.5 Multipliers DO EXIST! Proof Inside.

[CaBoOse999]

I have heard alot of talk that .5 multipliers don't work because it isn't supposedly supported by AMD and many hardware monitoring programs did not recognize them. Some people have said that in reality what it does is set the multiplier to the next whole number and changed the fsb accordingly. What this would do in affect is lower ram speeds. I decided to do a very basic test to see if this was true. I did not go very in depth with my testing because I don't have many good cpu benchmarking programs. I used the Counterstrike: Source video stress test. A memory divider of 5:6 was used.

Test #1

In this test I used a half multiplier
8.5 x 306 = 2601 Mhz
The results were: 130.17 fps

Test #2

The theory against the half multiplier was that it moved to the next whole number and changed the fsb. So in theory running at 9 x 289 should yield the same results.

9 x 289 = 2601 Mhz
The results were: 124.58 fps

This shows a major performance loss from the previous test. This disproves the theory of the mutiplier changing to the next highest whole number and lowering the fsb.

Test #3

325 x 8 = 2600 Mhz

My computer would not boot. The memory would not run at 270Mhz. This eliminates the possiblity that the multiplier was moved to the next lowest whole number.

I think that this clearly demonstrates that the .5 multipliers do in fact exist. I realize I did not use the best program for this test and did not go into a whole lot of detail but I feel that is more than enough proof to prove my point.

 

[brazzmunk]

http://forums.extremeoverclocking.com/showthread.php?t=149717

 

[jstarman]

brazzmunk,

You state in your guide:

GENERAL RULES ON A64:
Remember, never use .5 multis.

but don't explain why. I'm using 9.5 right now 260x9.5 on my winie 3200 because well, the number just worked out with all of my stability testing. So why should I change to a 9 or 10.

Thanks.

 

[brazzmunk]

you know what.. i figured it would explain why you should use .5 multi but i accidently pasted the wrong link.. and now i can't find the original... but i stay with you on that .5 multi's are of better performance... i had mine running at 8.5*300 and it was faster than 9*285

 

[GuitarDaddy]

Testing a CPU concept based on FPS? Game benchmarks are highly GPU dependent and the difference you show could easily be explained by normal fluctuations in scores (you can run the same bench several times and get slightly different scores each time)

Test your theory using CPU dependent benches like SuperPI or Sandra Arithmatic CPU.

And you should show memory speed per CPUZ/Clockgen or other utility and list the ram frequency and divider that you used. It is what .5 multis do to memory freqency that prompt many to beleive half multis don't exist.

 

[Shimmishim]

for those of you who don't know what he's talking about...

8.5 x 306 = 2601 mhz

running 5:6 ratio

mem mult = 200*8.5/166 = 10.2 --> rounds up to 11

so mem speed is 8.5 x 306/11 = 236.45


9 x 289 = 2601

running 5:6 ratio

mem mult = 200*9/166 = 10.8 --> rounds up to 11

so mem speed is 9 x 289/11 = 236.45


so the fact that you get 130 for test #1 and 124 for test #2 is very interesting to say the least...


then again CodeRed over at Xtremesystems makes a good point to in this thread

http://www.xtremesystems.org/forums/showthread.php?p=614771#post614771

he shows how he gets similar results from the .5 mult and the 1 up from it at the same clockspeed that is...

to be honest, i duno what to believe.... i want to say i believe codered because he's well respected and is the author of A64 tweaker but your results are interesting as well...

were you sure no variables were changed between tests?

possible to do more tests?

bandwidth, pifast, others?

cuz if what you're saying is true 8.5 x 306 should be faster than 9 x 289 right (speculating from your CSS test)?

i wanna see some more tests then you can go post over at xtreme your findings

 

[yliu]

Although the CPU and memory were using the same speed in both tests, the higher FSB used in the x8.5 test could increase performance slightly because the rest of the motherboard is running at a 5.8% faster rate. The HTT bus speed would be 918mhz for x8.5 (assuming x3 multi, as a x4 would be crazy) and 867 for x9.

Another possible reason for the performance difference is that the test used was mostly GPU bound. Should retest with superPI.

Finally, the only bad thing I've heard about .5 multis is that they may cause certain software utilities to report the wrong memory speed (such as CPU-Z) and/or causes stability problems on some motherboard/CPU combinations. The later is not proven, of course, and as long as you know how the multi and the memory divider works and you are sure about the speed of your memory clock, feel free to use half multies if they work and improves your system performance (because of the higher FSB)

 

[CaBoOse999]

I looked into this a bit more in depth. Yes the .5 multipliers "EXIST" but in essence what it does is use a different memory divider. This means that the ram frequency reported by cpuz is correct. The reason for the performance difference is the combination of a higher HTT and HT frequency. The ram was actually running at a faster speed at 289 x 9 but it was overshadowed by everything else. I should have probably researched the topic a bit more before I came to my own conclusion but it was interesting none the less.

EDIT: I was actually wrong again there. It just moves it up to the next multiplier. I have no idea why those results came out why they did. There's alot more too it then what I know now, so I'm just going to stick with whole dividers.

 

[GML3G0]

so basically, what you're saying is, the 0.5 multipliers do work for CPU speed, but they work differently for the memory controller because it rounds the multiplier value up. So let's say I am at 264 x 9.5 and the Memory divider is the 366 (166 on other boards, just labeled differently on mine) divider, then what speed would my memory be at, because CPU-Z shows it only at 209MHz.

EDIT: Never mind, CPU-Z is right, I used the forumula on the link to the XtremeSystems Forums and confirmed the numbers.

 

[CaBoOse999]

Wow wasn't expecting to see this retarded topic to come back to haunt me. It rounds the multiplier up and lowers your fsb.

 

[GML3G0]

i have so many contradicting reports... some say it works, others say it doesn't, I don't know what to believe.

 

[CKTurbo128]

Would this also hold true for Athlon XP CPUs as well? Is an integer multipler more stable than a decimal (0.5) multiplier?

 

[GuitarDaddy]

The absolute proof is a highly guarded secret by mobo makers

But the circumstancial evidence overwhelmingly supports the theory that 1/2 multi's Don'texist. Like CaBoOse999 said "it rounds the multiplier up and lowers your fsb"

Apparently the half multi's are a creation of the bios, designed by mobo makers. In my testing there seems to be no performance icrease or decrease associated with 1/2 multis. I can hit the same max O/C with a half or whole multi.

The one thing I have learned. Don't use A64 tweakers custom memory dividers with 1/2 multi Evidently the dividers screw up the bios's treatment of the 1/2 muli, Both times I tried this I got UGLY:shocked: bsod crashes.

*edit*
this only applies to A64's

 

[CaBoOse999]

I'm pretty sure this doesn't apply for the xp's. My old XP1800+ had a .5 multi by default.

 

[Shimmishim]

you are right caboose...

this only applies to the A64's... not the XP's