Bernd MachenschalkForum moderator
Project administrator
Project developer
Send message
Joined: Oct 15 04
Posts: 2973
Credit: 76,600,017
RAC: 33,729
I've been working on the PPC code a bit more with what I learned from the coding for x86 and made a new version. I've seen it run about 20% faster than the previous 4.23 App on some machines.
However, applying all the measurements we need for copmpatibility of an "official App", this code slows down even behind the 4.23 App, at least on G5. So for now I made an App that maintains the maximum possible compatibility without loosing speed. It turned out to be a G5 App that should run on MacOS 10.4 as well as on 10.3. It is available from our Power User Apps page. I haven't seen a single invalid result from this code in internal tests.
I'll continue to investigate this level of "maximum useful compatibility" for G4 machines. For G3 I don't have any more ideas for speeding up the Apps anyway.
Happy crunching!
BM
____________
BM
Dual 2.0 G5, 10.4.7, Boinc 5.4.9
No success with 4.27. Boinc Manager deletes the app immediately on start up. (Tried it twice.)
As experiment, tried copying the app in to the Project's Ein folder after Boinc Manager was already running. Dowloaded a WU but it errored out immediately with Code 2 (0x2).
4.23 ran fine.
Update:
Seems to be looking for App. 4.26. Error in the XML?
resultid
____________
Bernd MachenschalkForum moderator
Project administrator
Project developer
Send message
Joined: Oct 15 04
Posts: 2973
Credit: 76,600,017
RAC: 33,729
Sorry. There was indeed an error in the app_info.xml. I updated the package, new md5 is ecfa6c45e790b135c72587e1e2fad4ec.
BM
____________
BM
Bernd,
thanks for this fantastic new release! On my G5 Dual 2.7GHz machine crunch time went down from 23,000 seconds to 18,000. Now it is once again on par with the fastest AMD-processors and faster then comparable Intels.
Thanks again!
____________
My G5’s first WU with v4.27 has been reported and validated, and a second is almost crunched. So far it’s looking about 20% faster than v4.23, or 40% faster than v4.12. That should bring its productivity up to about ten credits per hour.
By way of comparison, the same system earns about 8.3 CS/h on typical SETI@home WUs with the Enhanced Beta v5.17. This version has a higher credit/Flop rate than the current standard v5.13, with which I made about 7 CS/h (before I started running an optimized app).
I second the motion of thanks!
____________
How do i INSTALL it on my G5 PowerPC running 10.4.7 I downloaded and unzipped and its a folder with an xml and a app
Now where do i put them? Sorry, i was born this stupid and its been uphill ever since.
Bernd MachenschalkForum moderator
Project administrator
Project developer
Send message
Joined: Oct 15 04
Posts: 2973
Credit: 76,600,017
RAC: 33,729
How do i INSTALL it on my G5 PowerPC running 10.4.7 I downloaded and unzipped and its a folder with an xml and a app
Now where do i put them? Sorry, i was born this stupid and its been uphill ever since.
Put them in the folder
/Library/Application Support/BOINC Data/projects/einstein.phys.uwm.edu/
and restart the BOINC Manager.
More detatiled instructions for such packages can be found on the Beta Test Page, look for the MacOS PPC section.
BM
____________
BM
I have installed this new item by dragging it into the boic data folder. The message taB displays the message 'now using version 427...' but boys:
It Doesnt Run Any Faster. Why is the Mac so slow compared to my pc which SCREAMS! Something like 5 times more points! Is there a setting that will demand shorter MU's?
____________
Bernd MachenschalkForum moderator
Project administrator
Project developer
Send message
Joined: Oct 15 04
Posts: 2973
Credit: 76,600,017
RAC: 33,729
Have you actually completed a Task with the new version? Don't count on the time estimated by the Client, the estimation is based on the old App.
There are some postings in the threads here that discuss short and long WUs. See e.g. Bruce's "Information about S5" (sticky thread in Cruncher's Corner).
BM
____________
BM
i use now the 4.27 version on my G5 1,8Ghz and a WU h1-1412,5 have been computed in 27 000 seconds against 44 000 with the 4.12 version so 60% faster !
thanks for this optimization !
