Advanced search

Message boards : Number crunching : Slower crunching time for NOELIA_c36x467x WUs on Windows xp

Author Message
Bedrich Hajek
Send message
Joined: 28 Mar 09
Posts: 397
Credit: 5,136,281,601
RAC: 1,461,617
Scientific publications
Message 42072 - Posted: 1 Nov 2015 | 15:12:51 UTC

I have noticed that the NOELIA_c36x467x crunch much slower on Windows xp than 10.

Here are examples:

XP computer first.

e67s16_e55s13p1f190-NOELIA_c36x467x2-0-1-RND8431_0 11277327 31 Oct 2015 | 13:36:40 UTC 1 Nov 2015 | 6:27:15 UTC Completed and validated 34,260.04 33,656.08 180,000.00 Long runs (8-12 hours on fastest card) v8.47 (cuda65)

Notice the GERARD tasks run faster on xp, see link below.

10 computer second.

e67s21_e35s16p1f10-NOELIA_c36x467x3-0-1-RND6782_0 11277572 31 Oct 2015 | 22:00:15 UTC 1 Nov 2015 | 7:48:56 UTC Completed and validated 21,376.66 21,280.27 180,000.00 Long runs (8-12 hours on fastest card) v8.47 (cuda65)

The NOELIA are faster on 10. see link below.

At first I thought it was my old computer that was the issue, but I noticed it on other volunteer's computers as well. See link below.

I thought this was odd, so started this post. Anyone have any explanation why this is happening?

Profile Retvari Zoltan
Send message
Joined: 20 Jan 09
Posts: 2182
Credit: 15,795,015,562
RAC: 1,341,394
Scientific publications
Message 42077 - Posted: 1 Nov 2015 | 16:51:10 UTC - in response to Message 42072.

These NOELIA_c36x467x workunits use much more CPU computation (double precision FP) than the GERARD tasks, but they are not faster on Windows 10 than on Windows XP. (however the total WDDM overhead is much less for these tasks)
My Windows XPx64 + i7-4790K + GTX 980Ti + DDR3 1866MHz host can finish these in 18967~19372 seconds (5h 16m 7s - 5h 22m 52s).
GERARD_VACXCL12_LIG: 20583~20808 seconds on the same host.
But I have an experimental host:
Windows XPx64, Core2 Duo E8500 (3.16GHz), DQ45CB motherboard, 2x2GB DDR2 800MHz RAM, GTX 980Ti (ASUS Strix).
This host can finish a NOELIA_c36x467x in 25504~25583 seconds.
GERARD_VACXCL12_LIG: 21411~21520
The runtime increase of the NOELIA_c36x467x workunits on the experimental host is nearly 6 times higher than of the GERARD_VACXCL12_LIG, so it's the CPU architecture and the other tasks running limits the performance of the NOELIA_c36x467x tasks (besides the OS and the WU itself).

Post to thread

Message boards : Number crunching : Slower crunching time for NOELIA_c36x467x WUs on Windows xp