What does the Vcore vs. Frequency plot look like?

[Hulk]

I am wondering how Vcore and clock frequency scale for cpu's in general? I realize that each chip will have different max frequency.

But if you charted max frequency vs. Vcore what would the resulting graph look like?

I'm wondering if it would be like this. At first small increases in voltage producing large increases in frequency, then as the max frequency is approached larger voltage increases would only provide small increases in frequency.

Just curious if anyone knows anything about this relationship.

 

[Henny]

I actually did a plot for C2D and it came out fairly linear in the range of Vcore values from stock to 1.5V.

 

[Hulk]

Interesting. For some reason I was thinking Vcore would shoot through the roof as max frequency was reached.

 

[f4phantom2500]

it depends on the chip. some chips go really far on little if any extra vcore but get to a point where theyd need a lot of extra voltage for a negligible clock increase. others might reach their max frequency on stock voltage and the extra voltage would only make it hotter. in general, i would say it's not linear.

 

[RichUK]

The relationship is most definitely nonlinear.

 

[EarthwormJim]

It entirely depends on the your individual chip. For my Opteron, there is no correlation between voltage and clock speed starting at stock vcore.

 

[Henny]

I have the graph in front of me.

It's indeed linear for the range of conditions I tested.

I took an E6600 and varied FSB to as high as it would go at stock voltage (ie 1.3V) before it became unstable. For my unit that was at about 330 FSB.

From that point on I gradually raised Vcore to get more speed and once again plotted the threshold of instablity. (Vcore vs. FSB) I did this for about 8 data points which ended at 380 FSB at 1.5V. The line from 330@1.3V to 380@1.5V is very linear.

If I had kept going beyond 1.5V the curve might have changed but I didn't want to risk my CPU. I do believe it would become non linear at some point well beyond where it likes to operate.

I suspect the overall curve would be the same for any C2D but the curve would shift higher or lower depending on the individual device.

 

[Idontcare]

You are talking about a shmoo plot...and there is very good fundamental device physics reasons why it will not be linear.

Idrive vs. Vcc is non-linear, is must be or else your xtor's are not behaving as amplifiers.

That is not to say that one cannot sample a select portion of the shmoo plot and come to the conclusion that the data are seemingly linear. This is true of any dataset, in the limit of narrowing the sample space you will always appear to have linear correlation in the data provided the underlying function has a continuous first and second derivative no matter how non-linear the data in "big picture" are.

For a real-life example of a shmoo plot, checkout slide 22 in this Niagara2 presentation from SUN: presentation

Ironically enough, the "shmoo" plot got its name from being named after a cartoon figure called a shmoo simply because the cartoon shmoo was curvy and non-linear.

 

[aka1nas]

Originally posted by: Henny
I have the graph in front of me.

It's indeed linear for the range of conditions I tested.

I took an E6600 and varied FSB to as high as it would go at stock voltage (ie 1.3V) before it became unstable. For my unit that was at about 330 FSB.

From that point on I gradually raised Vcore to get more speed and once again plotted the threshold of instablity. (Vcore vs. FSB) I did this for about 8 data points which ended at 380 FSB at 1.5V. The line from 330@1.3V to 380@1.5V is very linear.

If I had kept going beyond 1.5V the curve might have changed but I didn't want to risk my CPU. I do believe it would become non linear at some point well beyond where it likes to operate.

I suspect the overall curve would be the same for any C2D but the curve would shift higher or lower depending on the individual device.

That's mostly because you can't keep the chip cool enough to run it at the higher voltages that would be needed to squeeze more clock speed out of it.

 

[Hulk]

Originally posted by: Idontcare
You are talking about a shmoo plot...and there is very good fundamental device physics reasons why it will not be linear.

Idrive vs. Vcc is non-linear, is must be or else your xtor's are not behaving as amplifiers.

That is not to say that one cannot sample a select portion of the shmoo plot and come to the conclusion that the data are seemingly linear. This is true of any dataset, in the limit of narrowing the sample space you will always appear to have linear correlation in the data provided the underlying function has a continuous first and second derivative no matter how non-linear the data in "big picture" are.

For a real-life example of a shmoo plot, checkout slide 22 in this Niagara2 presentation from SUN: presentation

Ironically enough, the "shmoo" plot got its name from being named after a cartoon figure called a shmoo simply because the cartoon shmoo was curvy and non-linear.

Thanks. That's exactly what I was curious about. Slide 22 shows the plot as I thought it would look.