380 @ 2,4GHz
atm there ist only one compare
4.01 4.16
25938 21470
Speedup ~ 20 %
The last Linux Box doesn't have any new result. But I think the 1200er Tualatin should be around 11% faster.
RE: Comparing 4.01 - 4.16 -
)
Sorry, CPU is AMD64 2800+ (Socket 754)