|
|
New Altivec-enhanced Seti worker in need of testing (Page 16)
|
|
|
|
Fresh-Faced Recruit
Join Date: Feb 2006
Status:
Offline
|
|
Huh, sorry - TiloProbst answered my question I think. :-)
|
|
|
|
|
|
|
|
|
Fresh-Faced Recruit
Join Date: Jan 2006
Location: Maryland
Status:
Offline
|
|
Originally Posted by TiloProbst
there is a installation guide on page 7 of this thread.
In addition to the detailed instructions on page 7, you also need to include the bigfft_wisdom file appropriate for your machine. See the post by E.T. from Tellus on the previous page to get the wisdom file for your machine. Put that file in the same setiathome project directory as the customized client and app_info.xml file that you get from unpacking the alpha-52 client zipfile.
HTH, Mark
|
|
|
|
|
|
|
|
|
Fresh-Faced Recruit
Join Date: Feb 2006
Status:
Offline
|
|
Thanks everyone for clarifying this.
I have now: 1) installed the a5 worker; 2) installed BOINC Menubar SuperBench and 3) installed the appropriate wisdom file.
I'm interested to see how this affects my results. Will post an update in a week or so when I've accumulated a few.
Kindest regards and thanks again for your patience and tolerance with us "newbies". :-)
|
|
|
|
|
|
|
|
|
Forum Regular
Join Date: Aug 2005
Location: Cupertino, CA
Status:
Offline
|
|
Originally Posted by Urban Mainframe
Thanks everyone for clarifying this.
I have now: 1) installed the a5 worker; 2) installed BOINC Menubar SuperBench and 3) installed the appropriate wisdom file.
I'm interested to see how this affects my results. Will post an update in a week or so when I've accumulated a few.
Kindest regards and thanks again for your patience and tolerance with us "newbies". :-)
You might want to make sure you're running alpha-5.2, rather than alpha-5--as usual, I'm slow updating the first post of the thread. alpha-5.2 for G5s is now linked from the first post, as is E.T's repository of bigfft_wisdom files.
Besides checking the filename in your BOINC data directory, the other way to tell which worker you're running is to look at your completed work units to see if anything was written to stderr. If you're using alpha-5.2, you'll know.
Also, if you're having second thoughts about actually doing this upgrade, do it. Trust me, it's worth the trouble.
|
|
|
|
|
|
|
|
|
jamy
|
|
Originally Posted by E.T from tellus
E.T - the links don't seem to be working to get the files.
Also, does anybody know if the amount of ram in my G5 2.0 ghz has any bearing on the big_wisdom file? I have 2 gigs of ram in mine, so I wonder if I can use one from someone that might have only 1 gig (for example)
Thanks!
|
|
|
|
|
|
|
|
|
Fresh-Faced Recruit
Join Date: Feb 2006
Status:
Offline
|
|
Originally Posted by alexkan
Also, if you're having second thoughts about actually doing this upgrade, do it. Trust me, it's worth the trouble.
No second thoughts at all Alex - I am now running alpha-5.2.
Thank you.
|
|
|
|
|
|
|
|
|
Fresh-Faced Recruit
Join Date: Jan 2006
Status:
Offline
|
|
|
|
|
|
|
|
|
|
|
Dedicated MacNNer
Join Date: Oct 2005
Location: Switzerland
Status:
Offline
|
|
|
|
|
|
|
|
|
|
|
Mac Elite
Join Date: Apr 2000
Location: Minneapolis, MN USA
Status:
Offline
|
|
Single CPU 2.0 getting higher RAC of a dual?
Get that alpha-4 machine updated! You won't regret it.
|
|
|
|
|
|
|
|
|
Dedicated MacNNer
Join Date: Oct 2005
Location: Switzerland
Status:
Offline
|
|
Originally Posted by Todd Madson
Single CPU 2.0 getting higher RAC of a dual?
Get that alpha-4 machine updated! You won't regret it.
