DISCLAIMER: What is discussed in this piece has not been tested by anybody to my knowledge yet. The theory behind it looks fine, the mechanics of actually doing it have not yet been tested.
If you try this, and destroy your processor or
anything else in the process, this is solely your responsibility, not mine; you assume all risk.
It looks very likely that motherboards will allow you to overclock without doing any of this. The sole purpose of the article is merely to inform you that there may be a way to overclock these processors should that not be the case.
Most of the information contained in this article comes from a post at Ace’s Hardware from Armand Hirt here, along with a lot of good feedback in the thread. Thank you all.
Simply, it looks like the multiplier on the Athlon is initially set by some golden bridges on the L6 portion of the chip. Each has two points. Some are connected, some are not. If you change the connectors (by making or breaking them), that should change the multiplier on the processor.
The proposed idea is to use what is called a conductive pen to “draw” an electrically conductive trace to “connect the dots” and some sort of sharp knife/pen to break them.
The AMD tech documents, page 62, document this (and is the source for the tables below).
- We don’t know if a silver trace is sufficient to “connect the dots.” If this turns out to involve soldering, we’re back to the bad old days.
- If it is, connecting the dots is no big deal. Unconnecting the dots you made may be more of a problem, but probably quite doable with some care. Unconnecting AMD’s dots may be an entirely different story. Unless it’s easy to get a clean cut without touching the contacts, it could be very dangerous, since you might not only wipe out the connection to the dots, but the dots, too, which might destroy the chip.
- The following chart (from the abovementioned AMD source) shows the configuration for each multiple. If it’s a “1”, the dots are not connected; a “0”, they are.
If AMD’s “connect the dots” can’t be or can’t easily be removed, that will restrict us in two ways:
- If all we can do is “connect the dots,” we can’t go from 600Mhz or 700Mhz to 800 or 900 or 1000Mhz. The FIDs are arranged so that you have to break connections to reach those speeds.
- If we can only connect dots, we are restricted in the overclocking ranges we can attempt.
Here’s a chart for that:
Athlon O/Cing Options by “Connecting The Dots” and not erasing any of AMD’s
Rated Speed Theoretical O/C Options 600Mhz 1100Mhz, 1200Mhz 650Mhz 1100Mhz, 1150Mhz, 1200Mhz, 1250Mhz 700Mhz 1100Mhz 750Mhz 1100Mhz, 1150Mhz 800Mhz 1100Mhz, 1200Mhz 850Mhz 1100Mhz, 1150Mhz, 1200Mhz, 1250Mhz 900Mhz 1100Mhz 950Mhz 1150Mhz 1000Mhz 1100Mhz, 1200Mhz
As you can see, this is pretty bizarre, and stands conventional overclocking practice on its head. If we can’t break connections, it’s 1100Mhz or bust. Would a 600Mhz Duron or a 700-800Mhz Thunderbird run at 1100Mhz?
Who knows? I can tell you we’re looking at hellacious cooling just to make it possible; we’re looking at power consumption in the neighborhood of 60W and approaching 40A at 1100Mhz. However, if you have a system that can run an “old” Athlon at 1000Mhz, it should run a Duron/T-Bird at 1100Mhz.
Would on-die cache hold up? Another unknown, but if AMD’s yields are as good as we hear, with on-die cache, this might not be inconceivable.
This would be a real gut check, wouldn’t it? 🙂
Are We Going To Have To Do This?
I hope not, but I don’t think so.
The statements from QDI and others about multiplier overriding made the prospect of overclocking via the motherboard a lot iffier than it really is.
Page 38 of AMD’s tech docs says the following:
“The AMD Athlon processor transparently samples the BP/FID pins from the time PWROK is asserted until the deassertion of RESET# (in English, just at the very beginning–Ed).
The clock multiplier may be provided external to the part, or via resistors on the package”
In short, motherboard manufacturers are perfectly free technically to “provide” the clock multiplier. They may not, apparently; at least one early motherboard, the(MSI K7TPro, does not). At least according to the AMD documents, all the FID pins really do is provide a multiplier which may or may not be used, as opposed to K6-2 days when the motherboard had that task.
I suppose AMD could change its mind, but if they do, or motherboard makers decide for whatever reason not to take advantage of this, you get to replay kindergarten. 🙂
When Will We Know?
Well, nobody seems able to find QDI. The A7V is the next suspect on the list (though just because Asus has jumper settings for multiplier up to 12.5X doesn’t mean anything; they have lots of jumper settings for Intel chips, too.) Tom’s Hardware says they’re testing it right now, so if it can, I’m sure we’ll hear. Odd, though, he said he knew of two ways to overclock these chips, and mentioned QDI but not Asus.
Something important to know is that we’ll probably see Durons before Thunderbirds. The impression I’ve gotten is Thunderbirds will be rolled out fairly slowly to give resellers a chance to sell their “classic” Athlons chips. We should see Durons in about two weeks, and we plan to test it, preferably with an accomodating motherboard, if need be by connecting those dots. 🙂