.... I'd be curious if there are still machines out there without SSE and how they'd react on this App (the FFTW doku is somewhat ambiguous at that point).
Please test and report.
BM
OK, it's loaded and running on the trusty old Celeron 400 MMX. Done something over 3.5% at a bit over 1% per hour - be a while before we get the final answer.
Y'all might be amused to check the timings (and stderr_txt) for the previous task that box ran.....
Y'all might be amused to check the timings (and stderr_txt) for the previous task that box ran.....
Heh heh...
So I'm guessing you prepared an app_info with only ABP1 entries, upon restart the completed task was dumped only to be resent. But because the result file was already uploaded you were able to do some client_state fibbling to report it. And fair enough to avoid losing all that good crunching!
Okay, I see you pretty much confirm this in the CUDA thread. But I'll post anyway conceding that I only now noticed that cpu_time=report_date-(re)sent_date.
A new Einstein@home ABP1 App for Windows is available for Beta Test at Beta Test Page.
This is a big package to download, as it contains the 3.09 ABP1 application as well as the current 3.05 S5R5.
The new ABP1 App should mainly benefit from using SSE in the FFTw (and some other parts). I'd be curious if there are still machines out there without SSE and how they'd react on this App (the FFTW doku is somewhat ambiguous at that point).
app_info.xml does contain entries for ABP1 3.07, 3.08 and 3.09 and S5R5 3.05, tasks assigned to older apps will error out.
Please test and report.
BM
That Beta Test Page shows "einsteinbinary_ABP1_3.09_windows_intelx86.zip" for windows which I assume should not be run on my AMD opteron system and is the cause of the every wu erroring out after one second (all 2 of them before I stopped it).
Vista-64; dual AMD Opteron 270's; 9800gtx+
lemme know if my hunch is correct or not. Error log follows
Log Name: Application
Source: Application Error
Date: 2009-08-09 9:55:17 PM
Event ID: 1000
Task Category: (100)
Level: Error
Keywords: Classic
User: N/A
Computer: jys2x275
Description:
Faulting application einsteinbinary_ABP1_3.07_windows_intelx86_cuda.exe, version 0.0.0.0, time stamp 0x4a6f0656, faulting module cufft.dll, version 0.0.0.0, time stamp 0x49452e6c, exception code 0x40000015, fault offset 0x000015f8, process id 0x1354, application start time 0x01ca196541067774.
Event Xml:
That Beta Test Page shows "einsteinbinary_ABP1_3.09_windows_intelx86.zip" for windows which I assume should not be run on my AMD opteron system and is the cause of the every wu erroring out after one second (all 2 of them before I stopped it).
Vista-64; dual AMD Opteron 270's; 9800gtx+
lemme know if my hunch is correct or not. Error log follows
thanks.
Hm.
Faulting application einsteinbinary_ABP1_3.07_windows_intelx86_cuda.exe
looks like the problem occurred switching from the CUDA App to the 1.09 (CPU-only) App.
Faulting application einsteinbinary_ABP1_3.07_windows_intelx86_cuda.exe
looks like the problem occurred switching from the CUDA App to the 1.09 (CPU-only) App.
BM
See my note in the CUDA thread. Normally, that would cause BOINC to abandon all CUDA tasks with "No application found for task: windows_intelx86 307 cuda; discarding" (and no report to the project).
Unless BB tried to merge the two Beta app_info files together, and got something slightly wrong?
I've been running this 3.09 app on a Q9450, Windows 7 64-bit, for a few days now and all seem fine. Times have dropped from ~29000 seconds to ~22000 seconds. Nice speedup!
This is a Beta testing area. At the moment, there are two separate Beta tests being run (for Windows - let's leave the other operating systems out of it for now).
One test (this thread) is for the ABP1 v3.09 application for CPU only. The other test is for a CUDA app.
They way Einstein have set these tests up, they are mutually exclusive. This one tests CPU (no CUDA), the other tests CUDA (no CPU - except the reliable old S5R5, common to both and not being tested).
As I have demonstrated in the other (CUDA) thread, it is possible to write a combination app_info.xml file and participate in both tests at once - but the project isn't asking you to do that, and hasn't supplied the tools.
Unless you are prepared to go the extra mile, and have the knowledge and experience to prepare your own app_info.xml files, I suggest you restrict yourself to participating in one test at a time. And if you want to switch between them, I suggest you ensure that you have no Einstein tasks in progress (and learn from my mistake - report any completed ones!), then do a 'project reset' before installing the files for the next test - the same as you would have to do at the end of every Einstein Beta test, to return to normal automatic updating of revised applications.
As you can clearly see from my post, before your reply, I was under the impression that the ABP trials (both versions 3.07 and 3.09) were related to CUDA.
As I have an nVidia card, identified by BOINC as CUDA capable, I thought I would try out the new system.
My misunderstanding was compounded by the _cuda file present in the 3.07 client download.
Perhaps I can look back at the app_info files quoted here (your other thread) and see where the ABP CUDA client is activated. But, I would then need to see the 3.09 equivalent file present in the Einstein folder for the app_info file to access? Although, looking at your example app_info file I see a reference is made to the _3.07_cuda client.
Shih-Tzu are clever, cuddly, playful and rule!! Jack Russell are feisty!
