Information about the new S5 workunits

Annika
Annika
Joined: 8 Aug 06
Posts: 720
Credit: 494410
RAC: 0

RE: I edited the executable

Message 37837 in response to message 37836

Quote:
I edited the executable and it is running fine, but after this short time I can't say anything about performance.
But I'm very nosy in this case, so I will stay up for some more time and watch the progress. :-))

Will you please tell me how to do it, if it's not too hard?

M. Schmitt
M. Schmitt
Joined: 27 Jun 05
Posts: 478
Credit: 15872262
RAC: 0

RE: RE: I edited the

Message 37838 in response to message 37837

Quote:
Quote:
I edited the executable and it is running fine, but after this short time I can't say anything about performance.
But I'm very nosy in this case, so I will stay up for some more time and watch the progress. :-))

Will you please tell me how to do it, if it's not too hard?

It's easy. :-)

I googled for an hex editor(had one, but got lost), found HEX-Editor MX, downloaded, virus-checked an installed it.
Opened Admin Cmd window and typed "net stop boinc".
Started the editor and open the einstein.exe. Searched with the text search for Authentic and found AuthenticAMD. Edited that to AuthenticABC, saved file and started boinc again. Voila.

37:48 min - 3,38% could end up around 15h. WU is 408.50. Wingman has had some of them and got 318.02 credits. Rough guess: FASTER!
Remember, the same host got ~14 cr/h under Windows.

cu,
Micha

Annika
Annika
Joined: 8 Aug 06
Posts: 720
Credit: 494410
RAC: 0

Just did as you said. It

Just did as you said. It seems to have worked fine, let's see what it does to my host's performance. I have no Windows values from that box, but I think if it gets reasonably close to the Linux app it's okay...

M. Schmitt
M. Schmitt
Joined: 27 Jun 05
Posts: 478
Credit: 15872262
RAC: 0

RE: Just did as you said.

Message 37840 in response to message 37839

Quote:
Just did as you said. It seems to have worked fine, let's see what it does to my host's performance. I have no Windows values from that box, but I think if it gets reasonably close to the Linux app it's okay...

I think that would be funny, because I'm not aware of any problems of the Core - CPUs with the Win-app, except your one.
But I keep my fingers crossed for you. :-)

44:45 4,0% ~ 21,91 c/h (is ok!)
57:31 5,16% ~ 21,99 c/h :-) (should grow farther with progress)

cu,
Michael

Bikeman (Heinz-Bernd Eggenstein)
Bikeman (Heinz-...
Moderator
Joined: 28 Aug 06
Posts: 3522
Credit: 739084833
RAC: 1259840

RE: RE: Just did as you

Message 37841 in response to message 37840

Quote:
Quote:
Just did as you said. It seems to have worked fine, let's see what it does to my host's performance. I have no Windows values from that box, but I think if it gets reasonably close to the Linux app it's okay...

I think that would be funny, because I'm not aware of any problems of the Core - CPUs with the Win-app, except your one.
But I keep my fingers crossed for you. :-)

44:45 4,0% ~ 21,91 c/h (is ok!)
57:31 5,16% ~ 21,99 c/h :-) (should grow farther with progress)

cu,
Michael

Good morning!

I think Annika was referring to her Venice box.

Sooooo, any further results so far????

I must admit I didn't know about this Intel "policy". It's a real shame. Thanks for link, Richard!!!!!

AFAIK the Visual C(++) compiler was used to build the app, but it seems like the math library linked to the app is from intel.

A word of caution to all those out there who want to have their AMD boxes sped up:

If this works at all (and it does look like that at the moment), this can ONLY improve performance for those AMD CPUs that are SSE2 capable. So this won't work for Athlon and Athlon XP hosts which are SSE capable, but not SSE2 capable! Opterons and Athlon 64 do support SSE2.

Anyway, I guess it's best to wait for results and an official statement before everybody out there is modifying the executable :-).

CU

BRM

M. Schmitt
M. Schmitt
Joined: 27 Jun 05
Posts: 478
Credit: 15872262
RAC: 0

RE: RE: RE: Just did as

Message 37842 in response to message 37841

Quote:
Quote:
Quote:
Just did as you said. It seems to have worked fine, let's see what it does to my host's performance. I have no Windows values from that box, but I think if it gets reasonably close to the Linux app it's okay...

I think that would be funny, because I'm not aware of any problems of the Core - CPUs with the Win-app, except your one.
But I keep my fingers crossed for you. :-)

44:45 4,0% ~ 21,91 c/h (is ok!)
57:31 5,16% ~ 21,99 c/h :-) (should grow farther with progress)

cu,
Michael

Good morning!

