Welcome to the MacNN Forums.

If this is your first visit, be sure to check out the FAQ by clicking the link above. You may have to register before you can post: click the register link above to proceed. To start viewing messages, select the forum that you want to visit from the selection below.

You are here: MacNN Forums > Community > Team MacNN > Enhanced Optimized

Enhanced Optimized (Page 9)
Thread Tools
Knightrider
Dedicated MacNNer
Join Date: Sep 2004
Location: London
Status: Offline
Reply With Quote
Jul 17, 2006, 08:03 PM
 
Originally Posted by kennet
I downloaded the V6ng.
Put all the files into the folder:-

/Library/Applications Support/BOINC Data/projects/setiathome.berkeley.edu

The current running wu will have to finish before the new worker takes over.

K.
     
kennet
Fresh-Faced Recruit
Join Date: Jul 2006
Status: Offline
Reply With Quote
Jul 18, 2006, 04:45 PM
 
Doug and Knightrider:

Thanks for the info. I placed ALL of the files from the download into
/Library/ApplicationsSupport/BOINCData/projects/setiathome.berkeley.edu and have restarted the computer.

I have one work unit left to do (eight hours) and then the new worker should take over, and I will let you know if
there are any problems or unexpected progress.

Thanks

Kenn
( Last edited by kennet; Jul 18, 2006 at 05:04 PM. )
[FONT="Book Antiqua"]What is left unsaid: is neither heard nor heeded.[/FONT]
     
kennet
Fresh-Faced Recruit
Join Date: Jul 2006
Status: Offline
Reply With Quote
Jul 18, 2006, 05:05 PM
 
Doug;

I have the newest official app installed.

Kenn
[FONT="Book Antiqua"]What is left unsaid: is neither heard nor heeded.[/FONT]
     
kennet
Fresh-Faced Recruit
Join Date: Jul 2006
Status: Offline
Reply With Quote
Jul 19, 2006, 01:00 PM
 
Thanks for the input and thanks to Alex for his programming skills.

Before: 51,311.01 seconds for a credit of 62.52

After: 22,762.92 seconds for a credit of 62.39

Kenn
[FONT="Book Antiqua"]What is left unsaid: is neither heard nor heeded.[/FONT]
     
Odysseus
Guest
Status:
Reply With Quote
Jul 21, 2006, 06:50 PM
 
Originally Posted by Odysseus
If it behaves like a project without a screensaver of its own, you'd get a simple, generic "BOINCsaver" display, horizontally scrolling text IIRC, and maybe the BOINC logo.
Confirmed for seti_enhanced-ppc-v6-g4-nographics, except there's no BOINC logo; the scrolling text says "Project does not support screensaver graphics: SETI@home ##.#%".
     
bobpalmer
Fresh-Faced Recruit
Join Date: Jun 2006
Location: San Diego, CA
Status: Offline
Reply With Quote
Jul 24, 2006, 11:36 PM
 
My Quad is back from getting one of its CPU's replaced. Now all the temps are normal, and the machine is running well. In under 2 weeks it's back within the SETI top 10 hosts based on RAC. v6ng has really sped things up. Between the bad CPU and switching from v6 to v6ng, I'm seeing about a 30% reduction (10Ksec down to 7Ksec). Normally the Quad is completing WU's 3 to 5 times faster than the other hosts on the same WU. Good work Alex
     
halimedia
Dedicated MacNNer
Join Date: Oct 2005
Location: Switzerland
Status: Offline
Reply With Quote
Jul 27, 2006, 06:16 AM
 
Nice to see you back in the ranks, Bob! Also nice to see only three Windoze-boxen in the top 20!
     
alexkan
Forum Regular
Join Date: Aug 2005
Location: Cupertino, CA
Status: Offline
Reply With Quote
Jul 29, 2006, 11:15 PM
 
I am in the process of testing an optimized cruncher for Intel Macs. It uses SSE3, so no guarantees that it'll run on Developer Transition Kits, but it will run on every Intel Mac that Apple has shipped so far.

Stay tuned... (if you're interested in testing it, drop me a PM)
     
Elphidieus
Fresh-Faced Recruit
Join Date: Jun 2006
Status: Offline
Reply With Quote
Jul 30, 2006, 12:53 AM
 
Originally Posted by alexkan
I am in the process of testing an optimized cruncher for Intel Macs. It uses SSE3, so no guarantees that it'll run on Developer Transition Kits, but it will run on every Intel Mac that Apple has shipped so far.

