Optimized Boinc - SSE, SSE2...

Ed1934158
Ed1934158
Joined: 10 Nov 04
Posts: 62
Credit: 14481483
RAC: 0
Topic 193531

Could someone explain in detail how to install/use optimized Boinc versions for einstein@home (Windows and Linux)?
As I understand these versions use instructions SSE, SSE2 and you have a nice gain in speed compared to ordinary versions.
Now, what I don't understand is, why is that option not integrated in ordinary versions, since most of todays computers use these instructions? It seems like a terrible lose of CPU power.

Pooh Bear 27
Pooh Bear 27
Joined: 20 Mar 05
Posts: 1376
Credit: 20312671
RAC: 0

Optimized Boinc - SSE, SSE2...

Einstein code is very sensitive to different processing modes. It takes them time to develop versions for faster processing and make sure it's valid.

The issue is why not make it all that way? Well, there are still some very old computers out there. Not many, but there are some. Read about some being used in the Vintage Computer and Einstein@Home thread.

So, to stay compatible and have good science takes time.

Jord
Joined: 26 Jan 05
Posts: 2952
Credit: 5779100
RAC: 0

An optimized BOINC may claim

An optimized BOINC may claim 'fairer' credit when used with an optimized science application. But it won't speed things up as BOINC itself doesn't do any of the crunching. That's totally down to the science application.

The problem with adding MMX, SSE/SSE2/SSE3/SSSE3, 3DNow, etc. etc. is that they all need their own separate application, while they may not even be completely compatible between processors. For example, SSE2 on Intel CPUs does not do the same as SSE2 on AMD CPUs, so you'd need already 2 different applications for that.

Then there's the problem with the actual code of the science application. Yes, it may speed up the calculation, but at the same time it can produce a lot more erroneous results, where on the outside the task may look to have crunched OK, but the data returned is totally gobbledygook. Unusable in scientific terms.

Bikeman (Heinz-Bernd Eggenstein)
Bikeman (Heinz-...
Moderator
Joined: 28 Aug 06
Posts: 3522
Credit: 686128189
RAC: 573444

Hi! Basically there are

Hi!

Basically there are (at least) four approaches:

1) require at least SSE and exclude all other clients for Linus/Windows on i686. This would be kind of rude to all the users that still use non-SSE hosts.

2) implement the science app in various versions, for different platforms, and offer them for download so they can be installed manually by "Power Users". This is the current approach. The drawback is that only a minority will actually use the enhanced programs. It's great for beta-testing, tho.

3) Put different versions of the same science algorithm into a single program, which is distributed automatically to all users. The first thing the program will do is to detect the CPU's capabilities and then select the appropriate code-path.
There are some drawbacks as well, for example you have to generate a new app version which will be updated automatically on ALL clients even if only the code for one of the code-paths was changed.

4) Write different versions of the science app for different CPU capabilities. The server side BOINC code will then have to distribute the appropriate version to the clients, depending on the CPU capabilities that the clients report to the server.

There are pros and cons for all alternatives.

CU

Bikeman

Ed1934158
Ed1934158
Joined: 10 Nov 04
Posts: 62
Credit: 14481483
RAC: 0

OK, I see the problem.

OK, I see the problem. Nevertheless, lets say that gain is 10%, and as I saw many users have even greater gains. That means that for 2 extra clients (one for SSE AMD, and one for SSE Intel) you get 10TFlops, which is, I think, worth doing. Why did they make these instructions anyway if it's so difficult to make any use of them? Does any software use these instructions, if so, how did they solved their problems? If I'm not mistaking folding@home uses some extra instructions.
One more thing, it is not really clear to me how to get this to work. I download the zip file, unpack it, I get 4 files:
1. app_info.xml
2. einstein_S5R3_4.42_graphics_windows_intelx86.exe
3. einstein_S5R3_4.42_windows_intelx86.exe
4. einstein_S5R3_4.42_windows_intelx86 - some kind of debug database
When I try to run the 2. or 3. I get init_data.xml and stdrr.txt files. Is there any explanation for this except "Only run this if you're sure of what you're doing"? Is there any help or something similar or can someone explain it to me (and potential new users)?

Bikeman (Heinz-Bernd Eggenstein)
Bikeman (Heinz-...
Moderator
Joined: 28 Aug 06
Posts: 3522
Credit: 686128189
RAC: 573444

RE: OK, I see the problem.

Message 79574 in response to message 79573

Quote:
OK, I see the problem. Nevertheless, lets say that gain is 10%, and as I saw many users have even greater gains. That means that for 2 extra clients (one for SSE AMD, and one for SSE Intel) you get 10TFlops, which is, I think, worth doing.


Exactly, and that's the reason why SSE optimized apps are being offered, so that they can be tested here and later be distributed automatically.

Quote:

One more thing, it is not really clear to me how to get this to work. I download the zip file, unpack it, I get 4 files:
1. app_info.xml
2. einstein_S5R3_4.42_graphics_windows_intelx86.exe
3. einstein_S5R3_4.42_windows_intelx86.exe
4. einstein_S5R3_4.42_windows_intelx86 - some kind of debug database
When I try to run the 2. or 3. I get init_data.xml and stdrr.txt files.