____________
I have found the following improvements in my WU times:
Mac G4/1.0 Gig (4.26) - 9,300/7,150 CPU/s = 23% faster
Mac G5/2.0 Gig (4.27) - 39,500/32,100 CPU/s = 18.7% faster
MBP Duo 2.16 (4.28) - 45,400/39,800 CPU/s = 12% faster
I am surprised that the MBP is so slow compared with the G5. After all, nominally, the MBP should be, if anything, a little faster than the G5, should it not?
____________
Bernd MachenschalkForum moderator
Project administrator
Project developer
Send message
Joined: Oct 15 04
Posts: 2973
Credit: 76,600,017
RAC: 33,729
Mac G4/1.0 Gig (4.26) - 9,300/7,150 CPU/s = 23% faster
Mac G5/2.0 Gig (4.27) - 39,500/32,100 CPU/s = 18.7% faster
MBP Duo 2.16 (4.28) - 45,400/39,800 CPU/s = 12% faster
Which versions are these compared to, especially on the G5? I suspect it was 4.23, or was it 4.12, too?
BM
____________
BM
Mac G4/1.0 Gig (4.26) - 9,300/7,150 CPU/s = 23% faster
Mac G5/2.0 Gig (4.27) - 39,500/32,100 CPU/s = 18.7% faster
MBP Duo 2.16 (4.28) - 45,400/39,800 CPU/s = 12% faster
Which versions are these compared to, especially on the G5? I suspect it was 4.23, or was it 4.12, too?
BM
Hi Bernd
It was the 4.23 in the case of the G5
____________
Oh! Dear! ....
Sorry, Bernd ... I just see now that I am running 4.26 on my G5 ... !!!!
I am such a clutz!!!
Apologies ... I will change it soon ...
:-((
[edit] ... many apologies for wasting your time ... !!!!!
____________
Sorry. There was indeed an error in the app_info.xml. I updated the package, new md5 is ecfa6c45e790b135c72587e1e2fad4ec.
BM
Tried it. Comutation error on all work units. Now I need to wait a day until new units sent since I've reached my quota of 30 per day. :-(
____________
Bernd MachenschalkForum moderator
Project administrator
Project developer
Send message
Joined: Oct 15 04
Posts: 2973
Credit: 76,600,017
RAC: 33,729
Sorry, Bernd ... I just see now that I am running 4.26 on my G5 ... !!!!
No need to feel sorry at all!
Does that mean that the 4.26 runs 18% faster than the 4.23 on your G5? Or that the 4.27 ran 18% faster than the 4.26?
Anyway - I'm still curious to know how much faster the 4.27 is compared to the 4.26 on a G5. I'm sure that's the perfect question for you now!
BM
____________
BM
Bernd MachenschalkForum moderator
Project administrator
Project developer
Send message
Joined: Oct 15 04
Posts: 2973
Credit: 76,600,017
RAC: 33,729
Sorry. There was indeed an error in the app_info.xml. I updated the package, new md5 is ecfa6c45e790b135c72587e1e2fad4ec.
BM
Tried it. Comutation error on all work units. Now I need to wait a day until new units sent since I've reached my quota of 30 per day. :-(
(I deleted the double post)
Have these errors been with the old, buggy package? It was only up for ~19 hours.
You should make sure you get the new one. The error messages look like the client didn't leave you any App at all in the project directoy. Make sure you either remove the buggy app_info.xml (which will bring the client to downlaod the 4.12 again) or replace it with the one from the current package (which will tell the client to preserve and use the 4.27 App).
BM
____________
BM
Sorry, Bernd ... I just see now that I am running 4.26 on my G5 ... !!!!
No need to feel sorry at all!
Does that mean that the 4.26 runs 18% faster than the 4.23 on your G5? Or that the 4.27 ran 18% faster than the 4.26?
Anyway - I'm still curious to know how much faster the 4.27 is compared to the 4.26 on a G5. I'm sure that's the perfect question for you now!
BM
I am soooo sorry ...
It was the 4.26 that ran 18% faster than the 4.23
____________
Bernd MachenschalkForum moderator
Project administrator
Project developer
Send message
Joined: Oct 15 04
Posts: 2973
Credit: 76,600,017
RAC: 33,729
I am soooo sorry ...