Stay tuned... (if you're interested in testing it, drop me a PM)
I sense the death of PowerPC-optimised cruncher already...

With this coincide with the upcoming Mac Pro release, I bet there'll be a mass-selling of Quad G5s around....

drop me a PM if you guys wanna throw out your powerhouses....
     
Karl Schimanek
Junior Member
Join Date: Oct 2004
Location: Germany
Status: Offline
Reply With Quote
Jul 30, 2006, 08:40 AM
 
PPC optimizations has ended so far?

Regards
Karl
     
alexkan
Forum Regular
Join Date: Aug 2005
Location: Cupertino, CA
Status: Offline
Reply With Quote
Jul 30, 2006, 10:36 AM
 
Originally Posted by Karl Schimanek
PPC optimizations has ended so far?
Oh, I've still got a few things up my sleeve for PowerPC. Can't promise how much they'll help, though...
     
Todd Madson
Mac Elite
Join Date: Apr 2000
Location: Minneapolis, MN USA
Status: Offline
Reply With Quote
Aug 1, 2006, 05:40 PM
 
If all my machines did as well as my G5 with Alex's optimizations I'd be a
happy camper. Sadly, I only have one G5.

Sigh.
     
reader50
Administrator
Join Date: Jun 2000
Location: California
Status: Offline
Reply With Quote
Aug 1, 2006, 06:12 PM
 
Originally Posted by Todd Madson
If all my machines did as well as my G5 with Alex's optimizations I'd be a
happy camper. Sadly, I only have one G5.

Sigh.
A problem that is easily fixed with large sums of money.
     
arkayn
Dedicated MacNNer
Join Date: Aug 2005
Location: Golden Valley, AZ
Status: Offline
Reply With Quote
Aug 1, 2006, 09:29 PM
 
Care to loan us some!!
     
bobpalmer
Fresh-Faced Recruit
Join Date: Jun 2006
Location: San Diego, CA
Status: Offline
Reply With Quote
Aug 1, 2006, 11:26 PM
 
( Last edited by bobpalmer; Aug 2, 2006 at 12:46 AM. )
     
halimedia
Dedicated MacNNer
Join Date: Oct 2005
Location: Switzerland
Status: Offline
Reply With Quote
Aug 2, 2006, 02:33 AM
 
Seems like something's amiss on the server side at SETI. I've seen many validate errors lately, too. Also, although some of my WUs reported have received credit, my account total credit has not increased in almost 12 hours. I guess these problems will be resolved in time...

HTH,

Ron
     
halimedia
Dedicated MacNNer
Join Date: Oct 2005
Location: Switzerland
Status: Offline
Reply With Quote
Aug 2, 2006, 02:40 AM
 
Further examination reveals that no results have been reported by any of my machines since 21:30 UTC yesterday, while WU downloads are merrily taking place. BOINC client log checks show that result uploads are failing. Is anyone else seeing this? Looks to me like something's gone wrong during yesterday's outage...
( Last edited by halimedia; Aug 2, 2006 at 06:37 AM. )
     
zombie67
Senior User
Join Date: Jun 2006
Location: Dublin, CA
Status: Offline
Reply With Quote
Aug 2, 2006, 03:01 AM
 
Originally Posted by reader50
A problem that is easily fixed with large sums of money.
I hesitated to suggest this here but....

(cough) dell refurbs (cough)

It takes a diligent scanning of their offerings, but I've bought several dual core boxes for *well* under $400 each, far below what you could build one for (and I tried!). They almost always pop up in the weee hours of the morning.

The only downside is that they come with no monitor. The upside is that what a KVM is a lot cheaper and smaller than multiple monitors + their power consumption. A buddy of mine was ditching an old Pentium (pre-number) with 16mb ram and a 2gb HD. I stole the monitor to use for my KVM.

You can get there on the cheap if you try.
     
alexkan
Forum Regular
Join Date: Aug 2005
Location: Cupertino, CA
Status: Offline
Reply With Quote
Aug 3, 2006, 11:52 PM
 
Alright, I'm done testing what I've got to make sure it's solid. (What an inopportune time for the SETI servers to start having validation issues!)

Core Duo v6.1
Core Duo v6.1 (no graphics)

This is by no means as fast as it could be, owing to the lack of Intel compiler/libraries on my part. For that matter, it's mostly a straightforward conversion of what I already had in v6 to use SSEn instead of Altivec. I'll be working on v7 in the next couple days, while debating whether or not that Intel software is worth the price of entry, but I figured that I might as well put out a release now.

