0.04, 0.05 and 0.06 (all beta) Gravitational Wave Engineering run (test apps)

Richie
Richie
Joined: 7 Mar 14
Posts: 613
Credit: 1,690,698,845
RAC: 140,071

v0.06 Linux vs Windows One

v0.06
Linux vs Windows

One dual boot machine:

host/12462839

host/12768123

5 concurrent tasks, two separate runs in both environments.

Linux Mint 19.1 Tessa (MATE, kernel 5.1-rc5)
* snippet from task ID: 339.00 Hz
avg run time per task: 16584 & 16522

Windows 10 (18875)
* 181.85 Hz
avg run time per task: 17633 & 17573

Linux is 6 % faster.

Bernd Machenschalk
Bernd Machenschalk
Moderator
Administrator
Joined: 15 Oct 04
Posts: 3,998
Credit: 208,727,832
RAC: 61,214

We try pretty hard to make

We try pretty hard to make taks of the same "size" and thus runtime, but due to e.g. rounding (some things have to be integers) we don't fully succeed in that. A 5% variation is common and considered negligible.

Also too BOINC sets CPU apps to 'nice' / 'lower' priority level, so the actual run-time of a task depends partially on what else is happening on the computer, which at least on Linux depends a lot on the individual installation and configuration of the system.

Windows and Linux Apps are built from identical source code with the same compiler (gcc) with only slight difference in compiler versions (Windows:6.3, Linux:7.0 IIRC).

I don't think there's a significant difference in speed, but thanks a lot for checking!

BM

cecht
cecht
Joined: 7 Mar 18
Posts: 807
Credit: 922,597,340
RAC: 781,263

Gary Roberts wrote:E@H is

Gary Roberts wrote:
E@H is well positioned for the much harder detection of continuous emissions.  They must be there - they just need to be found :-).... It's nice to see a high RAC but it's insignificant in comparison to the prospect of finding continuous GW.

Thank you much for putting this into perspective.  I was pondering such a switch once a final GW GPU Linux app became available, but you've made that decision easy.  Dang, and I just recently hit my goal of >1M credits/day by running FGRPB1G tasks. In the best of all possible worlds, maybe BOINC will also give high credits to GW GPU tasks!

Ideas are not fixed, nor should they be; we live in model-dependent reality.

Richie
Richie
Joined: 7 Mar 14
Posts: 613
Credit: 1,690,698,845
RAC: 140,071

Just for the universal

Just for the universal record... two more results. There was some kind of v0.07 available temporarily. Two sets of 5 tasks.. crunched by the same machine as yesterday, but Windows only (couldn't get v0.07 tasks for Linux anymore).

Process priority "low" (default):  avg per task 17248 sec
Process priority manually forced to "high": avg per task 17370 sec

Very much the same speed as v0.06.

Gary Roberts
Gary Roberts
Moderator
Joined: 9 Feb 05
Posts: 5,411
Credit: 53,773,193,494
RAC: 71,283,657

Version 0.06 is no longer a

Version 0.06 is no longer a test CPU app.  It's now showing on the apps page without the beta tag.  I didn't see a 0.07 and there isn't one showing currently.  It must have been deprecated pretty quickly.  I guess they may use the 0.06 results in order to validate against whatever shows as the next test app, CPU or GPU.

There is a 0.12 GPU test app.  According to comments starting here, this new test app is already causing problems so likely to be deprecated quickly.

Edit:  There is also this thread on the Problems board.

 

Cheers,
Gary.

Bernd Machenschalk
Bernd Machenschalk
Moderator
Administrator
Joined: 15 Oct 04
Posts: 3,998
Credit: 208,727,832
RAC: 61,214

Sorry, I'm shuffelling around

Sorry, I'm shuffelling around things a bit faster than I can report. There was a 0.07 beta app version, and there is now again. I'm still trying to find the minimal sufficient table size (64k as in 0.06 seems sufficient, 16k as in 0.05 is too small, trying 32k in 0.07).

BM

Gary Roberts
Gary Roberts
Moderator
Joined: 9 Feb 05
Posts: 5,411
Credit: 53,773,193,494
RAC: 71,283,657

OK, no worries.  Just checked

OK, no worries.  Just checked the apps page again and 0.06 is gone completely and there is a 0.07 beta.  To help get results back quickly, I'll abort any unstarted 0.06 and get some 0.07 to replace them.

 

Cheers,
Gary.

Richie
Richie
Joined: 7 Mar 14
Posts: 613
Credit: 1,690,698,845
RAC: 140,071

v0.07 Two dual boot

v0.07

Two dual boot computers run 5 concurrent tasks under Windows and Linux.
Average run time per task (seconds):

Computer #1 (host/12768123 & host/12462839)
Windows 10 (18875) : 17299
Linux Mint 19.1 : 15855

Computer #2 (host/12331989 & host/12774162)
Windows 10 (18362) : 17100
Ubuntu 18.10 : 15801

Mad_Max
Mad_Max
Joined: 2 Jan 10
Posts: 147
Credit: 1,810,667,857
RAC: 303,795

In the app versions 0.04-0.06

In the app versions 0.04-0.06 part of the computation is sped up by using a "lookup table" for some usually slow computation (sinc()). This is less precise, but much faster. The difference between these app versions is the size of the table; we're looking for the size that gives enough precision for reliable validation. Speed between these App versions shouldn't vary at all, there's no difference in the code.

A doubling of speed is indeed pretty impressive, what I got from averaging over all hosts that ran both is only 20-25%. There may be outliers, we issued these app versions as "Beta Test" only due to the probably poor validation.

Even without difference in the code size of the lookup table can affect speed/runtimes by itself - via CPU cache efficiency based on how size of this often used table correspond to sizes of particular CPU caches.

I also see about 50% speed up for v.06 vs v.03 on AMD FX-8320 CPU (~24500 sec vs ~36800 sec of CPU time used per WU averaged on few WUs each)

Zalster
Zalster
Joined: 26 Nov 13
Posts: 3,063
Credit: 3,362,824,762
RAC: 270

So up until this morning I

So up until this morning I still had some V0.13 running. This morning I found the computer cool and idle. Checked the cache and no work units on it but the server seems to think I have 2 in progress.  Tried to update and no new work units have been sent.  Anyone still getting these work units or did they pull the plug on all of them them?

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.