It was the 4.26 that ran 18% faster than the 4.23
Thank you - that's almost too good to be true.
Do me a favor and look into the folder
/Library/Application Support/BOINC Data/projects/einstein.phys.uwm.edu/"
to see if there really is a 4.26 App. The old (erroneous) app_info.xml may under certain circumstances have reported the 4.27 as 4.26.
That actually was my fault - I tested the App more or less thoroughly, but not the app_info.xml...
BM
____________
BM
I am soooo sorry ...
It was the 4.26 that ran 18% faster than the 4.23
Thank you - that's almost too good to be true.
Do me a favor and look into the folder
/Library/Application Support/BOINC Data/projects/einstein.phys.uwm.edu/"
to see if there really is a 4.26 App. The old (erroneous) app_info.xml may under certain circumstances have reported the 4.27 as 4.26.
That actually was my fault - I tested the App more or less thoroughly, but not the app_info.xml...
BM
OK Bernd - I have taken screen shots as follows:
1. This is the Einstein folder (/Library/Application Support/BOINC Data/projects/einstein.phys.uwm.edu/)
Einstein Folder
2. This is the app_info.xml file inside the "/einstein.phys.uwm.edu/" file
"app_info.xml" file
3. A shot of the messages file on my G5 showing that 4.27 is the app being used.
Messages
It is useful to note that I am now settled down to 23,500 CPUs which is a blistering speed when compared to 4.12 which was ~ 39,500 [edit] They were both of the series "l1_0887.0 ... "
A BIG THANK YOU, Bernd
David
____________
BTW, Bernd, I have an even bigger speed-up to report with MacBookPro.
Running 4.12, it was returning a WU in 40,000 CPUs
With 4.28 it is completing in ~ 18,000 CPUs
Sadly it does not allow me to open the /app_info.xml/ file
DM
____________
Sadly it does not allow me to open the /app_info.xml/ file
Use TextEdit. Remember not to save the file when closing it.
____________
Bernd MachenschalkForum moderator
Project administrator
Project developer
Send message
Joined: Oct 15 04
Posts: 2973
Credit: 76,600,017
RAC: 33,729
Sadly it does not allow me to open the /app_info.xml/ file
Use TextEdit. Remember not to save the file when closing it.
TextEdit should be able to open the file. However the app_info wasn't buggy for 4.28, so I think I know which one you are using there.
I have written app_infos for the Mac with TextEdit, what's wrong with saving the file (as long as you didn't modify it to break)?
BM
____________
BM
I have written app_infos for the Mac with TextEdit, what's wrong with saving the file (as long as you didn't modify it to break)?
Just a safety precaution for the inadequately clued. I’ve edited a couple of app_info files myself, but never while BOINC was running … and I have lost work (on S@h) due to erroneous entries.
____________
BTW, Bernd, I have an even bigger speed-up to report with MacBookPro.
Running 4.12, it was returning a WU in 40,000 CPUs
With 4.28 it is completing in ~ 18,000 CPUs
Sadly it does not allow me to open the /app_info.xml/ file
DM
Bodley: What are you doing differently with your MBP? I have an MBP 2.16 and it is consistantly taking around 19,700 cpusec per WU running 4.28. 2 GB RAM, prefs set to never sleep, and dim screen after 5 minutes.
C
____________
Join Team MacNN
BTW, Bernd, I have an even bigger speed-up to report with MacBookPro.
Running 4.12, it was returning a WU in 40,000 CPUs
With 4.28 it is completing in ~ 18,000 CPUs
Sadly it does not allow me to open the /app_info.xml/ file
DM
Bodley: What are you doing differently with your MBP? I have an MBP 2.16 and it is consistantly taking around 19,700 cpusec per WU running 4.28. 2 GB RAM, prefs set to never sleep, and dim screen after 5 minutes.
C
I am not quite sure WHAT I am doing! Perhaps it is because I have NOTHING else running on the machine? It has settled to 18,500 cpus ... which is only 20 mins faster ... and about 1 credit more than yours.
I too am on "never sleep" and the display goes to sleep after 5 mins ... same as you. I also have a small room fan playing on it (I had to return my battery under the recall scheme!).