Have fun, and post links to your results!
( Last edited by alexkan; Aug 4, 2006 at 12:05 AM. )
     
Bad to the bone
Fresh-Faced Recruit
Join Date: Aug 2006
Status: Offline
Reply With Quote
Aug 4, 2006, 03:14 PM
 
Originally Posted by alexkan
Alright, I'm done testing what I've got to make sure it's solid. (What an inopportune time for the SETI servers to start having validation issues!)

Core Duo v6.1
Core Duo v6.1 (no graphics)

This is by no means as fast as it could be, owing to the lack of Intel compiler/libraries on my part. For that matter, it's mostly a straightforward conversion of what I already had in v6 to use SSEn instead of Altivec. I'll be working on v7 in the next couple days, while debating whether or not that Intel software is worth the price of entry, but I figured that I might as well put out a release now.

Have fun, and post links to your results!
Thank you VERY much for this release, Alex!

The first few results retuned after the project's downtime indicate slightly more than double the speed of the nographics version over the stock worker.

Here's the link to my MBP:

http://setiathome.berkeley.edu/resul...hostid=2462923
     
beadman
Dedicated MacNNer
Join Date: Nov 2004
Location: Virginia
Status: Offline
Reply With Quote
Aug 5, 2006, 10:03 AM
 
Alex:
Here are some results for you for comparison on my 2.16 GHz MBP.

Standard issue SETI: 192 cpusec per credit
Your 6.1 optimization: 148 cpusec per credit
Your 6.1 nographics: 124 cpusec per credit

These all compare to 117 cpusec per credit running the standard Einstein.

My older 1.67 GHz PowerBook, running your latest G4 optimization, is 253 cpusec per credit on SETI and 278 cpusec per credit running Einstein

beadman
     
Lauger
Fresh-Faced Recruit
Join Date: Nov 2005
Status: Offline
Reply With Quote
Aug 5, 2006, 03:17 PM
 
Originally Posted by alexkan
Alright, I'm done testing what I've got to make sure it's solid. (What an inopportune time for the SETI servers to start having validation issues!)

Core Duo v6.1
Core Duo v6.1 (no graphics)

This is by no means as fast as it could be, owing to the lack of Intel compiler/libraries on my part. For that matter, it's mostly a straightforward conversion of what I already had in v6 to use SSEn instead of Altivec. I'll be working on v7 in the next couple days, while debating whether or not that Intel software is worth the price of entry, but I figured that I might as well put out a release now.

Have fun, and post links to your results!
Works great on my MBP. Keeps the processors at a nice toasty 180 deg F

http://setiathome.berkeley.edu/show_...hostid=2544256
     
beadman
Dedicated MacNNer
Join Date: Nov 2004
Location: Virginia
Status: Offline
Reply With Quote
Aug 5, 2006, 08:31 PM
 
Lauger:
What type of MBP do you have? I see your host ID saying it's a MBP 1,2. All your parameters seem to be the same as mine, from what I see, but mine is called a MBP 1,1 and it's two weeks old - it's a 2.16GHz 15.x model. Our times for completion seem to be similar, memory is the same, flops are basically the same,...

beadman
     
Lauger
Fresh-Faced Recruit
Join Date: Nov 2005
Status: Offline
Reply With Quote
Aug 6, 2006, 11:57 AM
 
Originally Posted by beadman
Lauger:
What type of MBP do you have? I see your host ID saying it's a MBP 1,2. All your parameters seem to be the same as mine, from what I see, but mine is called a MBP 1,1 and it's two weeks old - it's a 2.16GHz 15.x model. Our times for completion seem to be similar, memory is the same, flops are basically the same,...

beadman
It's a 17 inch, Glossy Screen, 2.16 GHz, 2GB RAM. Shows MacBookPro1,2 in System Profile. Entering the S/N in at http://www.chipmunk.nl/klantenservice/applemodel.html shows a build date of July 2006.

Are you getting similar CPU temperatures? I'm not running full time until the temperature vs life story becomes more clear.
     
Elphidieus
Fresh-Faced Recruit
Join Date: Jun 2006
Status: Offline
Reply With Quote
Aug 6, 2006, 12:54 PM
 
That's one of the drawbacks of Intel processors, you can't control how much processor resources you wanna dedicate to your apps, unlike the G5s....

Poor Intel-ified Mac users lose control of their processor power management....

