akosf C37 performance observations thread

Akos Fekete
Akos Fekete
Joined: 13 Nov 05
Posts: 561
Credit: 4527270
RAC: 0

RE: My question was on your

Message 25986 in response to message 25984

Quote:
My question was on your (personal) recomandation.

Interesting question. I'm thinking on it...
Why do you ask?

Akos Fekete
Akos Fekete
Joined: 13 Nov 05
Posts: 561
Credit: 4527270
RAC: 0

RE: So far, all my WUs have

Message 25987 in response to message 25985

Quote:
So far, all my WUs have validated with C37...

Good news.

Quote:
I'm wondering when S38 will be posted for us to try? Perhaps to the d/l thread?

I will make a link in the d/l thread after my test in some hours or days.

Quote:
Again, thx for the binaries :)

Let's crunch! :)

LiborA
LiborA
Joined: 8 Dec 05
Posts: 74
Credit: 337135
RAC: 0

RE: RE: My question was

Message 25988 in response to message 25986

Quote:
Quote:
My question was on your (personal) recomandation.
Interesting question. I'm thinking on it...
Why do you ask?

Why? That's very easy:
What effect have WU's with uncorect results? I thing that this effect is very low - I can say it's negative. In this case it may be better use original binary or some other from your binaries without this problem.
Is this problem only by C37 or the same code (which caused the differencies from standard binary) is used in all your binaries?

LiborA
LiborA
Joined: 8 Dec 05
Posts: 74
Credit: 337135
RAC: 0

I have a one question again:

I have a one question again: What say for your optimalization (and its problem) peoples from projects as Bruce atc.?

Akos Fekete
Akos Fekete
Joined: 13 Nov 05
Posts: 561
Credit: 4527270
RAC: 0

RE: RE: RE: My question

Message 25990 in response to message 25988

Quote:
Quote:
Quote:
My question was on your (personal) recomandation.
Interesting question. I'm thinking on it...
Why do you ask?
Why? That's very easy:
What effect have WU's with uncorect results? I thing that this effect is very low - I can say it's negative. In this case it may be better use original binary or some other from your binaries without this problem.
Is this problem only by C37 or the same code (which caused the differencies from standard binary) is used in all your binaries?


Well, I did these unofficial codes because of fun, but I shared at the reques of some people.
I never said that you should use it, and I don't say that.
So, everybody got a chance to open Pandora's box. "Use at you own risk."
I don't want to guarantee anything, because I cannot do that.

Michael Roycraft
Michael Roycraft
Joined: 10 Mar 05
Posts: 846
Credit: 157718
RAC: 0

RE: ... Well, I did these

Message 25991 in response to message 25990

Quote:
... Well, I did these unofficial codes because of fun, but I shared at the reques of some people.
I never said that you should use it, and I don't say that.
So, everybody got a chance to open Pandora's box. "Use at you own risk."
I don't want to guarantee anything, because I cannot do that.

Akosf,

Don't worry about any guarantee - leave all that worry to the project owners and their choice(s) for "official" apps, where it rightfully should lie.

If you can continue to have fun at this, please, maintain that beautiful frame of mind. We are very grateful for all that you have done.

Michael R.

microcraft
"The arc of history is long, but it bends toward justice" - MLK

Alinator
Alinator
Joined: 8 May 05
Posts: 927
Credit: 9352143
RAC: 0

RE: Thanks for these

Message 25992 in response to message 25981

Quote:

Thanks for these infos!
I hoped that C37 will be much faster on your K6 machines...

I tried to find the reason of the validation problem.
I know that the precision of C37 has to be very near to the original.
I made a code that gave poor results, but the validator accepted them!
So, I examined the codes again and I saw there is a critical point.
Sometimes the exponent of the number format is not enough.
Probably this is the reason of the validation problem of SSE based linux application too.

edit:
exponent size of SSE -> 8 bit
exponent size of SSE2 -> 11 bit
exponent size of FPU -> 15 bit

I figured the cause of the failures was something like that, most likely in combination with a fairly tight tolerance band at the project validator end of things.

FWIW, it looked like C37 was somewhat faster than 387 on the K6-2, but that's hard to say conclusively with just a handful of results pushed through so far.

I just took a closer look at the K6-2's results, and it looks like one of the failures might be due to the project database work last week rather than a calculational precision problem from the app. In any event, I'm going to run a few more C37's through them to make for a better test.

Alinator

anders n
anders n
Joined: 29 Aug 05
Posts: 123
Credit: 1656300
RAC: 0

I have tried A36, C37 and

I have tried A36, C37 and S37a and with no problems with any of them.

Anders n

caferace
caferace
Joined: 21 Oct 05
Posts: 4
Credit: 286637
RAC: 0

RE: Well, I did these

Message 25994 in response to message 25990

Quote:
Well, I did these unofficial codes because of fun, but I shared at the reques of some people.
I never said that you should use it, and I don't say that.
So, everybody got a chance to open Pandora's box. "Use at you own risk."
I don't want to guarantee anything, because I cannot do that.

Akosf, you need not apologize, guarantee, nor make any amends for your work. What you have done is excellent, a grand stroke not only for Science but also for the people who just like crunching. If you were local to me, I'd buy you a large selection of the the beverage of your choice.

cheers,

-jim

rbpeake
rbpeake
Joined: 18 Jan 05
Posts: 266
Credit: 968093708
RAC: 1197619

RE: What you have done is

Message 25995 in response to message 25994

Quote:

What you have done is excellent, a grand stroke not only for Science but also for the people who just like crunching. If you were local to me, I'd buy you a large selection of the the beverage of your choice.
cheers,

-jim


Me too! You are a very gifted individual, and we appreciate that you are sharing the fruits of those gifts with all of us! :)

Comment viewing options

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