Can't (right now) - the dual runs Panther Server and a WebObjects app that's not Tiger-compatible (yet) - and a-4 is the latest version built for Panther. But I will upgrade as soon as feasible, trust me!
|
|
|
|
|
|
|
|
|
Mac Elite
Join Date: Apr 2000
Location: Minneapolis, MN USA
Status:
Offline
|
|
My RAC keeps rising on the DP 2.5 with 5.2a - 1725 as of yesterday rising from around 850-900.
Amazing.
|
|
|
|
|
|
|
|
|
Fresh-Faced Recruit
Join Date: Feb 2006
Status:
Offline
|
|
Originally Posted by Todd Madson
My RAC keeps rising on the DP 2.5 with 5.2a - 1725 as of yesterday rising from around 850-900.
Amazing.
Blimey! What the heck am I doing wrong? My RAC is currently reading 443.73 for a DP 2.7 with 2GB RAM!
|
|
|
|
|
|
|
|
|
Forum Regular
Join Date: Aug 2005
Location: Cupertino, CA
Status:
Offline
|
|
Originally Posted by Urban Mainframe
Blimey! What the heck am I doing wrong? My RAC is currently reading 443.73 for a DP 2.7 with 2GB RAM!
RAC takes a long time to stabilize, since it's an exponential average over weeks of computation. That being said, could you post a link to the page for your machine, so we can track its progress?
|
|
|
|
|
|
|
|
|
Fresh-Faced Recruit
Join Date: Feb 2006
Status:
Offline
|
|
Originally Posted by alexkan
RAC takes a long time to stabilize, since it's an exponential average over weeks of computation. That being said, could you post a link to the page for your machine, so we can track its progress?
Okay. I'll bear that in mind. Thanks Alex.
My PM data: http://setiathome.berkeley.edu/show_...hostid=2197618
|
|
|
|
|
|
|
|
|
Dedicated MacNNer
Join Date: Oct 2005
Location: Switzerland
Status:
Offline
|
|
Originally Posted by Urban Mainframe
Hmmm... that's oddly slow for a DP 2.7 GHz G5 and alpha-5.2. Could it be that you're actually running alpha-5 with Processor Performance set to Automatic or Reduced (see Energy Saver pref pane)?
|
|
|
|
|
|
|
|
|
Fresh-Faced Recruit
Join Date: Feb 2006
Status:
Offline
|
|
Originally Posted by halimedia
Hmmm... that's oddly slow for a DP 2.7 GHz G5 and alpha-5.2. Could it be that you're actually running alpha-5 with Processor Performance set to Automatic or Reduced (see Energy Saver pref pane)?
I'm definitely running with processor performance set to "Highest". I am certain I'm running alpha-5.2 - is there any way I can confirm this?
|
|
|
|
|
|
|
|
|
Dedicated MacNNer
Join Date: Oct 2005
Location: Switzerland
Status:
Offline
|
|
(
Last edited by halimedia; Feb 23, 2006 at 04:37 AM.
)
|
|
|
|
|
|
|
|
|
Fresh-Faced Recruit
Join Date: Feb 2006
Status:
Offline
|
|
Originally Posted by halimedia
It truly looks like your're not running a-5.2. What BOINC client are you using (Manager, Menubar or CLI)?
You're right Ron. I wasn't running a-5.2 ... I am now! :-)
I'm using SuperBench Menubar as my Boinc client as this machine is running 24/7 and I think I read earlier in this thread that SuperBench Menubar was the best client.
(
Last edited by Urban Mainframe; Feb 23, 2006 at 04:53 AM.
)
|
|
|
|
|
|
|
|
|
Dedicated MacNNer
Join Date: Oct 2005
Location: Switzerland
Status:
Offline
|
|
Originally Posted by Urban Mainframe
You're right Ron. I wasn't running a-5.2 ... I am now! :-)
Great!
Edit: Do you have a wisdom file installed?
I'm using SuperBench Menubar as my Boinc client as this machine is running 24/7 and I think I read earlier in this thread that SuperBench Menubar was the best client.
It looks like you're using version 4.44 of the client. While 4.44 is very stable and has fewer bugs, I'd advise you to use 5.2.13 Superbench beta, available here: http://members.dslextreme.com/~reade...boincbeta.html
The advantage of the newer version is that your benchmarks will be higher, which results in claiming higher credit. Bugs are that it sometimes stops computing out of the blue during SETI scheduler outages, and that it doesn't show numerical progress for the second computation in the menubar menu. But other than that, all is well.