RE: .... I'd be curious if
)
OK, it's loaded and running on the trusty old Celeron 400 MMX. Done something over 3.5% at a bit over 1% per hour - be a while before we get the final answer.
Y'all might be amused to check the timings (and stderr_txt) for the previous task that box ran.....
RE: Y'all might be amused
)
Heh heh...
So I'm guessing you prepared an app_info with only ABP1 entries, upon restart the completed task was dumped only to be resent. But because the result file was already uploaded you were able to do some client_state fibbling to report it. And fair enough to avoid losing all that good crunching!
Okay, I see you pretty much confirm this in the CUDA thread. But I'll post anyway conceding that I only now noticed that cpu_time=report_date-(re)sent_date.
RE: A new Einstein@home
)
That Beta Test Page shows "einsteinbinary_ABP1_3.09_windows_intelx86.zip" for windows which I assume should not be run on my AMD opteron system and is the cause of the every wu erroring out after one second (all 2 of them before I stopped it).
Vista-64; dual AMD Opteron 270's; 9800gtx+
lemme know if my hunch is correct or not. Error log follows
1000
2
100
0x80000000000000
29518
Application
jys2x275
einsteinbinary_ABP1_3.07_windows_intelx86_cuda.exe
0.0.0.0
4a6f0656
cufft.dll
0.0.0.0
49452e6c
40000015
000015f8
1354
01ca196541067774
thanks.
RE: That Beta Test Page
)
Hm.
Faulting application einsteinbinary_ABP1_3.07_windows_intelx86_cuda.exelooks like the problem occurred switching from the CUDA App to the 1.09 (CPU-only) App.BM
BM
RE: Hm. Faulting
)
See my note in the CUDA thread. Normally, that would cause BOINC to abandon all CUDA tasks with "No application found for task: windows_intelx86 307 cuda; discarding" (and no report to the project).
Unless BB tried to merge the two Beta app_info files together, and got something slightly wrong?
That reminds me -
)
That reminds me - continuation:
v3.07 (CPU): ~38,000 seconds
v3.07 (CUDA): ~21,000 seconds (7 validated, 1 pending, no errors except the initial DLL)
v3.08 (CPU): ~28,000 seconds
v3.09 (CPU): ~25,000 seconds
I've been running this 3.09
)
I've been running this 3.09 app on a Q9450, Windows 7 64-bit, for a few days now and all seem fine. Times have dropped from ~29000 seconds to ~22000 seconds. Nice speedup!
I am in the process of
)
I am in the process of replacing the CUDA version of Einstein 3.07 by 3.09.
Looking at the files I see the same ones for both -
app_info (common to both)
The 5 einstein_S5R5_3.05_windows_... files (common to both)
The 2 einsteinbinary_ABP1_3.07_graphics/windows_... files
(seems to be common to both except the new ones say _ABP1_3.09_
Now for where the difference lies -
The last file that the older CUDA version has is a file titled -
einsteinbinary_ABP1_3.07_windows_intelx86_cuda
This last file is missing from the 3.09 file group. I assume it is OK without this file telling the GPU that it is to be used by Einstein?
Now installed and 3.05 running on the CPU, together with Milkyway and Malaria (all vying for a core, with significant load going to Milkyway).
I see I have just downloaded 2 more 3.05 WUs, and no ABP1s at present.
Is there anything more I need to do to get the nVidia card working CUDA (missing .dlls, other files, drivers, etc.)
I installed the 185.85 nVidia driver earlier.
Shih-Tzu are clever, cuddly, playful and rule!! Jack Russell are feisty!
This is a Beta testing area.
)
This is a Beta testing area. At the moment, there are two separate Beta tests being run (for Windows - let's leave the other operating systems out of it for now).
One test (this thread) is for the ABP1 v3.09 application for CPU only. The other test is for a CUDA app.
They way Einstein have set these tests up, they are mutually exclusive. This one tests CPU (no CUDA), the other tests CUDA (no CPU - except the reliable old S5R5, common to both and not being tested).
As I have demonstrated in the other (CUDA) thread, it is possible to write a combination app_info.xml file and participate in both tests at once - but the project isn't asking you to do that, and hasn't supplied the tools.
Unless you are prepared to go the extra mile, and have the knowledge and experience to prepare your own app_info.xml files, I suggest you restrict yourself to participating in one test at a time. And if you want to switch between them, I suggest you ensure that you have no Einstein tasks in progress (and learn from my mistake - report any completed ones!), then do a 'project reset' before installing the files for the next test - the same as you would have to do at the end of every Einstein Beta test, to return to normal automatic updating of revised applications.
Thanks Richard As you can
)
Thanks Richard
As you can clearly see from my post, before your reply, I was under the impression that the ABP trials (both versions 3.07 and 3.09) were related to CUDA.
As I have an nVidia card, identified by BOINC as CUDA capable, I thought I would try out the new system.
My misunderstanding was compounded by the _cuda file present in the 3.07 client download.
Perhaps I can look back at the app_info files quoted here (your other thread) and see where the ABP CUDA client is activated. But, I would then need to see the 3.09 equivalent file present in the Einstein folder for the app_info file to access? Although, looking at your example app_info file I see a reference is made to the _3.07_cuda client.
Shih-Tzu are clever, cuddly, playful and rule!! Jack Russell are feisty!