I think Annika was referring to her Venice box.

Sooooo, any further results so far????

I must admit I didn't know about this Intel "policy". It's a real shame.
AFAIK the Visual C(++) compiler was used to build the app, but it seems like the math library linked to the app is from intel.

CU

BRM

Good morning too(didn't sleep yet) ;-)

Yeah the trick is great for AMD/Win and hopefully brings some speedup to Anika too!

My WU was at 33,33% after 5:40:57 h ~ 23,96 c/h and growing!
Remember this host got ~14 c/h befor this patch.

I'm very sure, not even Akos can top this result - change two bytes and get more
than 50% speed increase. :-)

Meanwhile I've had some first looks in my database and one thing I can already tell: The Win-app is slower than the Linux app. My X2@2,63 GHz will end up at about 28 c/h if my extrapolations are correct and this would be faster than under Linux.
My Athlon XPs running at about 2,2GHz(Linux) outperform every P4 even with 3,4GHz.
So it will be very interesting, how this patch will affect the performance of P4s and Conroes.

cu,
Michael

Bikeman (Heinz-Bernd Eggenstein)
Bikeman (Heinz-...
Moderator
Joined: 28 Aug 06
Posts: 3522
Credit: 739084833
RAC: 1259840

I find it always hard to

Message 37843 in response to message 37842

I find it always hard to compare the P4 results, as you never know whether hyperthreading is enabled. Most Credits per hour figures for P4s look lame, but for those running hyperthreaded doing two units on one physical core simultaneously, it would be fair to multiply those figures by 2 when you want to compare physical cores for AMD and Intel.

Well, those performance figures look darn impressive for the patch so far. As I said earlier, this is the funniest performance tuning I've ever seen !!!

Quote:

So it will be very interesting, how this patch will affect the performance of P4s and Conroes.

You mean the ranking of the P4s an Conroes relative to the AMDs under windows? because Intel CPUs won't benefit from that modification at all, just wanted to stress this.

CU

BRM

M. Schmitt
M. Schmitt
Joined: 27 Jun 05
Posts: 478
Credit: 15872262
RAC: 0

Argh, I made a mistake

Argh, I made a mistake looking up the credits my wingmen got for this type of WU. It's 318,02 instead of 405!
So this still makes up a performance boost, but so far only up to ~ 18,65 c/h, clearly slower than the Linux app. It might end up with ~21.5 c/h, we'll see.

cu,
Michael

Bikeman (Heinz-Bernd Eggenstein)
Bikeman (Heinz-...
Moderator
Joined: 28 Aug 06
Posts: 3522
Credit: 739084833
RAC: 1259840

RE: Argh, I made a mistake

Message 37845 in response to message 37844

Quote:

Argh, I made a mistake looking up the credits my wingmen got for this type of WU. It's 318,02 instead of 405!
So this still makes up a performance boost, but so far only up to ~ 18,65 c/h, clearly slower than the Linux app. It might end up with ~21.5 c/h, we'll see.

cu,
Michael


Every little bit helps. From what I've seen, one should not expect more than 30 % performance improvement from this trick.

CU

BRM

M. Schmitt
M. Schmitt
Joined: 27 Jun 05
Posts: 478
Credit: 15872262
RAC: 0

RE: I find it always hard

Message 37846 in response to message 37843

Quote:
I find it always hard to compare the P4 results, as you never know whether hyperthreading is enabled. Most Credits per hour figures for P4s look lame, but for those running hyperthreaded doing two units on one physical core simultaneously, it would be fair to multiply those figures by 2 when you want to compare physical cores for AMD and Intel.


I can clearly tell by the results if HT is enabled or not and there are Pentium D as well that have a pretty poor performance.

Quote:
Well, those performance figures look darn impressive for the patch so far. As I said earlier, this is the funniest performance tuning I've ever seen !!!


Well my fault. :(
I guess I was a little bit too tired.

Quote:

Quote:

So it will be very interesting, how this patch will affect the performance of P4s and Conroes.

You mean the ranking of the P4s an Conroes relative to the AMDs under windows? because Intel CPUs won't benefit from that modification at all, just wanted to stress this.

CU

BRM


You're right, it doesn't affect these cpus, and I don't know at the moment, if Athlon XPs are also affected by the performance gap.

cu,
Micha, going to rest. ;)

Comment viewing options

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