HTH,
Ron
|
|
|
|
|
|
|
|
|
Fresh-Faced Recruit
Join Date: Feb 2006
Status:
Offline
|
|
Originally Posted by halimedia
I do have the appropriate wisdom file installed.
Regarding Superbench... I'm confused, which of the following should I install:
Mac OS X 10.4.x
CLI 5.2.13 Clients.
non-superbench versions
GUI-compatible, Manager/Menubar
have G3/G4/G5 client preinstalled
BOINC Manager 2.7 M
BOINC Menubar 2.2 M
G3 / G4 / G5 1.7 M
G3 588 K
G4 588 K
G5 588 K
CLI 5.2.13 Clients.
superbench versions
GUI-compatible, Manager/Menubar
have G3/G4/G5 client preinstalled
BOINC Manager 2.7 M
BOINC Menubar 2.2 M
G3 / G4 / G5 1.7 M
G3 588 K
G4 588 K
G5 588 K
CLI 5.2.13 Clients.
superbench versions
not GUI-compatible
G3 / G4 / G5 1.7 M
Darwin x86 600 K
CLI 5.2.8 Clients.
superbench versions
G3 / G4 / G5 1.8 M
|
|
|
|
|
|
|
|
|
Fresh-Faced Recruit
Join Date: Feb 2006
Status:
Offline
|
|
Originally Posted by Urban Mainframe
Regarding Superbench... I'm confused, which of the following should I install:
No worries. I've figured it out.
Okay, let's see how it goes from here... :-)
|
|
|
|
|
|
|
|
|
Fresh-Faced Recruit
Join Date: Feb 2006
Status:
Offline
|
|
I think I've got it all sorted now.
stderr out:
<core_client_version>5.2.13</core_client_version>
<stderr_txt>
OS X Altivec-optimized vDSP/FFTW S@H application by Rick Berry and Alex Kan
version: alpha-5.2 (public release)
successfully loaded FFTW wisdom from bigfft_wisdom
FFT usage: vDSP out-of-place for FFT sizes < 2^13
FFTW out-of-place for FFT sizes >= 2^13
</stderr_txt>
CPU Time for each work-unit almost halved (from 4k+ to 2k+).
Measured floating point speed increased from: 4918.25 to 7026.97 mops
Measured integer speed increased from: 13069.09 to 22654.19 mops
Fscking Awesome!
(
Last edited by Urban Mainframe; Feb 23, 2006 at 06:51 PM.
)
|
|
|
|
|
|
|
|
|
Mac Elite
Join Date: Apr 2000
Location: Minneapolis, MN USA
Status:
Offline
|
|
You'll get an extra boost in speed if you install CHUD tools and disable NAP.
FYI.
|
|
|
|
|
|
|
|
|
Fresh-Faced Recruit
Join Date: Feb 2006
Status:
Offline
|
|
Originally Posted by Todd Madson
You'll get an extra boost in speed if you install CHUD tools and disable NAP.
FYI.
Need more information I'm afraid Todd.
|
|
|
|
|
|
|
|
|
Fresh-Faced Recruit
Join Date: Feb 2006
Status:
Offline
|
|
Originally Posted by Urban Mainframe
Need more information I'm afraid Todd.
No I don't ... Thank you Google! :-)
|
|
|
|
|
|
|
|
|
Dedicated MacNNer
Join Date: Oct 2005
Location: Switzerland
Status:
Offline
|
|
Feed your need for speed!
|
|
|
|
|
|
|
|
|
Mac Elite
Join Date: Apr 2000
Location: Minneapolis, MN USA
Status:
Offline
|
|
It does consume more power with nap disabled by about 100 watts - fyi.
At least on a DP 2.5 it does.
We used a meter called the Power Angel to detect this.
|
|
|
|
|
|
|
|
|
Fresh-Faced Recruit
Join Date: Feb 2006
Status:
Offline
|
|
Originally Posted by Todd Madson
It does consume more power with nap disabled by about 100 watts
Curious this. Disabling NAP means the processor doesn't go into low power mode (when idling) - but my computer never does anyway, since its always running a BOINC project.