Hi!

You have to unzip the files to a specific directory where BOINC stores the science apps.

Buts first you should stop BOINC. Right-Click on the bOINC symbol in the lower right hand corner "tray" area and select "Exit".

Then install the files:

Most likely, you have intstalled BOINC to a directory like

c:\\Program Files\\BOINC

The files in the ZIP archive need to be unpacked to the following directory then:

c:\\Program Files\\BOINC\\projects\\einstein.phys.uwm.edu\\

Then restart BOINC, and the new app will be used automatically for Einstein@Home.

Quote:

Is there any explanation for this except "Only run this if you're sure of what you're doing"?

The basic message is that this is mainly intended for experienced BOINC users. Installing these files will cause the automatic science app update mechanism to stop, unless you remove the "app_info.xml" file again, this app will stay indefinitely on your PC, which is probably not what you want because future versions will further improve performance. Therefore you will want to follow the discussions in this forum closely so that you can decide when to update manually to a new version or (by delting the app_info.xml file again) return to an automatic update.

Please let us know if you succeeded in installing the new app.

CU
Bikeman

RandyC
RandyC
Joined: 18 Jan 05
Posts: 6006
Credit: 111139797
RAC: 0

I'm making the following

Message 79575 in response to message 79573

I'm making the following assumptions:
1. The system you're installing on is Windows, has BOINC installed, and is attached to E@H
2. The system is SSE capable
3. You fat-fingered the release below and it's really 4.32...4.42 doesn't exist yet

Do the following:
1. Shutdown BOINC completely (not just Boinc-mgr)
2. (Optional) Back-up entire BOINC folder someplace else
3. Copy all 4 files below into folder: ...\\boinc\\projects\\einstein.phys.uwm.edu
4. Restart BOINC

HTH

[edit]Bikeman, either you're too fast or I'm too slow ;) [/edit]

Quote:
OK, I see the problem. Nevertheless, lets say that gain is 10%, and as I saw many users have even greater gains. That means that for 2 extra clients (one for SSE AMD, and one for SSE Intel) you get 10TFlops, which is, I think, worth doing. Why did they make these instructions anyway if it's so difficult to make any use of them? Does any software use these instructions, if so, how did they solved their problems? If I'm not mistaking folding@home uses some extra instructions.
One more thing, it is not really clear to me how to get this to work. I download the zip file, unpack it, I get 4 files:
1. app_info.xml
2. einstein_S5R3_4.42_graphics_windows_intelx86.exe
3. einstein_S5R3_4.42_windows_intelx86.exe
4. einstein_S5R3_4.42_windows_intelx86 - some kind of debug database
When I try to run the 2. or 3. I get init_data.xml and stdrr.txt files. Is there any explanation for this except "Only run this if you're sure of what you're doing"? Is there any help or something similar or can someone explain it to me (and potential new users)?


Seti Classic Final Total: 11446 WU.

Ed1934158
Ed1934158
Joined: 10 Nov 04
Posts: 62
Credit: 14481483
RAC: 0

RE: The basic message is

Message 79576 in response to message 79572

Quote:

The basic message is that this is mainly intended for experienced BOINC users. Installing these files will cause the automatic science app update mechanism to stop, unless you remove the "app_info.xml" file again, this app will stay indefinitely on your PC, which is probably not what you want because future versions will further improve performance. Therefore you will want to follow the discussions in this forum closely so that you can decide when to update manually to a new version or (by delting the app_info.xml file again) return to an automatic update.

Please let us know if you succeeded in installing the new app.

Bikeman


Yes, I did it, there was no problem at all. Thank you very much!
It would be nice if this explanation was a part of zip file, as a "readme.txt". I also have one Linux machine. I assume that the procedure is the same?

P.S. RandyC, thank you too. Maybe you'll be faster with Linux question :)

Bikeman (Heinz-Bernd Eggenstein)
Bikeman (Heinz-...
Moderator
Joined: 28 Aug 06
Posts: 3522
Credit: 686128189
RAC: 573444

RE: I also have one Linux

Message 79577 in response to message 79576

Quote:
I also have one Linux machine. I assume that the procedure is the same?

Yes, it's exactly the same, except perhaps that you will have to terminate BOINC differently, depending on how you start it there.

CU
Bikeman

Jord
Joined: 26 Jan 05
Posts: 2952
Credit: 5779100
RAC: 0

RE: It would be nice if

Message 79578 in response to message 79576

Quote:
It would be nice if this explanation was a part of zip file, as a "readme.txt".


It's already on the Einstein Beta download page. If you don't read that, then why the need for a readme text file that people then probably won't read either?

th3
th3
Joined: 24 Aug 06
Posts: 208
Credit: 2208434
RAC: 0

From your kernel version it

From your kernel version it looks like you are using some kind of Ubuntu, if you installed boinc from the respository you can use this method:

sudo /etc/init.d/boinc-client stop
cd /var/lib/boinc-client/projects/einstein.phys.uwm.edu
sudo tar -xf /home//einstein_S5R3_4.35_i686-pc-linux-gnu.tar.gz
sudo /etc/init.d/boinc-client start

Comment viewing options

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