(This is part of some correspondence)
After some further testing, I can see why Ed has some suspicions about the A7V. I decided to try some tricks I learnt back in the days of Celery 266’s @ 448 and my good old P2-333 that ran at 560 – slow everything else down.
I disabled L2 cache, set the memory bus to 100 and the A7V’s “system performance” option to normal. After these, the CPU would then post all the way up to 1200MHz (and all increments in between), but still simply would not boot beyond 950MHz. Could it be a limitation of the A7V? AMD seems to think so, and I’m inclined to agree at this point.
Jumping into BIOS at these speeds showed that temperature was not the issue – even at 1200 the core temp in BIOS at idle was a reasonably low 34c. Unfortunantly, the attempts I made resulted in partial frying of my C partition (taking out Win98 at 1000 clock speed); fortunately, Win2k on a separate partition held up.
Perhaps it’s just because I’ve always worked with Intel’s offerings, but wouldn’t you expect that a CPU that can post (and successfully run the BIOS menus) some 250MHz higher than it can run, would also be capable of running basic disk IO at 50MHz above it’s RSS (rock solid stable) speed? I would, but then again my experience with these platforms is limited to the config I have here now.
On the other side of the coin, I did have to disable L2 cache to POST above 1050 (1066 indicated by the A7V), so it could be dirty processing on the CPU’s part after all.
For the record, I have active cooling in the Northbridge, and a passive heatsink on the clock generator (temps stable at 32-34c).
Editorial comment: You push a CPU a little past its limits, it can trash hard drives, have had it happen to me, that’s no surprise. Came very close to that trying to run this TBird at 1050Mhz. My experience has been W2K is actually more sensitive to O/Cing than 98; you probably only trashed 98 because the CPU didn’t get a chance at W2K.
I’ve found that unlike the Intel chips, you don’t get too much harmless warning with the AMD chips when you’re exceeding limits, like freezing at the Windows screen or a Registry error. It either just won’t boot at all, or it may start trashing files. Not that you should leave important data vulnerable when first overclocking any chip, but that’s particularly true here.
Why are people for the most part finding 1Ghz a barrier?
Right now, there are too many variables. We have heat, mobo circuitry, cache, and the core CPU itself as likely suspects, and it’s probably a combination of all of them.
Now it’s a matter of eliminating possible suspects.
One way or the other, we should be able to eliminate heat as a factor in the next month (though we are getting very erratic temp readings from these chips; some have reported the TBird stays relatively cool doing certain things, then heats up big time doing something that doesn’t seem to be particularly more strenuous).
The KT7 should address the possibility of less-than-robust mobo circuitry.
This leaves the CPU; and I’m just dumbfounded at AMD saying things like they could release 1.5Ghz processors now based on what I’m seeing. My gut is telling me that there’s something about these CPUs; that we’re not getting Intel-style yields, and that buying a low-end 700 is not as close to buying a high-end TBird as it has been with Intel chips. However, I’m far from certain about that.
There are reports that Dresden is pumping out nothing less than 950Mhz chips. That may be nice if you could get your hands on Dresden chips, but how likely that is in the U.S. at the moment is an open question. How much good it would do you if you had one is another question; I have a 900 core Slot A, as do quite a few others, but the database and newsgroups are indicating that they aren’t getting much past 1Ghz, either. So buying a higher-end chip might get you another
100Mhz, but that’s it.
James’ comments do indicate that cache is running into a problem before the rest of the chip, which does not bode well for any big-time overclocking. Then again, since the chips are cheaper than Intel Mhz equivalents, you don’t need quite as much to justify the purchase.
Looking at this overall, it looks like when you take all factors into account, we have a roughly equal playing field between the Intel and AMD chips. Outside of the exceptions I’ve noted before (Durons, high-end applications under W2K), there’s no compelling reason to generally shift over to AMD from Intel yet. However, if AMD can indeed get out 1.5Ghz processors in a few months (and they are due for at least one core revision before then), and DDR boards aren’t delayed too much, that could change quickly.