So given that my processors are running at 100% 24/7, I have 2 questions:
1) Is there any real, proven benefit in disabling NAP?
2) Will I be consuming any more power with NAP disabled?
|
|
|
|
|
|
|
|
|
Mac Elite
Join Date: Apr 2000
Location: Minneapolis, MN USA
Status:
Offline
|
|
(1) I found for my own situation, yes.
(2) Yes.
|
|
|
|
|
|
|
|
|
Fresh-Faced Recruit
Join Date: Feb 2006
Status:
Offline
|
|
Okay. Thanks for the info Todd. I have had NAP disabled for a good 6 hours now - so I'll see if it makes a noticeable difference to stats. Processor temps haven't changed (some NAP documentation I read warned of a +10degC change (but I guess they weren't already saturating their processors with a highly optimised SETI worker eh?)
|
|
|
|
|
|
|
|
|
Fresh-Faced Recruit
Join Date: Feb 2006
Status:
Offline
|
|
Originally Posted by Todd Madson
It does consume more power with nap disabled by about 100 watts.
I guess I better turn a light off then eh?
|
|
|
|
|
|
|
|
|
Fresh-Faced Recruit
Join Date: Feb 2006
Status:
Offline
|
|
A quick update: My RACs are stabilising now and everything seems to be working consistently.
Thanks to all the help I have received on this forum I am currently enjoying a combined RAC of around 1500 on SETI@Home (and it's increasing daily atm) - which has taken me to position #10 on the Team MacNN leaderboard!
Brilliant. Thanks for all your help guys.
|
|
|
|
|
|
|
|
|
Forum Regular
Join Date: Sep 2005
Location: Chicago Suburbs
Status:
Offline
|
|
Urban Mainframe,
If you check out the MacNN Stats page it shows your RAC much faster as it has a shorter avarage span that the SETI's RAC. Will give you a better feel for how you are running now. Here is the link for you:
http://teamstats.macnn.com/seti2/sta...nge=50&UID=219
Has you at a (7 Day?) RAC of 2,317.34. That sounds even better for you to me. Of course not that good for the rest of us trying to keep up ;-)
(
Last edited by BTBlomberg; Feb 27, 2006 at 02:20 PM.
)
|
|
|
|
|
|
|
|
|
Forum Regular
Join Date: Oct 2003
Location: Germany
Status:
Offline
|
|
Is there any info available on how big the speedup gained by disabling the nap-mode is?
|
|
|
|
|
|
|
|
|
Fresh-Faced Recruit
Join Date: Feb 2006
Status:
Offline
|
|
I am a newbie trying to get my G4 dual 1.25 GHz to crunch better than 4-6 hours per block I am currently using seti 4.18 any help woulgd greatly be appreciated.
|
|
|
|
|
|
|
|
|
Fresh-Faced Recruit
Join Date: Jan 2006
Status:
Offline
|
|
Originally Posted by phantomac
Is there any info available on how big the speedup gained by disabling the nap-mode is?
Yeah, I'd be interested in hearing more about the purported benefits of disabling it, as well. From my understanding, it should only make a difference in idle modes where the CPU transitions into a lower power state than the automatic DOZE. When the power management has been set to Highest and the CPUs constantly loaded, I can't imagine it making a huge difference. Though I could be horribly horribly wrong.
|
|
|
|
|
|
|
|
|
Dedicated MacNNer
Join Date: Nov 2004
Location: Virginia
Status:
Offline
|
|
Originally Posted by SASHA_H
I am a newbie trying to get my G4 dual 1.25 GHz to crunch better than 4-6 hours per block I am currently using seti 4.18 any help woulgd greatly be appreciated.
