Overclocking - which is better and why?

[Tiamat]

Say you have a Athlon XP system on a Nforce 2 Mobo.
Which is better? 200x11 or 210x10? (Assume complete stability at both conditions)
Even tho you lose a net of 100 mhz, you gain in cache speeds. But what does this loss and gain really mean?
Are you processing less per unit time, but processing larger chunks of data per process? Where is the balance? and is there a point where you can gain better performance due to faster processor cache while swallowing a somewhat slower processor speed?

 

[Nohr]

I'd go with 10x210. The higher FSB speed will make up for the 100MHz CPU loss. It's not just the cache that's faster, it's the RAM as well.

 

[Lyfer]

Higher FSB = more bandwidth.

 

[TheCanuck]

I'd actually go with the 200x11. Performance is going to be slightly higher since bandwidth can only help so much. I did a linear regression on the Bartons wrt clock speed / FSB a while ago and here's the calculation to get the PR rating:

PR = 1.401*mhz+2.834*FSB-1016 ; where mhz = 1800 - 3000, and FSB = 300 - 600

So for 200x11 (i.e. 400FSB) you get a PR of 3200
and for 210x10 (i.e. 420FSB) you get a PR of 3116 (to get a PR of 3200 at 2100Mhz you'd need an FSB of 450)

 

[Jeff7181]

Originally posted by: TheCanuck
I'd actually go with the 200x11. Performance is going to be slightly higher since bandwidth can only help so much. I did a linear regression on the Bartons wrt clock speed / FSB a while ago and here's the calculation to get the PR rating:

PR = 1.401*mhz+2.834*FSB-1016 ; where mhz = 1800 - 3000, and FSB = 300 - 600

So for 200x11 (i.e. 400FSB) you get a PR of 3200
and for 210x10 (i.e. 420FSB) you get a PR of 3116 (to get a PR of 3200 at 2100Mhz you'd need an FSB of 450)

Would be cool if you could create that graph again and take a screen shot of it.

 

[BlueWeasel]

Originally posted by: Jeff7181

Originally posted by: TheCanuck
I'd actually go with the 200x11. Performance is going to be slightly higher since bandwidth can only help so much. I did a linear regression on the Bartons wrt clock speed / FSB a while ago and here's the calculation to get the PR rating:

PR = 1.401*mhz+2.834*FSB-1016 ; where mhz = 1800 - 3000, and FSB = 300 - 600

So for 200x11 (i.e. 400FSB) you get a PR of 3200
and for 210x10 (i.e. 420FSB) you get a PR of 3116 (to get a PR of 3200 at 2100Mhz you'd need an FSB of 450)

Would be cool if you could create that graph again and take a screen shot of it.

I'd love to see it as well. If it's an excel spreadsheet, would you consider emailing to some of us?

 

[TheCanuck]

Sure no problem. Equation has changed a little now that I've added the 2600+ info, but here's the link to the file:

Barton Regression Link

This is the info you're really concerned with:


Coefficients
Intercept -1051.853627
X Variable 1 1.418149373
X Variable 2 2.829812518

So PR = 1.418*Mhz + 2.83*FSB - 1052

To do it in Excel you click on Tools-->Data Analysis-->Regression, select your Y inputs and then your X Inputs (which equals the last two columns -- mhz & FSB), then select what plots, residuals etc you want and hit ok.

Who would've thought math is useful!

 

[Fern]

A little help for the mathematically challenged please.

I'm not familiar with "PR", but since 11 x 200 =3200+ speeds, I'm assuming it means "performance rating". I.e., how AMD arrives at their Intel-type equivilency.