Hope for miracles on the Meroms...
     
beadman
Dedicated MacNNer
Join Date: Nov 2004
Location: Virginia
Status: Offline
Reply With Quote
Aug 6, 2006, 09:13 PM
 
Lauger:
I consistantly run 88C on one processor and 85C on the other. Occasional spikes up to 91 happen, but it apparently throttles back for a minute or so until the processor gets down to around 80C, and then it goes back to high speed. I do keep the laptop elevated on a couple of sticks so the air can flow around the bottom to aid in cooling. I also go to black screen after five minutes and don't run any screen savers. I have my BOINC prefs set to only write to HD every 15 minutes, and the MBP's prefs say to put the HD to sleep when possible.

I thought your MBP1,2 might be the 17inch model, but wasn't sure so I asked. Mine is the 15 ind model.

beadman
     
zombie67
Senior User
Join Date: Jun 2006
Location: Dublin, CA
Status: Offline
Reply With Quote
Aug 8, 2006, 01:18 AM
 
Originally Posted by alexkan
Alright, I'm done testing what I've got to make sure it's solid. (What an inopportune time for the SETI servers to start having validation issues!)

Core Duo v6.1
Core Duo v6.1 (no graphics)

This is by no means as fast as it could be, owing to the lack of Intel compiler/libraries on my part. For that matter, it's mostly a straightforward conversion of what I already had in v6 to use SSEn instead of Altivec. I'll be working on v7 in the next couple days, while debating whether or not that Intel software is worth the price of entry, but I figured that I might as well put out a release now.

Have fun, and post links to your results!
Thanks again Alex! What about the wisdom thing? Will it work with intel? If so, will it help?

TIA
     
alexkan
Forum Regular
Join Date: Aug 2005
Location: Cupertino, CA
Status: Offline
Reply With Quote
Aug 8, 2006, 02:40 AM
 
Originally Posted by zombie67
Thanks again Alex! What about the wisdom thing? Will it work with intel? If so, will it help?
I don't recall if I've released a Universal version of fft_test, but it doesn't help much and it's not really worth the trouble (since you'd have to edit app_info.xml to get SETI to actually use your pregenerated wisdom).

Anyone going to be at WWDC tomorrow? If I get a chance, I'm thinking of taking some Intel SETI clients for a spin on the new Mac Pros to see how they perform.
     
Knightrider
Dedicated MacNNer
Join Date: Sep 2004
Location: London
Status: Offline
Reply With Quote
Aug 8, 2006, 07:52 AM
 
Originally Posted by alexkan
Anyone going to be at WWDC tomorrow? If I get a chance, I'm thinking of taking some Intel SETI clients for a spin on the new Mac Pros to see how they perform.
At these prices, plus Flights from London plus Hotel plus expenses - I would love to be able to say yes. As it is, you will have to be our representative, so a report would be appreciated.

Enjoy.

K.
     
jfosback
Fresh-Faced Recruit
Join Date: Jul 2006
Status: Offline
Reply With Quote
Aug 8, 2006, 05:06 PM
 
Originally Posted by alexkan
Alright, I'm done testing what I've got to make sure it's solid. (What an inopportune time for the SETI servers to start having validation issues!)

Core Duo v6.1
Core Duo v6.1 (no graphics)

This is by no means as fast as it could be, owing to the lack of Intel compiler/libraries on my part. For that matter, it's mostly a straightforward conversion of what I already had in v6 to use SSEn instead of Altivec. I'll be working on v7 in the next couple days, while debating whether or not that Intel software is worth the price of entry, but I figured that I might as well put out a release now.

Have fun, and post links to your results!
Oooh, the new optimized apps run *very* well on my MacBook Pro 2GHz. I seem to be crunching the 67 credit WUs in 8,600 cpuseconds, or around 128 cpuseconds per credit:

MacBook Pro

Thanks a ton Alex!

BTW, my machine runs consistently between 81-86C on both cores.

-jason
     
Todd Madson
Mac Elite
Join Date: Apr 2000
Location: Minneapolis, MN USA
Status: Offline
Reply With Quote
Aug 8, 2006, 10:49 PM
 
Be real interesting to see what the hopped up MacPros does in terms of workunit
performance.
     
alexkan
Forum Regular
Join Date: Aug 2005
Location: Cupertino, CA
Status: Offline
Reply With Quote
Aug 8, 2006, 11:52 PM
 
Unfortunately, I wasn't on the Mac Pros for long enough each time to test performance on long WUs, but from a preliminary examination, these machines are fast! (Incidentally, the Mac Pros also compiled the Intel build of my app what seems like an order of magnitude faster than my PowerBook normally does.) If want some numbers to look forward to, look for one of msattler's posts on the SETI boards and take a look at his overclocked Conroe's times...then knock off up to another 15% for the 3 GHz Mac Pro, based on the short WUs I've tried it with.