Sasha: If you are running OS X 10.4.x, go to this web site and download the 5.2.13 BOINC client, and then go to this other web site and download the appropriate SETI client. As a point of reference, my PowerBook G4, OS X 10.4.x, 1.67GHz, averages around 5800 CPU seconds (roughly 1.6 hours), and my iBook G4, OS X 10.4.x, 1.33 GHz, averages around 7400 CPU seconds (roughly 2.1 hours). I'm running BOINC 5.2.13 and SETI alpha-4-G4.
beadman
|
|
|
|
|
|
|
|
|
Fresh-Faced Recruit
Join Date: Feb 2006
Status:
Offline
|
|
Originally Posted by BTBlomberg
Urban Mainframe,
If you check out the MacNN Stats page it shows your RAC much faster as it has a shorter avarage span that the SETI's RAC. Will give you a better feel for how you are running now. Here is the link for you:
http://teamstats.macnn.com/seti2/sta...nge=50&UID=219
Has you at a (7 Day?) RAC of 2,317.34. That sounds even better for you to me. Of course not that good for the rest of us trying to keep up ;-)
Holy s^&*... now this is more like it.
Now I have to get my hands on a couple of quads!
|
|
|
|
|
|
|
|
|
Fresh-Faced Recruit
Join Date: Feb 2006
Status:
Offline
|
|
Sasha: If you are running OS X 10.4.x, go to this web site and download the 5.2.13 BOINC client, and then go to this other web site and download the appropriate SETI client.
What do I do then?
|
|
|
|
|
|
|
|
|
Fresh-Faced Recruit
Join Date: Jan 2006
Location: Maryland
Status:
Offline
|
|
Originally Posted by beadman
As a point of reference, my PowerBook G4, OS X 10.4.x, 1.67GHz, averages around 5800 CPU seconds (roughly 1.6 hours), and my iBook G4, OS X 10.4.x, 1.33 GHz, averages around 7400 CPU seconds (roughly 2.1 hours). I'm running BOINC 5.2.13 and SETI alpha-4-G4.
beadman
As another point of reference, I'm averaging about 9000 CPU seconds for a 500 MHz G4 using alpha-5-g4. I've been *very* pleased with alpha-5 on my old G4. It does, however, do nothing but crunch SETI all day long while I use my G5. Personally, I use the CLI BOINC set to startup after boot on all my machines and then I log out whenever I am not actively using my computer(s).
SASHA_H: For details on how to install BOINC (using BOINC Manager / BOINC Menu) and the SETI worker, read up the past page and page 7 in particular. There is a thorough description of the install process.
|
|
|
|
|
|
|
|
|
Fresh-Faced Recruit
Join Date: Mar 2003
Location: Nutley, NJ
Status:
Offline
|
|
I've purchased a Mac Mini, duel core intel.
I've got it running the beta 5.2.13i386 client from Mnn beta page, and the Alpha G4 a5.2 worker.
I've added the bigwisdom with this, and I'm running one WU in approx 3hours 30 minutes or so.
I've reviewed all the messages here and cannot find references to a universal worker. Any idea?
Bill
|
|
|
|
|
|
|
|
|
Mac Elite
Join Date: Apr 2000
Location: Minneapolis, MN USA
Status:
Offline
|
|
I do not believe a universal worker exists as yet - I know that there is
a link at the Seti @ Home main website that discusses their intention to
create one however.
After the recent S@H outage it seems to have taken days to get back
to my usual level of production and we're still not back yet. Sigh.
|
|
|
|
|
|
|
|
|
Forum Regular
Join Date: Oct 2003
Location: Germany
Status:
Offline
|
|
BTBlomberg, how do you get so much work from the project? I never seem to get more than 3 to 6 workunits in advance.
|
|
|
|
|
|
|
|
|
Mac Elite
Join Date: Apr 2000
Location: Minneapolis, MN USA
Status:
Offline
|
|
That's another thing I was going to ask - my own machine often sits for
hours waiting for work.
I'm running the menubar boinc manager but there is no option to ask
for x blocks or the like.
The preferences menu in that software is grayed out anyway.
Ideas?
|
|
|
|
|
|
|
|
|
Dedicated MacNNer
Join Date: Sep 2004
Location: London
Status:
Offline
|
|
In the Menu Bar version you can try 'update' in the project manager, however, to know the number of wu's you have you will have to open the project folder and count them.