Other than that ... it might be in the WU batch? I am crunching a swatch of:
h1_1138.5_S5R1__1222_S5R1a_0 (and _1)
DM
____________
Dual 2.7 G5, 10.4.8, BOINC 5.4.9
Is the app_info.xml file ok now if I install 427 downloaded from the "Power User Apps" page suggested by BM's original message nr 46252?
My Mac is using 426 at the moment. Is it worth upgrading? How do I measure the effect? Be specific and not to technical please, I'm not used to the lingo here (yet).
____________
Bernd MachenschalkForum moderator
Project administrator
Project developer
Send message
Joined: Oct 15 04
Posts: 2973
Credit: 76,600,017
RAC: 33,729
Is the app_info.xml file ok now if I install 427 downloaded from the "Power User Apps" page suggested by BM's original message nr 46252?
Yes.
My Mac is using 426 at the moment. Is it worth upgrading?
Yes. I think the 4.27 is roughly 20% faster than 4.26 (it only runs on G5, so we can't make it the "official", i.e. automatically downloaded App, though).
How do I measure the effect? Be specific and not to technical please, I'm not used to the lingo here (yet).
You should note a reduction of the CPU times you need for a Task, i.e. simply crunching faster. The variation between Tasks from the same set (i.e. having names that start with the same letters and numbers) is about +-5%, so the change should be notable even in "worst case". If the time your machine spends for BOINC is rather constant, over a week you should also notice an increasing RAC for that machine.
BM
____________
BM
I did a comparison on my G5 between "official" current app and the 4.27 optimized app:
http://einstein.phys.uwm.edu/results.php?hostid=416972
I measured a consistent ~35% increase in credits/time.
If you're reading this while the recent units are still displayed, the official-app WUs are the two reported on 21 Oct 2006 17:31:34 UTC and on 20 Oct 2006 15:03:53 UTC. Average execution time 75,600 seconds (@ 113.22 credits).
The 4 WUs dated after that are using the 4.27 optimized app. Average execution time 55,200 seconds (@ 113.22 credits).
This is a 1.8 GHz G5 iMac, with 1 GB RAM.
____________
Are you a musician? Join the Musicians team.
Meet the Musicians
I did a comparison on my G5 between "official" current app and the 4.27 optimized app:
http://einstein.phys.uwm.edu/results.php?hostid=416972
I measured a consistent ~35% increase in credits/time.
If you're reading this while the recent units are still displayed, the official-app WUs are the two reported on 21 Oct 2006 17:31:34 UTC and on 20 Oct 2006 15:03:53 UTC. Average execution time 75,600 seconds (@ 113.22 credits).
The 4 WUs dated after that are using the 4.27 optimized app. Average execution time 55,200 seconds (@ 113.22 credits).
This is a 1.8 GHz G5 iMac, with 1 GB RAM.
are you sure of 55,200 seconds of calculation credit 113 ?
on my G5 1,8Ghz with 4.27 app(and BOINC 5.4.9) i have WU of 27000 seconds of calculation that credit 113
____________
on my G5 1,8Ghz with 4.27 app(and BOINC 5.4.9) i have WU of 27000 seconds of calculation that credit 113
Most of the long WUs (111–113 CS) my dual-core G5/2.3 GHz has crunched recently with v4.27 took between 20000 and 28000 seconds (always near either end of the range, with none of them falling between 21000 and 27000 s), but I’ve had many that ran much longer: quite a few over 40000s, some as high as 46000 s. As a result this system’s productivity has varied widely over the last couple of months, ranging from about 9 to 20 CS/h.
____________
on my G5 1,8Ghz with 4.27 app(and BOINC 5.4.9) i have WU of 27000 seconds of calculation that credit 113
Most of the long WUs (111–113 CS) my dual-core G5/2.3 GHz has crunched recently with v4.27 took between 20000 and 28000 seconds (always near either end of the range, with none of them falling between 21000 and 27000 s), but I’ve had many that ran much longer: quite a few over 40000s, some as high as 46000 s. As a result this system’s productivity has varied widely over the last couple of months, ranging from about 9 to 20 CS/h.
ok , i did not know that it's surprizing there are so much differences in granted credits between WU;all WU on my computer are between 26000 to 29500 seconds for 112-114 credits.
