Stress on MCH vs. CPU Multi

[zipz0p]

At a given FSB frequency, does increasing the CPU multiplier (and thus the CPU frequency) increase stress on the MCH (thus requiring more voltage to the MCH for stability)?

The specific issue is: At an FSB of 410MHz, my Q9450 has no issues with a 7x multi, but has a great deal of trouble with an 8x multi. It could be that I need to increase my CPU voltage more, but I'm curious about the answer to the question above. Any feedback is appreciated.

 

[Lupiron]

I don't think that it would have too much to do with the simple act of swapping Multipliers. Once a FSB is stable, then it is stable, and the CPU speed has very little to do with the rest of the chipset.

I suppose it is possible, to a small degree, but more than likely its due to VCore, or another CPU related issue.

Can you run either Real Temp or Core temp and list your VID, please?

--Lupi

 

[zipz0p]

No worries, I backed down to 400 and I'm going 1:1 with memory at tighter timings. I'll see if I can squeeze in higher frequency or tighter tRD (currently tRD is at 7) later.

I'm really interested in how the multiplier actually works. If it requires more power for a higher multiplier, or puts more stress on the base FSB frequency generator (whatever that is?), then I would think it would increase the stress on the MCH, slowly decreasing the returns of higher multipliers. But that's pure speculation from my end... it'd be interesting to understand how the chip works.

In case you're very interested, Core Temp lists my VID as 1.2375V (Q9450). I'm not sure how that stacks up to other 45nm quads, but I hope it's good I've raised my Vcore in bios from 1.2375 to 1.2560 to keep it stable at 3.2GHz.

That doesn't seem altogether too bad - only a 1.5% increase in voltage for a 20% frequency increase (though I realize power consumption will go like V^2).

 

[Lupiron]

Hehe, well, you have to think about intels picture. They give you more than enough voltage at stock to get like 300-500 "Free" Mhz in speed, because they have their own buffers, to make sure there are zero problems in 3 years at least at stock, they make sure they can run faster than 2.4 Ghz, and with the same voltage.

Basically what you do when you get the free Mhz is take away intels headroom they have set aside to be sure they don't have problems at such a slow speed of 2.4 Ghz.

--Lupi

 

[zipz0p]

2.4 GHz is the Q6600, I'm running the Q9450 - which is 2.66 GHz stock Other than that, yeah, that seems to make sense... I'll have to see what I can find on the actual structure of Intel's northbridge chips to see what I can discover about how the multipliers actually work.

 

[Lupiron]

As far as I know, the Multiplier does nothing but change the CPU speed. So I highly doubt that it puts much more of a strain on the system.

I am sure it puts a MINOR strain, because the Data has to pass through somewhere at a higher speed, right?

But as far as anything else, like needing more voltage, I don't believe so.

--Lupi

 

[zipz0p]

Well, I've been reading up (that is, going through some Wikipedia articles) on some circuitry elements, quartz clocks and frequency multipliers.

A couple of interesting observations:
multipliers are simply taking higher harmonics of the base frequency. Sometimes this is done in series in order to get reasonably high multipliers. Successively higher multipliers will, of course, have lower amplitude signals, so they must be amplified. Still, if you're driving it VERY VERY FAST, this could degrade the chip's ability to distinguish individual wave crests.

So essentially, if you're using a higher multiplier, it may require significantly higher amplification of the signal compared to a lower multiplier, and this may degrade the signal, particularly at higher base frequencies.

Also, when it comes to the resonators that drive these frequencies, they are temperature-dependent! That is, the actual frequency output degrades as temperature drifts from a specific temperature. This means that at very high temperatures and very low temperatures, the frequency of components will actually decrease (you probably won't notice this, because it is the very reference frequencies that are doing the drifting!). I think this effect is probably extremely slow, but it's interesting to know...

 

[zipz0p]

Oh, and another thing, some motherboards have "PLL Voltage" settings, right? Well, I figured out that the PLL is a Phase-locked loop. The PLL generates the base FSB frequency, and at higher frequencies, again, this will need a higher voltage in order for independent wave peaks to be distinguishable.

 

[Lupiron]

Thats correct! Phase lock loop keeps the voltage and signals straight, by feeding the same voltage back into itself, which allows it to correct for any errors, because it knows what it should always be receiving!

And yes, it can help with stabilizing the FSB because it is a signal purifier of sorts! Neat that your comp has a 14 Mhz timer.

--Lupi