The amount of wu you down load is set by your prefrences in the projects 'your prefrences' page. The 'connect every 0.5 days' gets you a low amount of wu's, connect every 10 days will get you bundles. I think the exact amount is determined by your per day work rate
K.
|
|
|
|
|
|
|
|
|
Fresh-Faced Recruit
Join Date: Nov 2001
Location: Milky Way
Status:
Offline
|
|
As a side issue, there is some strategy in setting the queue length on your machines; you can take advantage of the home/work/school distinction to set different queues for different classes of machines as well.
The strategic issues include being able to ride out the temporary outages, but with longer queues, your results will lag production by several days, meaning that your work will generally not be one of the first three sent in, determining the score of the work unit. (which is the median of the first three valid reports). If your scores tend to be lower than average, this is to your advantage - you'll be awarded a higher credit than you claimed. The reverse is true when your scores tend to be higher; while you may not always be the low or median score in the first three, the probability that you might be the median score is always there, thus generating additional credit for the work unit.
Of course, if all you care about is the science, then reporting early and often allows the project to work at maximum efficiency. That implies work queues that are only long enough to ride out the typical outage to one sigma (about 1.5 days by my estimate).
|
|
|
|
|
|
|
|
|
Junior Member
Join Date: Feb 2006
Location: Paris, France, Europe, Earth, Sol
Status:
Offline
|
|
First hello all and thanks for the clients, you rock!
next a few observations on my G5 crunching
G5 1.8 mono, mk1, 1.5 GB of RAM.
BOINC 5.2.13 (the optimized superbench)
SETI client optimized alpha 5-2
I'm currently using the FFTwisdom file of TiloProbst, seems to be slightly faster than the other one posted… there are a few differences between the files btw. If I have time I'll run the script too…
right now I'm at about 321/day and going up… I'm wondering where it'll stop
Only one or two units went bad in "zillions" (I used the 5-2 for 3 weeks), they did so from the start.
On the "nap" difference, with nap checked, I crunch a unit in about 47', without nap, it's between 41 and 42'…
Not that baaaad
(
Last edited by lepetitmartien; Mar 13, 2006 at 01:14 AM.
)
|
|
|
|
|
|
|
|
|
Fresh-Faced Recruit
Join Date: Sep 2005
Status:
Offline
|
|
Originally Posted by Todd Madson
I do not believe a universal worker exists as yet - I know that there is
a link at the Seti @ Home main website that discusses their intention to
create one however.
Yeah, there will definitely be a universal binary version (with separate optimisations for G4, G5 and x86) at some point. I'll be getting started on this just as soon as my MacBook Pro arrives (hopefully by the end of the month).
The initial x86 version won't be as massively optimised as the G4 / G5 versions. Quite a bit of our code is custom Altivec stuff and I don't yet have the experience to convert this to SSE (the x86 equivalent of Altivec). Should still be pretty quick though.
It'll be many weeks before you guys see any sort of update (from me at least, dunno if Alex is similarly busy). My PowerBook keeps coughing up blood so it's not a great development environment at the moment (stubborn little bastard refuses to die though...).
|
|
|
|
|
|
|
|
|
Junior Member
Join Date: Jul 2005
Status:
Offline
|
|
Originally Posted by lepetitmartien
G5 1.8 mono, mk1, 1.5 GB of RAM.
I'm currently using the FFTwisdom file of TiloProbst, seems to be slightly faster than the other one posted… there are a few differences between the files btw. If I have time I'll run the script too…
right now I'm at about 321/day and going up… I'm wondering where it'll stop
mk1 = ?
uhm, I have 900 Mhz FSB, maybe that explains the performance differences?
I once achieved ~440 RAC with my Single 1,8 .. Can't give a more precise prediction since this machine runs maybe 14 hours a day, often not even that.
|
|
|
|
|
|
|
|
|
Mac Elite
Join Date: Apr 2000
Location: Minneapolis, MN USA
Status:
Offline
|
|
Latest update:
http://setiathome.berkeley.edu/show_...hostid=1876496
Getting almost 800 per day now on my DP2.5. RAC is 1600, was almost
2000 before the recent outage. More as things develop. My team is just
nearing 5K rac total. And we have a total of five people on it. More later.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
Forum Rules
|
|
|
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts
|
HTML code is Off
|
|
|
|
|
|
|
|
|
|
|
|