____________
Bernd MachenschalkForum moderator
Project administrator
Project developer
Send message
Joined: Oct 15 04
Posts: 2973
Credit: 76,600,017
RAC: 33,729
To get new S5RI work with this App, you must install a new app_info.xml. Stop the client, replace the old app_info.xml with this one, and start the client again.
____________
BM
To get new S5RI work with this App, you must install a new app_info.xml. Stop the client, replace the old app_info.xml with this one, and start the client again.
Excuse me for being a bit elementary here, but I did not find an old app_info.xml so I am confused when you say that I should replace the old one with this updated one.
Could you help me to understand into which directory I should copy this file?
I am soooo sorry ...
It was the 4.26 that ran 18% faster than the 4.23
Thank you - that's almost too good to be true.
Do me a favor and look into the folder
/Library/Application Support/BOINC Data/projects/einstein.phys.uwm.edu/"
to see if there really is a 4.26 App. The old (erroneous) app_info.xml may under certain circumstances have reported the 4.27 as 4.26.
That actually was my fault - I tested the App more or less thoroughly, but not the app_info.xml...
BM
I looked in this directory on my computer and discovered the following files:
config_S5R1a.cfg
earth_05_09
sun_05_09
h1_0396.5_S5R1
I copied the app_info.xml and einstein_S5R1_4.27_powerpc-apple-darwin files into this directory (after I exited the client). I then restarted the client and clicked on the UPDATE button for Einstein and I was still getting the following messages:
Thu Jan 18 22:46:46 2007|Einstein@Home|Sending scheduler request to http://einstein.phys.uwm.edu/EinsteinAtHome_cgi/cgi
Thu Jan 18 22:46:46 2007|Einstein@Home|Reason: Requested by user
Thu Jan 18 22:46:46 2007|Einstein@Home|(not requesting new work or reporting completed tasks)
Thu Jan 18 22:46:51 2007|Einstein@Home|Scheduler request succeeded
What am I doing wrong here?
I looked farther up the message log and noticed this:
Thu Jan 18 22:57:21 2007|Einstein@Home|Found app_info.xml; using anonymous platform
Thu Jan 18 22:57:21 2007||Processor: 2 i386 MacBookPro1,1
Thu Jan 18 22:57:21 2007||Memory: 1.50 GB physical, 0 bytes virtual
Thu Jan 18 22:57:21 2007||Disk: 34.00 GB total, 3.01 GB free
Thu Jan 18 22:57:21 2007|Einstein@Home|URL: http://einstein.phys.uwm.edu/; Computer ID: 721615; location: home; project prefs: default
The second line of your log quotation below explains the problem. You are running a MacBookPro which contains an i386 Intel processor. The 4.27 app, as its name says, is built for the PowerPC processor, and as the Einstein page where you got it says, very specifically only for the G5 processor. (Which Apple doesn't even sell any more.)
I copied the app_info.xml and einstein_S5R1_4.27_powerpc-apple-darwin files into this directory (after I exited the client). I then restarted the client and clicked on the UPDATE button for Einstein and I was still getting the following messages:
I looked farther up the message log and noticed this:
Thu Jan 18 22:57:21 2007|Einstein@Home|Found app_info.xml; using anonymous platform
Thu Jan 18 22:57:21 2007||Processor: 2 i386 MacBookPro1,1
Thu Jan 18 22:57:21 2007||Memory: 1.50 GB physical, 0 bytes virtual
Thu Jan 18 22:57:21 2007||Disk: 34.00 GB total, 3.01 GB free
Thu Jan 18 22:57:21 2007|Einstein@Home|URL: http://einstein.phys.uwm.edu/; Computer ID: 721615; location: home; project prefs: default
____________
Are you a musician? Join the Musicians team.
Meet the Musicians
To get new S5RI work with this App, you must install a new app_info.xml. Stop the client, replace the old app_info.xml with this one, and start the client again.
thanks for the new xml :)
____________
Sorry but i have some problems with the new xml file ; the name is S5R1 or S5RI ?