Speaking of compiling, I had the brief opportunity to try compiling the client with the Intel C Compiler and IPP. The only executables I produced were dynamically linked, but showed yet another performance increase, so perhaps you'll have that to look forward to in the future as well.
     
alexkan
Forum Regular
Join Date: Aug 2005
Location: Cupertino, CA
Status: Offline
Reply With Quote
Aug 10, 2006, 03:28 AM
 
Soooo...is anyone thinking of getting a Mac Pro anytime in the near future?
     
Knightrider
Dedicated MacNNer
Join Date: Sep 2004
Location: London
Status: Offline
Reply With Quote
Aug 10, 2006, 04:31 AM
 
Originally Posted by alexkan
Soooo...is anyone thinking of getting a Mac Pro anytime in the near future?
Thinking about it - but I would have to sell my quad first.

K.
     
gorbag
Fresh-Faced Recruit
Join Date: Nov 2001
Location: Milky Way
Status: Offline
Reply With Quote
Aug 10, 2006, 01:50 PM
 
Originally Posted by alexkan
Soooo...is anyone thinking of getting a Mac Pro anytime in the near future?
Near future, yes. Immediate future, no.
     
halimedia
Dedicated MacNNer
Join Date: Oct 2005
Location: Switzerland
Status: Offline
Reply With Quote
Aug 10, 2006, 05:12 PM
 
Originally Posted by alexkan
Soooo...is anyone thinking of getting a Mac Pro anytime in the near future?
Thinking - yes, but that's where it will stay for the time being. My Quad still has to pay for itself a little while longer That said, I can't wait to see how the Mac Pros will stack up against the Quads in SETI-crunching...
     
arkayn
Dedicated MacNNer
Join Date: Aug 2005
Location: Golden Valley, AZ
Status: Offline
Reply With Quote
Aug 10, 2006, 06:10 PM
 
I would love to get a Mac Pro, but first I would have to win the lottery!!
     
Gecko_r7
Forum Regular
Join Date: Oct 2005
Location: Las Vegas, NV
Status: Offline
Reply With Quote
Aug 10, 2006, 11:19 PM
 
Originally Posted by alexkan
Soooo...is anyone thinking of getting a Mac Pro anytime in the near future?
I have recently been pricing parts to build a Dual Xeon 5130 to replace my G4 as our main household computer. Prices for 2 CPUs and a Tyan MB @ $1000 + memory, HD, PS etc. Apple rather surprised me w/ the relatively affordable price for the Dual 3.0GHZ. I'd expected it to be more. IMO, it's a much better buy than an Extreme Conroe platform.

So my quandry is whether to "go cheap" and build the Dual 5130 Wintel box, or go all the way for 3GHZ. I prefer OSX, but the Pro 2GHZ isn't that good of a buy at $2199.00 (compared to building similar, or an OC-friendly Conroe).
I'm open for suggestions, but I think this group may be a little biased.

And another important consideration...what if one day a Mrs. Kan or Alex Jr. enters the scene and your spare time for optimizing goes the way of the Dodo? Who will keep the flame burning for OSX revisions?

On a more serious note, do we know how Windows and OSX compare w/ S@H on same speed Core CPUs? Is there a significant speed advatage between one OS or the other?
( Last edited by Gecko_r7; Aug 10, 2006 at 11:32 PM. )
     
bobpalmer
Fresh-Faced Recruit
Join Date: Jun 2006
Location: San Diego, CA
Status: Offline
Reply With Quote
Aug 11, 2006, 12:12 AM
 
Right now 9 of the top 11 hosts, and 12 of the top 20 hosts are PowerMac11,2 running Darwin 8.7.0. If the new Mac Pro 3.0GHz Quad is really 1.8x faster than the PowerMac Quad, then our days at the top of the hosts leader board are numbered. At least the OS will still be Darwin.
     
Todd Madson
Mac Elite
Join Date: Apr 2000
Location: Minneapolis, MN USA
Status: Offline
Reply With Quote
Aug 11, 2006, 11:39 AM
 
Yep, I'm basically still a long way off from paying for my G5 2.5 dual. Maybe when the
fourth or fifth generation of this particular model comes out I might be able to swing it.

