I don't really think so. From what I heard from other crunchers C/H is normally fairly constant, and all WUs give about equal credit in relation to crunching time. Maybe it really has to do with running BOINC in a VM... never tried it out but it sounds plausible to me.
Really great news about the new science app :-D very nice work from Akos (again). Am I correct that the performance increase will be for all platforms?
EDIT: I was referring to the post about different WU sizes and differences in C/H.
I have collected pretty many data about this stuff(Perl is my friend ;-))
Yesterday I've fed a database with all results of the top 20 crunchers in our team and as soon as I have the real clock frequencies I can do some research. Formerly I used a spreadsheet and collected data from my and some other hosts. And I'm very shure that we have differences of about ±5% all the time since credits are fixed serverside. Nevertheless this is no reason to whine, let's hope the client based credits never come back. ;)
Coming back on the discussion of SSE2 detection, here's somthing that you probably won't believe :
Doing some inspection on the assembly code, I found that the detection code is making some strcmp call. Why would it do that? I think I just found out:
Sorry but what is the string compared to, and what does that do? You lost me there.
As for credit differences: 5%, okay, I can easily believe that. But 30 and more? Never ever. There was sth else involved on my host than just differently-sized WUs.
Sorry but what is the string compared to, and what does that do? You lost me there.
As for credit differences: 5%, okay, I can easily believe that. But 30 and more? Never ever. There was sth else involved on my host than just differently-sized WUs.
I think what's happening is this: the detection code will use the CPUID instruction to get the brand of the CPU as a text string. Then it will compare it to the string "AuthenticAMD" and if it maches ==> SSE2 support is disabled. That simple. No wonder that AMDs suffer under Win!
Yes, this is hard to believe, and if today was 1st of April I wouldn't either, but it's relatively easy to check by changing that string to something else.
I don't have a SSE2 capable AMD here, if anybody want's to give it a try please report the results.
CU
BRM
P.S.: This time I'm sober :-). If this really works it would be the funniest tuning ever ;-)
Well my 3500+ Venice should be well suited, I'm sure it can do SSE2 (not sure about SSE3, think it can, doesn't matter here anyway...) So is that tool you mentioned free, can I get it now and how big is the risk I'll kill something major on my system? I'd like to volunteer...
Could be interesting :-D but let's concentrate on the hard facts... I'll try to get said tool and install it on my AMD, give me 5 minutes...
EDIT: Just did it... no idea what this is going to do to my hostid, and it certainly looks strange to have a modified CPU ID, but lets hope it actually helps. I suspended all other projects, so results should come in rather soon...
Could be interesting :-D but let's concentrate on the hard facts... I'll try to get said tool and install it on my AMD, give me 5 minutes...
I just checked again, I think that tool only allows to edit the Processor name, not the vendor name which is "AthenticAMD" for AMD and "GenuineIntel" for intel.
RE: I don't really think
)
I have collected pretty many data about this stuff(Perl is my friend ;-))
Yesterday I've fed a database with all results of the top 20 crunchers in our team and as soon as I have the real clock frequencies I can do some research. Formerly I used a spreadsheet and collected data from my and some other hosts. And I'm very shure that we have differences of about ±5% all the time since credits are fixed serverside. Nevertheless this is no reason to whine, let's hope the client based credits never come back. ;)
cu,
Michael
Coming back on the discussion
)
Coming back on the discussion of SSE2 detection, here's somthing that you probably won't believe :
Doing some inspection on the assembly code, I found that the detection code is making some strcmp call. Why would it do that? I think I just found out:
# strings einstein_S5R2_4.17_windows_intelx86.exe | grep Auth
AuthenticAMD
Yes, the math lib doesn't like AMDs at all. Michael, you know what to do ;-) ?
CU
BRM
RE: Yes, the math lib
)
Go and get a gun? (Not so easy here in Germany, except for crazy pupils) ;-)
cu,
Michael
p.s. Did you think of an Hex-Editor? That's afaik the way how e.g. c't patched the Intel compiler.
Sorry but what is the string
)
Sorry but what is the string compared to, and what does that do? You lost me there.
As for credit differences: 5%, okay, I can easily believe that. But 30 and more? Never ever. There was sth else involved on my host than just differently-sized WUs.
RE: Sorry but what is the
)
I think what's happening is this: the detection code will use the CPUID instruction to get the brand of the CPU as a text string. Then it will compare it to the string "AuthenticAMD" and if it maches ==> SSE2 support is disabled. That simple. No wonder that AMDs suffer under Win!
Yes, this is hard to believe, and if today was 1st of April I wouldn't either, but it's relatively easy to check by changing that string to something else.
I don't have a SSE2 capable AMD here, if anybody want's to give it a try please report the results.
CU
BRM
P.S.: This time I'm sober :-). If this really works it would be the funniest tuning ever ;-)
Well my 3500+ Venice should
)
Well my 3500+ Venice should be well suited, I'm sure it can do SSE2 (not sure about SSE3, think it can, doesn't matter here anyway...) So is that tool you mentioned free, can I get it now and how big is the risk I'll kill something major on my system? I'd like to volunteer...
RE: # strings
)
Nothing like that in the Linux app and also nothing about Intel in both apps.
What do you think about corruption and conspiracy theory? ;-)
cu,
Michael
Could be interesting :-D but
)
Could be interesting :-D but let's concentrate on the hard facts... I'll try to get said tool and install it on my AMD, give me 5 minutes...
EDIT: Just did it... no idea what this is going to do to my hostid, and it certainly looks strange to have a modified CPU ID, but lets hope it actually helps. I suspended all other projects, so results should come in rather soon...
RE: Could be interesting
)
I just checked again, I think that tool only allows to edit the Processor name, not the vendor name which is "AthenticAMD" for AMD and "GenuineIntel" for intel.
Yeah, no "Authentic AMD"
)
Yeah, no "Authentic AMD" there... already wondered. Another idea?