Ven 19 jan 23:47:44 2007|Einstein@Home|Reason: To fetch work
Ven 19 jan 23:47:44 2007|Einstein@Home|Requesting 8640 seconds of new work
Ven 19 jan 23:47:49 2007|Einstein@Home|Scheduler request succeeded
Ven 19 jan 23:47:49 2007|Einstein@Home|Message from server: To get more Einstein@Home work, finish current work, stop BOINC, remove app_info.xml file, and restart.
Ven 19 jan 23:47:49 2007|Einstein@Home|Message from server: No work sent
Ven 19 jan 23:47:49 2007|Einstein@Home|No work from project
____________
Hi,
I see major differences in the new S5R1 work-units. On my 3 computers (1 Windows XP, 2 Macs) I receive WUs that all take approximately the same crunch-time of roughly 9,000 seconds. However, on my Windows computer and on one of the Macs they receive 84 credits (e.g. Result ID 68717626 , Result ID 68753890 ) while on my second Mac they receive only 55 credits (e.g. Result ID 68843151, Result ID 68843139). What is the difference?
____________
Hi,
I see major differences in the new S5R1 work-units. On my 3 computers (1 Windows XP, 2 Macs) I receive WUs that all take approximately the same crunch-time of roughly 9,000 seconds. However, on my Windows computer and on one of the Macs they receive 84 credits (e.g. Result ID 68717626 , Result ID 68753890 ) while on my second Mac they receive only 55 credits (e.g. Result ID 68843151, Result ID 68843139). What is the difference?
There was a mistake in the multiplier of the first units that were released. So those are giving higher than normal credit. They have since fixed that, and the ~55 numbers are the ones you should be seeing from now on.
This information was gotten from the S5R1 and Beyond post.
____________
Please help! I cannot get any new work.
Here is the error message I receive:
2007-04-23 13:23:25 [Einstein@Home] Message from server: To get more Einstein@Home work, finish current work, stop BOINC, remove app_info.xml file, and restart.
I did exact that - it downloaded version 4.16, which is much slower than 4.27. So I reinstalled BOINC 5.8.17 and also reinstalled 4.27. I reset the project - no go. Same problem.
What can I do to get 4.27 running again?
____________
Please help! I cannot get any new work.
Here is the error message I receive:
2007-04-23 13:23:25 [Einstein@Home] Message from server: To get more Einstein@Home work, finish current work, stop BOINC, remove app_info.xml file, and restart.
I did exact that - it downloaded version 4.16, which is much slower than 4.27. So I reinstalled BOINC 5.8.17 and also reinstalled 4.27. I reset the project - no go. Same problem.
What can I do to get 4.27 running again?
Build a time machine? You would need to go back about a week, to before the recent S5RI run issued its last workunit. Read the various threads about S5R2: the current app you're downloading - 4.16 for PPC - was rushed out in an un-optimised state so you would have something - anything at all - to crunch on. If you follow the advice, and remove the app_info.xml, you'll get newer, faster applications during the course of the run as development work proceeds.
the 4.27 app is for s5r1/s5ri which is the just finished science run. The 4.16 app is the one for the s5r2 run. You need to remove the app_info.xml file so that the 4.16 app will run.
The application name itself is changed, but starting its version numbering from 4.0 is somewhat confusing. The s5r2 work units are significantly larger than in s5r1 so will take several times longer to complete, although since the application is completely new and not as fully optimized runtimes should decrease in the future. Credits per WU should be somewhere in the 100-400 range depending on size. In addition, credit per hour in s5r1 was significantly higher on einstien than in almost every other boinc project and was reduced as a result. On windows platforms the reductions are ~40% on AMD systems, 25% on intel. Noone's posted any numbers on the performance shift on a mac system yet.
____________
the 4.27 app is for s5r1/s5ri which is the just finished science run. The 4.16 app is the one for the s5r2 run. You need to remove the app_info.xml file so that the 4.16 app will run.
The application name itself is changed, but starting its version numbering from 4.0 is somewhat confusing. The s5r2 work units are significantly larger than in s5r1 so will take several times longer to complete, although since the application is completely new and not as fully optimized runtimes should decrease in the future. Credits per WU should be somewhere in the 100-400 range depending on size. In addition, credit per hour in s5r1 was significantly higher on einstien than in almost every other boinc project and was reduced as a result. On windows platforms the reductions are ~40% on AMD systems, 25% on intel. Noone's posted any numbers on the performance shift on a mac system yet.