So: PR = 1.401*mhz+2.834*FSB-1016 ; where mhz = 1800 - 3000, and FSB = 300 - 600

In my case, my last OC attempt I got to FSB of 192. Help me out cuz I can't come up with a reasonable number. Viz:

1.401 x 2112+2.834 x 384-1016

2112 = 192 fsb x 11 multi
384= 192 x 2

So:

1.401 x (2112+2.834) x (384-1016)=

1.401 x 2114.834 x -632= giberish

Can someone help me out with a calc for FSB of 192

Thanks

 

[TheCanuck]

You're calculating it wrong. This is how you do it:

Performance Rating = (1.401*2112) + (2.834*384) - 1016
so, PR = 3031+

Or 3030+ if you use the updated numbers from my last post. Not that it really makes much difference! lol

Remember, always do multiplication and division first, then perform addition and subtraction.

 

[BlueWeasel]

Originally posted by: Fern
A little help for the mathematically challenged please.

I'm not familiar with "PR", but since 11 x 200 =3200+ speeds, I'm assuming it means "performance rating". I.e., how AMD arrives at their Intel-type equivilency.

So: PR = 1.401*mhz+2.834*FSB-1016 ; where mhz = 1800 - 3000, and FSB = 300 - 600

In my case, my last OC attempt I got to FSB of 192. Help me out cuz I can't come up with a reasonable number. Viz:

1.401 x 2112+2.834 x 384-1016

2112 = 192 fsb x 11 multi
384= 192 x 2

So:

1.401 x (2112+2.834) x (384-1016)=

1.401 x 2114.834 x -632= giberish

Can someone help me out with a calc for FSB of 192

Thanks

You've got the order of operations wrong.

Using the latest equation, it should be this: (1.418 x 2112) + (2.83 x 384) - 1052 = 3029

Essentially, running at that speed and FSB, you've got a 3000+.

 

[Tiamat]

Originally posted by: TheCanuck
Sure no problem. Equation has changed a little now that I've added the 2600+ info, but here's the link to the file:

Barton Regression Link

This is the info you're really concerned with:


Coefficients
Intercept -1051.853627
X Variable 1 1.418149373
X Variable 2 2.829812518

So PR = 1.418*Mhz + 2.83*FSB - 1052

To do it in Excel you click on Tools-->Data Analysis-->Regression, select your Y inputs and then your X Inputs (which equals the last two columns -- mhz & FSB), then select what plots, residuals etc you want and hit ok.

Who would've thought math is useful!

Nice, How do you figure in Cache Size. Since i have a Tbred-B. it would be awesome if i could get a PR rating for 11x200 for Tbred-B (I think its like 2900+ since its not a Barton)

 

[RobCur]

what about 225x10 2250mhz vs 210x11 2310mhz? now that's a hard one to tell!

 

[BlueWeasel]

Originally posted by: Tiamat
Nice, How do you figure in Cache Size. Since i have a Tbred-B. it would be awesome if i could get a PR rating for 11x200 for Tbred-B (I think its like 2900+ since its not a Barton)

IMHO, this equation should also be valid for T-breds running at the same FSB speeds. While the Barton has double the L2 cache of a TB, some tests have shown that a Barton running at 11x200 and a TB at 11x200 benchmark about the same.

 

[Falloutboy]

Originally posted by: Tiamat
Say you have a Athlon XP system on a Nforce 2 Mobo.
Which is better? 200x11 or 210x10? (Assume complete stability at both conditions)
Even tho you lose a net of 100 mhz, you gain in cache speeds. But what does this loss and gain really mean?
Are you processing less per unit time, but processing larger chunks of data per process? Where is the balance? and is there a point where you can gain better performance due to faster processor cache while swallowing a somewhat slower processor speed?

if the chip is stable at 200x10 and the memory can run at 210 fine. run the chip at 10.5x210 this yeilds 2.205

5 mhz shouldn't make a chip unstable.

 

[Tiamat]

Thanks for your responses. I upped my mushkin PC3500 Level 2 to 1.9V and am stable at 214*10.5
So, i basically kept my processor the same while upping my system speed

 

[Fern]

"Remember, always do multiplication and division first, then perform addition and subtraction. "

Thanks for help. I've been curious how to calculate this for some time now.

Fern

 

[magratton]

Trying to understand... Your RAM is at 1.9v or your CPU? I would have expected that RAM to be at 2.9v.