Sigh.
     
alexkan
Forum Regular
Join Date: Aug 2005
Location: Cupertino, CA
Status: Offline
Reply With Quote
Aug 13, 2006, 01:42 AM
 
Originally Posted by Gecko_r7
I have recently been pricing parts to build a Dual Xeon 5130 to replace my G4 as our main household computer. Prices for 2 CPUs and a Tyan MB @ $1000 + memory, HD, PS etc. Apple rather surprised me w/ the relatively affordable price for the Dual 3.0GHZ. I'd expected it to be more. IMO, it's a much better buy than an Extreme Conroe platform.

So my quandry is whether to "go cheap" and build the Dual 5130 Wintel box, or go all the way for 3GHZ. I prefer OSX, but the Pro 2GHZ isn't that good of a buy at $2199.00 (compared to building similar, or an OC-friendly Conroe).
I'm open for suggestions, but I think this group may be a little biased.
From what I've seen, the 2.66 GHz Mac Pro sounds like your best bet. Going down 667 MHz to save $300 and paying $800 for 333 MHz both sound like bad deals. On another note, it does seem like a good idea to have at least 4 FB-DIMMs installed, since you want to make sure all channels are in use to take advantage of the new architecture's greater memory bandwidth.

Then again, I'm not clear on what your budget is for this computer-building/buying project, so perhaps telling you to pay more for a system is not the best option.
And another important consideration...what if one day a Mrs. Kan or Alex Jr. enters the scene and your spare time for optimizing goes the way of the Dodo? Who will keep the flame burning for OSX revisions?
I don't think that's going to happen anytime soon, guys. I'm a 22-year-old new college graduate who's just entered the workforce, so I don't think either of those things are in the cards for me just yet.

As far as the more distant future goes, now that Apple has gone Intel, people ought to be able to do what's going on in Windows/Linux land--namely, use the Intel compiler and performance libraries to optimize SETI. There's a limit to how much can be done automatically, but hopefully you won't have to worry about that for a while. I enjoy doing this too much sometimes, I think.
On a more serious note, do we know how Windows and OSX compare w/ S@H on same speed Core CPUs? Is there a significant speed advatage between one OS or the other?
Sadly, since I didn't run full-length WUs on the Mac Pros, I don't have any real-world performance numbers. (That's where you all come in!) However, if you'd like a comparison of performance on Core Duos, I present the following three sets of results:
  • a 1.66 GHz laptop, running the most recent optimized Windows client, belonging to Demi from the official SETI boards
  • a 1.66 GHz Mac mini, running v6.1, belonging to njmaugbill from these forums
  • a 1.66 GHz Mac mini, running a testing (but nowhere near final) version of v7, belonging to halimedia from these forums
Note that these may or may not bear any similarities to performance numbers on Core 2 chips, but still...
     
Gecko_r7
Forum Regular
Join Date: Oct 2005
Location: Las Vegas, NV
Status: Offline
Reply With Quote
Aug 13, 2006, 03:51 PM
 
Originally Posted by alexkan
Sadly, since I didn't run full-length WUs on the Mac Pros, I don't have any real-world performance numbers. (That's where you all come in!) However, if you'd like a comparison of performance on Core Duos, I present the following three sets of results:
  • a 1.66 GHz laptop, running the most recent optimized Windows client, belonging to Demi from the official SETI boards
  • a 1.66 GHz Mac mini, running v6.1, belonging to njmaugbill from these forums
  • a 1.66 GHz Mac mini, running a testing (but nowhere near final) version of v7, belonging to halimedia from these forums
Note that these may or may not bear any similarities to performance numbers on Core 2 chips, but still...
Thanks Alex! You have an amazing talent that's well beyond your years. There's no doubt you have a brilliant career ahead. I'll likely build a dual 5130 for the savings and easy upgrade to Clovertown (CPU swap, bios update) next year. By chance, are you planning to do a Windows compile of V7 at some point later on for Core 2?

Since I didn't find an absolutely identical WU common to all three, I ran the hosts through AndyK's cool status calculator since all things being equal, it's probably the best way to compare the aggregate average of WUs for each host against each other.

http://www.andyk.de/boinc/pc.php?pid...u=on&hsort=not

WOW! You can certainly see the difference that all your hand-assembly/optimizing makes versus compiler auto-vectorizing in the Windows version. No surprise that Halimedia is the fastest w/ v7, but just barely quicker than 6.1 at this point.

Switching topics slightly....

