Calculate your A64 overclocking potential
You wonder what frequency your A64 can reach at a certain Vcore?
Well, I did, and tried to find a way to calculate it instead of doing the time consuming cycle of SuperPi/Prime95 after every voltage change over again. I will show you how to calculate and then afterwards why this calculation holds (roughly). edit-> PLEASE READ THOROUGHLY, it is not trivial! <-edit
1. Find the highest stable frequency "y" in Mhz, at a certain Vcore "x"
2. At some other "Vcore" the max. stable "Freq" is
Freq = y + (Vcore - x) * z (*)
z = 870 for Toledo (also dc-Opteron)
z = 860 for Manchester
z = 620(?) for SanDiego (also sc-Opteron)
z = 310(?) for Venice
z = 860 for Winchester
Notes:
a. Take the initial Vcore/Freq (x,y) point not too far from the point you aim for.
b. I only listed z values for 90mm A64 cores, as I did not include data from older cores like Clawhammer and NewCastle
c. The Venice and SD values for z seem too low to me. It was the most precise value I could obtain, but you might need a HIGHER value of about 800 (never tried it though). I expected all 90mm single-cores to behave about the same. A possible reason for Venice to be that low? (see below)
d. This relation should roughly hold if you manage to keep your CPU and MB cool and your PSU rails clean. Since the heat generated by the CPU grows linearly with the frequency and Vcore^2 and (*) is linear, it follows for the power of your PSU and the heat your CPU Cooler has to be able to carry away is:
Q ~ Freq^3 (**)
Here is a table for an imaginary setup with a Toledo. The first column contains the Vcore, the second the heat output relative to that at 1.4V, the third one contains the frequency that should be able to be reached about.
Vcore Heat freq_tol
[V] [%] [MHz]
---------------------------------
1.250 -28.8 2470.8
1.275 -24.5 2492.6
1.300 -19.9 2514.3
1.325 -15.2 2536.1
1.350 -10.3 2557.8
1.375 -5.3 2579.6
1.400 0.0 2601.3
1.425 5.5 2623.1
1.450 11.1 2644.8
1.475 16.9 2666.6
1.500 23.0 2688.3
1.525 29.2 2710.1
1.550 35.7 2731.8
1.575 42.4 2753.6
1.600 49.3 2775.3
1.625 56.4 2797.1
1.650 63.7 2818.8
--------------------------------------------------------------------------------------------------------------------------------
Why (*) should work for you?
All overclocking results in the databases on the internet give a nice sample that can be treated statistically.
I gathered the o/c results from these pages:
http://www.forumdeluxx.de/forum/showthread.php?t=204404
http://www.forumdeluxx.de/forum/show...teron+stepping
http://www.ocforums.com/showthread.p...08#post2987108
http://www.hwupgrade.it/forum/showthread.php?t=935362
http://www.xtremesystems.org/forums/...3&page=1&pp=25
, 591 in total. I categorized them as in (**), so I made no difference between opteron 165 or FX-60 for example, as both of them have Toledo cores.
Now if you assume that the extreme results roughly weigh out the too loose results, all elements of the compiled lists make up an average processor. As statistical treatment requires, I included all entries in the databases and didn't filter out dubious cases.
Here are some plots of the samples for the different cores, with linear-chi^2-fits. I chose a linear fit for simplicity.
1. Single-Cores
http://img46.imageshack.us/img46/287...locking8uv.gif
2. Dual-Cores
http://img102.imageshack.us/img102/4...locking8hr.gif
3. All fit-curves with four points measured at my own Opteron 165.
http://img211.imageshack.us/img211/7...locking0cv.gif
This picture shows, that the curves are nearly parallel. This actually is the most convincing argument that any processor overclocking potential can be calculated with these curves. Depending on the quality of your
1. die
2. cooling
3. mainboard
4. PSU
, your curve (the one that matches your processor) is like one of these, just shifted up or down. edit-> There apparently is more evidence for the linear relation to hold, as hitechbj1 pointed out (http://www.ocforums.com/showthread.p...06#post2588006):
Quote:
Originally Posted by hitechbj1
How does Vcore improve CPU frequency?
When certain higher Vcore is applied to a CPU or any logic chip, the current, called Idsat, in transistors would increase, hence the transistors can operate the logic gates faster. The delay of a logic gate, transistor current (Idsat), voltage (Vcore), loading (Cload) are related by
delay = Vcore Cload / Idsat
Idsat = k (Vcore - Vt)^n
where Vt is transistor threshold voltage and k is some constant, n is between 1 and 2.
Simply speaking, the higher the Vcore, the higher the current Idsat, the smaller the delay (hence chips run faster).
It can be shown that for a given temperature and without temperature constraint, increase Vcore can speed up frequency about linearly, that is the higher the Vcore, the higher the max frequency.
frequency / Vcore ~ constant
<-edit
My Opteron 165 performs on air worse than the average Toledo core (see lowest picutre). However, that average is based on results from overclocking-freaks, that proudly present the returns of good hard labour. So don't be disappointed if your curve is below average.
