Hot Days, Hot Processors

Add Your Comments

The Inquirer has an article about forthcoming Intel moves, and XBit Labs raises some questionable questions about just how well (if at all) future Northwoods will work with current PIV mobos.

The significant piece of news is that Intel will announce what used to be called Granite Bay (now the E7205) in late October. This would seem to indicate that no matter what Via or SiS does or doesn’t do, you can get a dual DDR board for Christmas.

Will Intel move up its schedule as suggested? I doubt it, simply because anybody who would actually buy a 2.5GHz+ mobo is on vacation, and there isn’t exactly a big back-to-school market for $500+ processors. This kind of move might get people in certain Californian corporate halls all hot and bothered, but just going outside will do the same if not more. 🙂

The key to Intel’s fall moves is the C1 stepping for the Northwood. Any new Northwood announced from this point on will have it, which brings us to the next point.

Getting Up To Speed On Steppings

Since people haven’t been waiting for a new Intel stepping to show up for some time, it’s probably a good idea to talk about this a bit.

“Stepping” is a geek term for “revision.” A new stepping usually fixes a few bugs and tweaks the design a bit to allow for somewhat higher speeds.

Both Intel and AMD periodically revise their CPUs. Only difference is that Intel makes these things public, and AMD doesn’t; we only get notified with AMD CPUs when we see different codes on the chip change.

When Intel talks about new steppings, they usually give a date by which developers should be ready for the new CPUs. That is the earliest possible date they can show up.

I’m already getting emails from people thinking that they’ll just await the price cut, they’ll order, and voila, one C1 stepping CPU. Doesn’t work that way.

When Intel comes up with a new stepping, the old one doesn’t go away. Resellers sell the old ones before ordering new ones, and Intel usually keeps shipping old ones until their inventory is out.

In the past, it’s usually been about six weeks after ETA for new chips to show up, and I recall that for one Coppermine stepping, the wait was close to three months.

So please don’t start ordering parts now and assume you’ll get a C1 stepping CPU if you order September 2. You probably won’t.

When ETA approaches, we’ll go over what you need to look for, and rest assured we’ll keep a close eye on availability if for no other reason than I want to buy one. 🙂

However, trying to find these things before they generally become available can be a real hassle (unless a reseller flat out says he’s selling you a specific stepping in writing, you have no beef if you order and you get an older one).

I personally go to computer shows and visually inspect these things. This does make you look like Übergeek to the average seller, who at best thinks you’re bizarre and at worst thinks you’re trying to put one over on him.

November and Dual DDR

If you’re looking to build the dream Northwood system with less social embarrassment, it probably would be wise to look at the beginning of November, when not only the C1 stepping 2.4B should be universally available, but when Intel dual DDR boards should make the scene.

What will you get out of dual DDR boards? It will pretty much be like going from SDRAM to DDR; very little improvement in office-type apps, 10% of more in memory-bandwidth intensive games. About 7% overall performance improvement won’t be too far off.

You’ll need two DDR sticks to do this, which will cost you a bit more, and the DDR boards will cost a bit more, too, but an extra $50 for 7% is a reasonably good investment.

There may be a more pressing reason for you to wait, also.

A Dicey Fit

As mentioned above, XBit Labs has a piece on how PIV processors running at over 3GHz may be “incompatible” with current socket 478 mobos.

The article cites three reasons:

  • Incompatibility with hyperthreading
  • Inadequate heat protection for mobos
  • Inadequate power supply circuitry

    Compatibility with hyperthreading seems to be more a BIOS issue than anything else. In any case, hyperthreading is hardly a universal improvement; it helps some things, does little for others, actually hurts sometimes.

    From at least an overclockers’ standpoint, it’s a little hard to take the following two claims all too seriously when hardcorers have gotten well past 3GHz and over 4GHz.

    True, if you’re using liquid nitrogen, heat around the CPU is hardly a problem, but it’s hard to see how a few extra watts can’t be handled with a little effort.

    Power supply circuitry may be a problem with shoddier boards, but until we have some more evidence of this, given what the hardcorers are doing, it’s hardly, “Throw your PIV mobo out now?”

    In any case, if you wait until late October, you’ll end up with a mobo that will take these possible potential problems into account, but the reason for you waiting should be more getting dual DDR than avoiding these “issues.”

    If you already have a PIV board, certainly don’t even think about tossing it until you have real reason to.

    The Real Issue

    I want to see what Intel is going to use for a heatsink on the 3.06GHz.

    The big issue for the rest of this year is going to be heat. Both high speed Intel and AMD chips will be tossing out considerably over .5 watts per sq. mm; the AMD chips much in excess of that.

    Combine that with a desire for quiet fans on coolers, and you have a big problem, and from what we’ve seen already, I don’t think the kind of heat spreaders we’ve seen so far are a solution (Hint, Hint to entrepreneurs: replacement copper heatspreaders maybe?).

    I think we’re going to hear a lot about crashing TBreds and semipermanently throttled PIVs this fall.

    Yes, there are available answers, but the reality is most current overclockers are just not going to go to water cooling. They’ll give up overclocking rather than do that.

    Nor is it any longer a matter of “end-of-generation heatwave, a die shrink will cool things down again.” It sure didn’t happen with TBred, and from preliminary numbers, Prescott isn’t going to be any better.

    What’s been happening is that both AMD and Intel keep reducing die size about 50% each generation, but only reducing power 25%. This make the core hotter and hotter and hotter, and we’re just about at the limits of conventional cooling right now.

    Yes, there’s a few more items in the bag of tricks, but they’ll get more and more expensive, and if things keep going the way they’ve been, the day is coming when the cooling will cost more than the CPU.

    I think the CPU companies will have to do a rethink about CPU design, and sooner rather than later.

    Ed

  • Leave a Reply

    Your email address will not be published. Required fields are marked *