Half multipliers

[Guest Spartacus]

I know the story is that half multipliers are not supposed to be good on A64 rigs due to the RAM clock being auto set at a fraction of the "FSB" clock.

 

I'm not seeing that though on any diags or benches I run.

 

I get better bench numbers (3DM, Sandra) at 253x9.5 than I do at 240x10. But that shouldn't happen right?

 

Also, CPU-Z shows the RAM clock to be 253 (1:1) when the multiplier is set for 9.5.

 

Is CPU-Z wrong?

 

If so, what diags will show this effect accurately?

 

 

:confused:

[jess1313]

Originally posted by Spartacus

I know the story is that half multipliers are not supposed to be good on A64 rigs due to the RAM clock being auto set at a fraction of the "FSB" clock.

 

I'm not seeing that though on any diags or benches I run.

 

I get better bench numbers (3DM, Sandra) at 253x9.5 than I do at 240x10. But that shouldn't happen right?

 

Also, CPU-Z shows the RAM clock to be 253 (1:1) when the multiplier is set for 9.5.

 

Is CPU-Z wrong?

 

If so, what diags will show this effect accurately?

 

 

:confused:

 

hey this is strange that your getting better results at 253x9.5, that you are at 10x240.

 

what HTT multi are you using?

 

also, that really might be 253.8x9.5, so to get the exact same results, you might need 241x10.

 

CPU-z is WRONGE

 

GL

-JEss-

 

could you post exact results?

[Guest Spartacus]

Yup did some more testing and the minor speed increase I was seeing was due to the CPU speed advantage of 253x9.5 vs. the slightly slower CPU final speed at 240x10. Got really nothing at all from the FSB increase. The numbers are all too close for any good comparisons at these particular speeds anyway, all pretty much within the margin of error.

 

I've seen enough to convince me that the half multipliers do appear to penalize the RAM speed in some way. The higher FSB scores should have been a bit higher than they were.

 

Tried 252x9.5 vs. 240x10 and got almost identical scores.

 

[jess1313]

Originally posted by Spartacus

Yup did some more testing and the minor speed increase I was seeing was due to the CPU speed advantage of 253x9.5 vs. the slightly slower CPU final speed at 240x10. Got really nothing at all from the FSB increase. The numbers are all too close for any good comparisons at these particular speeds anyway, all pretty much within the margin of error.

 

I've seen enough to convince me that the half multipliers do appear to penalize the RAM speed in some way. The higher FSB scores should have been a bit higher than they were.

 

Tried 252x9.5 vs. 240x10 and got almost identical scores.

 

 

yes, if your set to 1:1 in bios half multies slow down your ram consideraly

 

these are the ram ratios that you get, when using half multies and 1:1 in bios....

 

10.5x= "10.5:11"

 

9.5x= " 9.5:10"

 

8.5x= " 8.5: 9"

 

7.5x= "7.5: 8"

 

6.5x= "6.5: 7"

 

 

etc.

 

-Jess-