But can you give me a reference unit that I can use for all the optimized apps?
And describe how to use this reference unit?
Thanks, you do a great job!
I have a Pentium 930 D @ 3,6 GHz
And the time with the s5t0304 is about 54 min (avarage for 10 wus)
My aproach:
1. Make new instalaltion of boinc (or you can also finished all your work in current instalation)
2. conect to E@H project
3. download new WU
4. during downloading:
a. on tab "Projects" -> "No new work"
b. on tab "Work" -> "Suspend" (You can not start crunching immediately)
5. stop Boinc
6. backup Boinc directory after downloading all files (I use packing to zip archive)
7. start Boinc
8. on tab "Work" -> "Resume"
After finishing of work I:
1. unpack archive (I use new directory)
2. change program einstein_S5R1_4.02_windows_intelx86.exe with new version
3. start Boinc
4. on tab "Work" -> "Resume"
I think the official app could be about 2 times faster than the current speed. There are lots of needless FPU -> memory -> FPU operations, but i could not remove them because these movements change the last bit of the numbers ( only one bit! ).
The results would be better ( the last bit would be also good ) without these movements and the app would be more faster, but... the current S5 validator doesn't accept these ( better, faster ) results.
@ Akosf - ouch! that must be difficult to work with. Surprised we're all returning valid results even with the official app. Usually some possibility of extremely tiny differences even between two units done with official app.
Maybe you could ask the project leaders if they would consider letting the validators be a tiny bit forgiving? but maybe there a good reason why they need it to be that strict.
If you are having issues with your ftp, I can setup an ftp account for you on my server. This way you would always have a backup if needed. Least I can do for all your work. It can hold all the patches at that if needed, will have 1 gig of space for your use if you would like it.
Also maybe Bernd or Bruce could load a few test WU's for testing use here as well. That way we do not impact the project while testing.
If you are having issues with your ftp, I can setup an ftp account for you on my server. This way you would always have a backup if needed. Least I can do for all your work. It can hold all the patches at that if needed, will have 1 gig of space for your use if you would like it.
Let me know.
Thanks, but i see that www.esnips.com works well!
( 1GB space, updload/download client, easy to use, etc... )
The results would be better ( the last bit would be also good ) without these movements and the app would be more faster, but... the current S5 validator doesn't accept these ( better, faster ) results.
@ Akosf - ouch! that must be difficult to work with. Surprised we're all returning valid results even with the official app. Usually some possibility of extremely tiny differences even between two units done with official app.
Maybe you could ask the project leaders if they would consider letting the validators be a tiny bit forgiving? but maybe there a good reason why they need it to be that strict.
I think they use this 'standard' method because all platfrom ( win, linux, -freebsd, etc... ) can produce 'standard' results, so they are absolutely similar. Only the calculations are a bit slower...
unfortunately I only have an AMD Athlon XP 1700+ (without SSE3).
But S5T0307 seems to be a real fast one...
I started h1_0804.0_S5R1__685_S5R1a_1 with S5T0003, switched to S5T0304 after 30 seconds and let it run.
When I noticed S5T0307 being available, I switched to it.
The (long) WU was processed 4,5% with an estimated duration of 99,1% (compared to h1_0804.0_S5R1__700_S5R1a)
With S5T0307 the estimatated duration after 8% dropped to 93,2%!
@akosf: OK, I would do the
)
@akosf: OK, I would do the measurement.
But can you give me a reference unit that I can use for all the optimized apps?
And describe how to use this reference unit?
Thanks, you do a great job!
I have a Pentium 930 D @ 3,6 GHz
And the time with the s5t0304 is about 54 min (avarage for 10 wus)
RE: @akosf: OK, I would do
)
I don't have any reference units.
I usually did this:
- download a wu
- disable boinc network access
- stop boinc
- backup boinc directory
for test:
- restore boinc directory
- copy a patched file here
- start boinc (network is disabled)
RE: @akosf: OK, I would do
)
My aproach:
1. Make new instalaltion of boinc (or you can also finished all your work in current instalation)
2. conect to E@H project
3. download new WU
4. during downloading:
a. on tab "Projects" -> "No new work"
b. on tab "Work" -> "Suspend" (You can not start crunching immediately)
5. stop Boinc
6. backup Boinc directory after downloading all files (I use packing to zip archive)
7. start Boinc
8. on tab "Work" -> "Resume"
After finishing of work I:
1. unpack archive (I use new directory)
2. change program einstein_S5R1_4.02_windows_intelx86.exe with new version
3. start Boinc
4. on tab "Work" -> "Resume"
Thats all :)
EDIT: I forgot "disable boinc network access" :)
S5T0307.dat - eliminated
)
S5T0307.dat
- eliminated double jumps
- reduced amount of FPU macro ops
- removed double loads on general purpose registers
- better SSE register usage
- reduced memory and integer register usage
- optimized branch structure
- faster FPU comparisons
CPU: SSE compatible
RE: I think the official
)
@ Akosf - ouch! that must be difficult to work with. Surprised we're all returning valid results even with the official app. Usually some possibility of extremely tiny differences even between two units done with official app.
Maybe you could ask the project leaders if they would consider letting the validators be a tiny bit forgiving? but maybe there a good reason why they need it to be that strict.
To Akosf If you are having
)
To Akosf
If you are having issues with your ftp, I can setup an ftp account for you on my server. This way you would always have a backup if needed. Least I can do for all your work. It can hold all the patches at that if needed, will have 1 gig of space for your use if you would like it.
Also maybe Bernd or Bruce could load a few test WU's for testing use here as well. That way we do not impact the project while testing.
Let me know.
Come Join us at Hawaiian Beach Bums
RE: To Akosf If you are
)
Thanks, but i see that www.esnips.com works well!
( 1GB space, updload/download client, easy to use, etc... )
RE: Thanks, but i see that
)
Ok, offer is always open if needed. Server sets on dual OC-48's so downloads would not be a problem.
Come Join us at Hawaiian Beach Bums
RE: RE: The results would
)
I think they use this 'standard' method because all platfrom ( win, linux, -freebsd, etc... ) can produce 'standard' results, so they are absolutely similar. Only the calculations are a bit slower...
RE: S5T0307.dat -
)
unfortunately I only have an AMD Athlon XP 1700+ (without SSE3).
But S5T0307 seems to be a real fast one...
I started h1_0804.0_S5R1__685_S5R1a_1 with S5T0003, switched to S5T0304 after 30 seconds and let it run.
When I noticed S5T0307 being available, I switched to it.
The (long) WU was processed 4,5% with an estimated duration of 99,1% (compared to h1_0804.0_S5R1__700_S5R1a)
With S5T0307 the estimatated duration after 8% dropped to 93,2%!
Udo
