Do CPUs favor certain multis?

Y

YARDofSTUF

[H]ard|Gawd
Joined
Jun 19, 2001
Messages
1,469
It seems like my 4400 doesnt like the 9.0 and 10.0 multipliers.

At 8.5 I can crank the fsb to 300 and at 9.5 I have the it at 285. But anything over 240 using the 9 or 10 multi and the system will freeze.
 
{NcsO}ReichstaG said:
could it be because your increasing the CPU speed when raising the multiplier and the FSB? ;)
Click to expand...
no because 9.5x285 is higher than 240x9. Someone needs to pull a Billy Madison and head back to elementary school. :D
 
I don't know about them favoring certain integers over others, but I DO know that the A64's by FAR prefer whole number multi's as opposed to .5x multis. I know eclipse has touched on this topic before...suffice it to say, the frequency of the memory controller gets screwy at .5x multis. That's why I always laughed back in the day when people complained about the K8N Neo Plat not properly doing .5x multis...
 
Eva_Unit_0 said:
I don't know about them favoring certain integers over others, but I DO know that the A64's by FAR prefer whole number multi's as opposed to .5x multis. I know eclipse has touched on this topic before...suffice it to say, the frequency of the memory controller gets screwy at .5x multis. That's why I always laughed back in the day when people complained about the K8N Neo Plat not properly doing .5x multis...
Click to expand...


Weird, was the A64 a single core that eclipse talked about?
 
YARDofSTUF said:
Weird, was the A64 a single core that eclipse talked about?
Click to expand...

doesn't matter...it universally applies to all A64's. Basically, the memory speed is always derived from a integer multiplier, even if a .5x multi is used. Remember that the 1:1 ratio on A64's really means [overall clock]/[multiplier]. Well let's say that you're running 1:1 at 9.5 x 285 (2707 mhz) Now normally you'd assume that since you're running 1:1 the ram speed would also be 285. But it's not. It rounds up to 10x for the multi and derives the mem speed as 2707/10 = 270mhz. So if you're running 1:1 at a .5x multi you will APPEAR to have more headroom before a divider is needed, but in reality it's just the mem controller playing games.
 
I agree, stay away from .5x multipliers if you can...and other then that, a higher muliplier with a lower HTT as opposed to a low muliplier and high HTT makes things a LOT easier on the CPU's memory controller, and will also allow you to tighten up your memory timings a fair bit. If you've got some nice TCCD(5) based ram that you KNOW can do something like 300Mhz 1:1 and your chip has a strong enough memory controller to do 300HTT 1:1 then I'd go for it...but lowering your CPU multiplier lowers your memory bandwidth (even though the ram speed stays the same) along with some aspects of the CPU's performance.

There is a line here though, comparing something like 250x9 with 225x10 on a 3200+ Venice for example...I'd take 225x10 every single time. Even if both setups were using 2x512MB of Mushkin PC4000 Redline and both setups had memory timings of 2-2-2-5...the 10x multiplier would still show a performance advantage. However, comparing results between 267x9 and 200x12 isn't quite the same...both examples result in a 2.4GHz clock speed, but once the setup with the lower multiplier has enough of an HTT advantage, the results start swinging the other way.

Hope I could help, if there's clarification needed on anything let me know.
 
ya I saw the the memory divider never never matched my math, but I cant ignore the .5 multis.

240x11 is stable and the highest I can get on the even multis. but I can get 285x9.5 with the .5 multi

I lose scores in sandra but I gain some points in 3d mark05 and cut my super Pi time down too.

Thanks for the info though.
 
are your results effected when you drop the memory divider down to 166mhz?
 
Yeah .5x multis are no good

I notice many CPUs (mostly 1mb like SD core and Toledo core) prefer certain multis. I know many people running 3700+s who have had bad results at 11x multi and have had to use 10x instead. Seems like certain CPUs do like certain multis.
 
(cf)Eclipse said:
are your results effected when you drop the memory divider down to 166mhz?
Click to expand...


When I get up my 300 I drop it to the 166 divider. at 285 I have it at the 180, cpuz says its 246mhz. So for 270x10 you think I should try the 166 divider?
 
I am seeing something similar to you; my Opty 146 will run up to 316 on a 9.5 multi, but no farther than 288 on a 10. If you know your actual RAM speed, and don't mind it, I don't see what the problem is with a .5 multi- it's working allright for me.
 
YARDofSTUF said:
ya I saw the the memory divider never never matched my math, but I cant ignore the .5 multis.

240x11 is stable and the highest I can get on the even multis. but I can get 285x9.5 with the .5 multi

I lose scores in sandra but I gain some points in 3d mark05 and cut my super Pi time down too.

Thanks for the info though.
Click to expand...

Interesting...got any before and after shots comparing them? IMO if 3Dmark, SuperPi, and (especially) games show gains as opposed to losses when switching to .5 multipliers then w/e.

So, is the only difference between a whole and .5 multiplier that the memory speed is shown incorrectly? It's not hard to figure it out yourself, so if that's the only thing "wrong" with it...I don't really see what the big deal is...interesting...
 
Eva_Unit_0 said:
That's why I always laughed back in the day when people complained about the K8N Neo Plat not properly doing .5x multis...
Click to expand...


I'm not sure if it just applied to the k8n neo plat, but that particular MSI board emulated .5x multis by upping the memory freq. Maybe some other companies are doing it too :(
 
Talonz said:
I'm not sure if it just applied to the k8n neo plat, but that particular MSI board emulated .5x multis by upping the memory freq. Maybe some other companies are doing it too :(
Click to expand...

I had heard that quite awhile ago, which is kinda why I just dis-regarded using them so early on. Why I decide to start questioning it now I have no idea.
 
the .5 multis arent that bad.. you just have to account for the memory divider rounding up. graba copy of A64 Info and use the calculator tab. it makes it all easy as pie. see sig...... :p
 
cornelious0_0 said:
Interesting...got any before and after shots comparing them? IMO if 3Dmark, SuperPi, and (especially) games show gains as opposed to losses when switching to .5 multipliers then w/e.

So, is the only difference between a whole and .5 multiplier that the memory speed is shown incorrectly? It's not hard to figure it out yourself, so if that's the only thing "wrong" with it...I don't really see what the big deal is...interesting...
Click to expand...


Havent been saving benchmarks much, what I do have I dont know what the settings were lol
 
You must log in or register to reply here.
Back
Top