The company I work for, we sometimes "farm-out" business studies to a local University where these studies are assigned as an MBA project to be researched and solutions/recommendations presented by an MBA student group. Good for them as a learning exercise, great for us in practicality as it efficiently expands our resource base and brings-in a fresh perspective.

I was thinking something similar could offer a comparable benefit from a mathematical stand-point if 4 or 5 of the most mathematically intensive tasks re: various functions/analysis algorithims could be presented w/ the goal of finding the most efficient mathematical solutions & implementations that could easily be x-ferred to programming. Would this idea be worth pursuing? If so, your profile results have probably identified some likely candidate areas that a mathematician could likely improve upon.

Surely Seti has enough recognition and academic respect where it would present a credible subject & project for a graduate-level, aspiring mathematician? The real trick might be in aligning an individual's mathematical knowledge w/ prerequisite fundamentals supporting the kind of signal processing/analysis that S@H performs.

With a clear set of objectives and deliverables, I'm sure a few folks have the necessary contacts at respective universities where this could be possible. I may be able to accomplish this at my former Alma Mater.
Could even be interesting to see differing solutions to the same problems.
Just an idea if anyone thinks there's any merit to it.
Cheers!
( Last edited by Gecko_r7; Aug 13, 2006 at 10:39 PM. )
     
alexkan
Forum Regular
Join Date: Aug 2005
Location: Cupertino, CA
Status: Offline
Reply With Quote
Aug 14, 2006, 03:18 AM
 
Originally Posted by Gecko_r7
Since I didn't find an absolutely identical WU common to all three, I ran the hosts through AndyK's cool status calculator since all things being equal, it's probably the best way to compare the aggregate average of WUs for each host against each other.

WOW!  You can certainly see the difference that all your hand-assembly/optimizing makes versus compiler auto-vectorizing in the Windows version.  No surprise that Halimedia is the fastest w/ v7, but just barely quicker than 6.1 at this point.
Great! It's good to see confirmation that what I'm working on for v7 is making a difference. It looks like I still have some tuning to do, though, so don't expect me to release v7 tomorrow.
The company I work for, we sometimes "farm-out" business studies to a local University where these studies are assigned as an MBA project to be researched and presented by a MBA student group.  Good for them as a learning exercise, great for us in practicality as it efficiently expands our resource base.

I was thinking something similar could offer a comparable benefit from a mathematical stand-point if 4 or 5 of the most mathematically intensive tasks re: various functions/analysis algorithims could be presented w/ the goal of finding the most efficient mathematical solutions & implementations that could easily be x-ferred to programming. Would this idea be worth pursuing? If so, your profile results have probably identified some likely candidate areas that a mathematician could likely improve upon.

Surely Seti has enough recognition and academic respect where it would present a credible subject & project for a graduate-level, aspiring mathematician? The real trick might be in aligning an individual's mathematical knowledge w/ prerequisite fundamentals supporting the kind of signal processing/analysis that S@H performs.

With a clear set of objectives and deliverables, I'm sure a few folks have the necessary contacts at respective universities where this could be possible. I may be able to accomplish this at my former Alma Mater.
Could even be interesting to see differing solutions to the same problems.
From what I've seen...yes and no. Based on my own academic background, an undergraduate signals and systems course and a DSP course are all you'd need to understand what SETI is doing, and in the case of Gaussian-finding, substitute a more efficient implementation. Then again, there are other things that don't seem to pop up as often, like the fast folding algorithm used for pulse-finding. Perhaps some questions might be interesting to pose to graduate students, but I think a lot of simpler optimization questions can be left to undergraduates with too much free time on their hands.

On a related note, I've heard the line on the main SETI forums about GPUs not being accurate enough for SETI computation to want to plug GPUFFTW into SETI just to prove them wrong. The only problem is that the authors don't explicitly mention OS X support, and if I thought remotely testing/debugging Intel binaries was hard, getting this working can't be any easier. Also, not everyone is going to have the kinds of graphics cards that will actually be useful for this.
     
halimedia
Dedicated MacNNer
Join Date: Oct 2005
Location: Switzerland
Status: Offline
Reply With Quote
Aug 14, 2006, 03:25 AM
 
Originally Posted by alexkan
On a related note, I've heard the line on the main SETI forums about GPUs not being accurate enough for SETI computation to want to plug GPUFFTW into SETI just to prove them wrong. The only problem is that the authors don't explicitly mention OS X support, and if I thought remotely testing/debugging Intel binaries was hard, getting this working can't be any easier. Also, not everyone is going to have the kinds of graphics cards that will actually be useful for this.
Thanks once again for all your efforts, Alex! If you're interested in exploring GPUFFTW, I'd be more than willing to lend you a hand - given that PPC-Macs could be used for testing. I could offer a variety of GPUs from Rage 128 to GeForce 7800GT...