Thanks!
____________
Is there any progress/activity going on for optimized apps for the G4? I have a PowerMac G4 2x1ghz machine and on Seti the optimizations have nearly doubled its performance from the stock app, could the same thing be done for a G4 in Einstein? I'm assuming that the optimized app for the G5 won't run on a G4, is that accurate?
____________
Is there any progress/activity going on for optimized apps for the G4? I have a PowerMac G4 2x1ghz machine and on Seti the optimizations have nearly doubled its performance from the stock app, could the same thing be done for a G4 in Einstein? I'm assuming that the optimized app for the G5 won't run on a G4, is that accurate?
There are no current optimized apps here AFAICT, for any platform or CPU. During the S5R1 run a G4-optimized app, v4.26 IIRC, was beta-tested and then made standard; the G5-optimized v4.27 remained separate and only available by manual download—and yes, it contained processor-specific code that would not work on a G4. I assume the development for S5R2 will follow a similar course.
(I’ll be removing my v4.27 on this G5 as soon as I get a chance to stop BOINC; at the moment it’s got a couple of SzDG tasks on the go. It could be two or three weeks before this host gets any E@h work.)
Is there any progress/activity going on for optimized apps for the G4? I have a PowerMac G4 2x1ghz machine and on Seti the optimizations have nearly doubled its performance from the stock app, could the same thing be done for a G4 in Einstein? I'm assuming that the optimized app for the G5 won't run on a G4, is that accurate?
There are no current optimized apps here AFAICT, for any platform or CPU. During the S5R1 run a G4-optimized app, v4.26 IIRC, was beta-tested and then made standard; the G5-optimized v4.27 remained separate and only available by manual download—and yes, it contained processor-specific code that would not work on a G4. I assume the development for S5R2 will follow a similar course.
(I’ll be removing my v4.27 on this G5 as soon as I get a chance to stop BOINC; at the moment it’s got a couple of SzDG tasks on the go. It could be two or three weeks before this host gets any E@h work.)
Hello, witch compiler are use for the actually S5R2 app (for each CPU/OS)?
And is the autovectorization use (For Each CPU/OS)?
-OS X PowerPC gcc 3.3 or 4.0 ? autovectorized ?
-OS X Intel gcc or Intel compiler ? autovectorized or vectorized (in the code) ?
-Windows x86 gcc or intel compiler ? vectorized or autovectorized ?
And what will you do on specific optimized app for mac ? (compiler, vectorized/autovectorized, flags ..)
I noticed that my G5 1,8Ghz is as fast as an windowsX2 1,6Ghz (compared one core to one core) so it's not very good but it's not catastrophic too.
And a Macpro 3Ghz (so OS X) is 2,3 faster than my G5 (compared core to core) and
at same clock frequency 40% faster.(20% compared to Turion X2)
____________
The s5r2 work units are significantly larger than in s5r1 so will take several times longer to complete, although since the application is completely new and not as fully optimized runtimes should decrease in the future.
Any news on this one?
____________
Bikeman (Heinz-Bernd Eggenstein)Forum moderator
Project administrator
Project developer
Send message
Joined: Aug 28 06
Posts: 3020
Credit: 46,835,049
RAC: 60,483
The s5r2 work units are significantly larger than in s5r1 so will take several times longer to complete, although since the application is completely new and not as fully optimized runtimes should decrease in the future.
Any news on this one?
You mean news about optimization?
I was speculating earlier that only Intel and AMD CPUs will benefit significantly from the next round of optimization, but I now think I was wrong.
From what info is available to the public, you can see that the devs are trying to re-arrange the code in a way so that modern compilers like gcc 4.x would be able to auto-vectorize it. On a PPC this would mean Altivec !
Bernd and Akos will know more, not sure they are willing to share that info at this moment, tho ;-). I guess it all comes down to: it's ready when it's ready.
CU
BRM
____________
delete
____________
I guess it all comes down to: it's ready when it's ready.
CU
BRM
Is it?
____________
delete
____________