Summer PIV O/Cing

Intel will cut prices May 26, bringing a few more PIV Northwood into the typical O/Cer’s budget.

Based on studying CPU databases, both ours and others, I came up with this little chart which
summarizes your likelihood of success with the various processors at the moment with high end air at the moment.
Please note that speeds of 2.5GHz or better will likely require more than a 10% increase in voltage, most
air-cooled scores show voltage increases to a range of 1.70-1.85V.

If your idea of a summer upgrade is at most to pop a cheap Thoroughbred into your socket A box and hold your hard-earned cash for future developments, that is wise. If I were in your shoes, that’s what I would do.

Items To Consider

1) What Settings? It is unclear whether a high FSB/high memory combination provides any real benefit over a medium FSB/high memory combination on the PIV platform like it does on the Athlon platform.

High FSB and memory help the Athlon because the FSB becomes a bottleneck for memory, and increasing the bandwidth eases the bottleneck.

In theory, the total PIV platform bandwidth should not create a bottleneck for memory, but we have a few doubts about this, and plan to test at 166MHz to resolve them.

Running a PIV at 166MHz will not be for the casual overclocker this summer. The only candidates for this treatment are the 1.6A now, and the 2.26 a little later. While some are getting it done with the 1.6A, that chip will be shuffling off the stage shortly. We haven’t even seen a 2.26 yet; it’s on the edge. Our best guess is that the first ones won’t generally
do it, but those 2.26s made with the optical shrink will have a fighting chance (more on this later).

In any case, we don’t know yet if 166MHz FSB is worth much of a struggle to those less inclined to struggle. By this time next week, we should know if it is, or whether medium FSB/high mem speed is just as good.

Unless Intel comes up with low-speed 133MHz FSB Northwoods for OEMs, we don’t expect 166MHz PIV overclocking to become routine until the fall.

2) What Stepping? Right now, we have the B0 stepping for these chips.

Probably sometime in August, PIVs made after Intel’s optical shrink conversion (the link includes future sspecs for these chips) will become available. Though Intel
swears it didn’t tweak these chips for performance, anecdotal evidence seems to indicate such chips do perform a little better.

Intel’s own actions (and lack thereof) seems to indicate the same). The optical shrink is occurring first with the fastest processors almost right away, then the rest.

The Intel dog that hasn’t barked (at least not yet) is any announcement of a new stepping for the Northwoods from the B0 stepping. Normally, they do this several months ahead of time to give developers time to test these new products. They haven’t done it yet.

It’s conceivable that when Intel introduces the 2.6 and 2.66GHz processors, these newbies will have a more advanced stepping. New stepping or not, they’ll almost certainly be made using the optical shrink from the start.

If they don’t have a new stepping, though, Intel is basically saying that they can put out CPUs that can do 2.66GHz at stock voltage without a new stepping. Unless they’re just skimming off the cream of the CPU crop (which Intel has done in the past on rare occasion), no new stepping would be a silent admission by Intel that the shrink does help.

Moving a little further on, Intel expects to start selling a slightly over 3GHz PIV next January, maybe a bit sooner. I don’t think this optical shrink will be enough to make that possible, so I expect to see a C stepping for the PIV to become available sometime in the fall. This should make 3GHz overclocking at least reasonable feasible and probably fairly easy.

3) Double DDRing? There seems to be little performance difference between the new Intel chipsets and the forthcoming Via P4X333. The Via may rack up slightly better numbers. SiS will have its equivalent in a couple months, pulling an answer from nether regions, it will probably do a few percentage points better yet.

If you’re planning on medium FSB/high memory overclocking, the best of all possible worlds would be to have a 3:4 and a 4:5 ratio setting for that. That allows the most tweaking options for your equipment. If you have to choose, if you’re less confident in your memory than your PCI components/hard drives overclocking, go 4:5. If it’s the other way around, go 3:4. If you’re inbetween, do some math with the ratios before you buy.

Probably in the fall, we’ll start seeing dual DDR PIV boards, first from Via and SiS, later from Intel. To me, “What improvement will we see from it?” is the second biggest question left in 2002 (the first being “How will Hammer do?”) “How much more will such a board cost?” will be another good question.

For those who have to make their money count (or just those planning to get a Christmas computer), these are the two questions you need answers to before you can wisely buy.

P.S. Intel actually has a dual DDR board out now for Xeon processors. There’s a report out which I’m sure some RDRAM retards are and will be citing as “proof” that dual DDR stinks.

Actually, it does no such thing. All the evidence shows is that if you are using a multithreaded server process for a database, you’re using hyperthreading, and you’re servicing 25 or more clients, RDRAM does rather better. That’s hardly the desktop environment, and it would be foolish to say “dual DDR is a dud, RDRAM rules” based solely on this single, specialized example.

We’ll get much better proof one way or the other soon enough.

Ed