FWIW,

Ron
     
Knightrider
Dedicated MacNNer
Join Date: Sep 2004
Location: London
Status: Offline
Reply With Quote
Aug 14, 2006, 09:01 AM
 
Originally Posted by alexkan
Core Duo v6.1
Core Duo v6.1 (no graphics)

it's mostly a straightforward conversion of what I already had in v6 to use SSEn instead of Altivec. I'll be working on v7 in the next couple days,
Hi Alex,

Can (could) you get this to work on a windows/intel pc ?

K.
     
QSilver
Dedicated MacNNer
Join Date: Jun 2006
Location: Chicago
Status: Offline
Reply With Quote
Aug 14, 2006, 09:46 AM
 
Originally Posted by Knightrider
Can (could) you get this to work on a windows/intel pc ?
Knightrider, check out this thread on the S@H Number Crunching board.

QS
     
halimedia
Dedicated MacNNer
Join Date: Oct 2005
Location: Switzerland
Status: Offline
Reply With Quote
Aug 15, 2006, 08:02 AM
 
Just found the first Mac Pro that entered the top-1000 list. A RAC of over 600 after less than one week of crunching - impressive!

http://setiweb.ssl.berkeley.edu/show...hostid=2599334
http://setiweb.ssl.berkeley.edu/resu...hostid=2599334

It's running stock workers as of this writing. Shame it's an anonymous machine - otherwise we might be able to alert the owner to the optimized workers...
( Last edited by halimedia; Aug 15, 2006 at 08:17 AM. )
     
halimedia
Dedicated MacNNer
Join Date: Oct 2005
Location: Switzerland
Status: Offline
Reply With Quote
Aug 15, 2006, 08:12 AM
 
The above Mac Pro with stock workers crunches a 67 credit WU in roughly 7700 seconds. This is just a touch slower than a G5 Quad with optimized workers (v. 6 NG). Judging from my experience with my Intel mini, Alex' latest optimized Intel worker results in a speed gain of roughly 30 %. All things being linear (which they might well not be), it looks like Mac Pros could quite easily push into the 4000 RAC territory. The days of the Quad reign are surely numbered...
     
Gecko_r7
Forum Regular
Join Date: Oct 2005
Location: Las Vegas, NV
Status: Offline
Reply With Quote
Aug 15, 2006, 01:15 PM
 
Originally Posted by Piednoel Francois
If you look at the FFT using 4 vectors in parallel, you have to try to code your FFT in a way you minimize the penalities: Branching, Memory footprint, and in the case of Core, you want to use as many SSEx 128Bits instruction as you can. To use SIMD efficenly, you want to move your data from Array of Structure to Structure of Structure.
Alex: While his comments were in reference to Core 2, do you think this could also be applied w/ Altivec on G4s and G5s?
     
beadman
Dedicated MacNNer
Join Date: Nov 2004
Location: Virginia
Status: Offline
Reply With Quote
Aug 15, 2006, 04:32 PM
 
Originally Posted by halimedia
Just found the first Mac Pro that entered the top-1000 list. A RAC of over 600 after less than one week of crunching - impressive!

http://setiweb.ssl.berkeley.edu/show...hostid=2599334
http://setiweb.ssl.berkeley.edu/resu...hostid=2599334

It's running stock workers as of this writing. Shame it's an anonymous machine - otherwise we might be able to alert the owner to the optimized workers...

Halimedia:
My MBP 2.67 GHz Core Duo is averaging a little over 1000 per day. I have it split 50/50 between SETI and Einstein.
http://setiathome.berkeley.edu/show_...hostid=2546291
http://einstein.phys.uwm.edu/show_ho...?hostid=693388

I'm running the stock Einstein and Alex's optimized app for SETI.

beadman
     
 
Thread Tools
 
Forum Links
Forum Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts
BB code is On
Smilies are On
[IMG] code is On
HTML code is Off
Top
Privacy Policy
All times are GMT -4. The time now is 07:25 AM.
All contents of these forums © 1995-2017 MacNN. All rights reserved.
Branding + Design: www.gesamtbild.com
vBulletin v.3.8.8 © 2000-2017, Jelsoft Enterprises Ltd.,