The Nag Gets Out of Rehab

Bottom Lines

  • AMD stopped neglecting the Neglected Nag, and you can’t call revision B a Neglected Nag anymore. It looks like 2.4GHz (real) is reasonably doable.

    A processor capable of 2.4 GHz indicates a fairly level playing field for the rest of the .13 micron cycle. On the whole, we’ll be back to Athlons being better at certain things, PIVs being better at others.

    This is very good for overclocking AMD fans.

  • Although AMD’s revision cut power consumption somewhat, any serious overclocking attempt in the 2.4GHz range will mean 85 watts or better. That means 1 watt per sq. mm. This is very bad for the typical overclocker.

    There’s two big unanswered questions:

  • The big unanswered question is the fate of the Neglected Nag (now known as revision A of the Thoroughbred). Will they go out and shoot the poor thing to put it out of its misery, or will AMD keep selling it (at least for a while) in the lower speed ranges? In the near term, that could be bad for bargain-hunters.
  • The second is: How much will this delay Barton?

    The Fake Problems

    There’s a few items that some Intelverts might mindlessly toss out when confronted with a newly-leveled playing field, but they don’t hold water.

    Strictly speaking, this is a vaporware announcement. The CPUs won’t be available for a while, probably September for OEMs, October for the rest of us. Please note that the reviewer CPUs were made in week 31, which makes it an early August chip. These are hot-off-the-griddle CPUs.

    However, there are times when such announcements are justified, and this is one of those. Long-suffering AMD fans have been defecting to Intel, and a lot more waverers would have gone in September without any good reason to hang on, and this is good reason to hang on.

    AMD should also be commended for handling what is a preview right (unlike at least one video card company I know). The CPUs were sent out, and reviewers weren’t told how to look at them.

    AMD also adjusted their PR downward a bit. A few Intelverts might spew that this somehow invalidates the PR process, but this is hardly the case. AMD gave itself a cushion when it created PR, and used some of it when it named new CPUs. Maybe three anal-retentives would have been happy if the 2200+ had been called the 2184.367, but no one else cares.

    Keeping matter simple and making occasional adjustments is fine and reasonable.

    The Real Problem


    From what looks to be a yet unreleased AMD tech doc, running at 2400MHz will require 77 watts of power.

    That means the future 2400+, running at stock voltage is going to put out almost 1 watt per sq. mm. Even just increasing the voltage from 1.75 to 1.85 will put you over 1 watt/sq. mm.

    This is roughly what the Neglected Nag is putting out (though of course at a significantly higher speed). These are unprecedented levels of heat, and will require unprecedented cooler capabilities.

    Combine that with an ever-growing desire for quiet cooler operation, and you have a real problem, both from the OEM and typical overclocker standpoint.

    OEMs don’t want to pay a lot for a heatsink. Neither do the majority of overclockers. Too many of both still treat a cooler as an afterthought. When folks selling current TBreds tell you that the RMA rate on current TBreds is high due to overheating, and Nag pioneers report big heat problems, and the new product puts out the same kind of heat, this is something that needs to be pointed out big-time.

    You can’t afford to be nonchalant about this.

    We take this issue so seriously that when we end up getting this chip and the C1 stepping PIV, Joe will get them first to determine how bad the heat problem is, and how well air solutions work.

    We think CPU heat is going to be the biggest issue the rest of the year. We think both AMD and Intel will be faced with serious heat problems. The AMD chips will generate the nastier numbers, but the PIVs seem more delicate when on the griddle.

    What About The Neglected Nag?

    At some point, AMD will run out of Palominos. What will replace them?

    I’m sure a lot of you are probably thinking, “Boy, oh boy, pretty soon I’m going to buy a 1700+ or 1800+ Thoroughbred Revision B pretty soon for $60 and get 2.4GHz out of it.” Why shouldn’t you, that was my first thought. 🙂

    But then I kept thinking. If AMD stopped making Palominos in late June, and hasn’t started making Revision Bs in quantity yet, just what has Dresden been doing the last two months.

    A reasonable guess would be Neglected Nags, wouldn’t it?

    I think there’s a pretty good chance that if you order a low-speed TBred in a month or so, you’re going to end up with a Neglected Nag.

    Who knows, maybe AMD will continue to make Neglected Nags for the lower speed bins. It saves them money on conversion, and makes you spend more money on the high-flying CPUs.

    That certainly would makes sense for AMD. Not so good for overclockers, but you can hardly fault AMD for not wanting to spend extra money so you can give them a lot less.

    Do I know that for sure? Of course not. Personally, I hope I’m wrong, but it would be foolish if you didn’t make sure before you buy whether or not that was actually the case.

    And if that’s too much to do when it could make a 500MHz difference, well, pardon me, but I just can’t call that smart.

    Are We Headed Towards A Glut?

    The “official” AMD price for these new processors are:

    2400+: $193
    2600+: $297

    These prices are unlikely to hold for any length of time, if indeed they ever end up being charged. By the time these processors actually show up, $130 and $200-220 will be more like it (and if Intel gets aggressive, they could be lower).

    These prices are hardly unreasonable, unless you’ve gotten used to the idea of paying $60. And if you’re patient enough, these processors will eventually cost only $60 (though if the Neglected Nag hangs out, that might be a long, long time from now).

    There’s at least the potential danger that the combination of Palomino price clearance and “eventually, everything we make will cost $50” will make people reluctant to pick up “premium” priced processors.

    On the other hand, this downshift to cheaper processors is also affecting Intel; Celeron PIV are selling much better than Intel would like.

    Behind all this is both Intel and AMD building up more and more CPU capacity for what is essentially a market that isn’t growing. There’s growing rumbles that the computer industry is not picking up as expected.

    More capacity, no more demand = glut = lower prices, maybe much lower prices.


    There seems to be some indications that these Revision Bs are more a substitute for Barton (in the short term) rather than an immediate predeccessor.

    This may end up giving Intel an edge once dual DDR becomes common.

    What dual DDR will do is give Intel a further edge in those areas where it already has an advantage over Athlon. Overall, about 7% won’t be far off the mark, but that 7% is an average of practically no improvement in office-type apps, and 15% in bandwidth intensive programs.

    Going to a 512K cache would improve AMD’s performance in the areas where it shines, by about the same 7%, though again, more in the areas where it currently shines.

    No Barton soon means no equivalent adjustment.

    There’s a rather interesting statement about Barton in this review.

    In the future, AMD plans to use the SOI technique (nearly pure copper interconnects) for the first time with its Barton core. CPUs produced with SOI technology are supposed to process signals faster and with less power consumption. This enables a 35% increase in clock speed with the same power consumption.

    That would mean eventually a 3GHz+ Barton at the same furnace level, or a 2.4GHz Barton running at Palomino heat levels, which would be very good news for the noise-adverse.

    However, we’ll probably not see such beasts until well into 2003.

    What About Hammer?

    In April, it was “Where’s TBred?” Now you have to ask yourself, “Where’s Hammer?”

    It makes you wonder a bit when AMD is now scaling up what they expect from Athlons, and apparently scaling down Hammer expectations: the figure now commonly being mentioned is 3000+ rather than 3400+.

    If AMD has figured out its problems at .13 micron for Athlon, you’d think the same would be true for Hammer.

    Of course, Hammer has that built-in memory controller, and AMD has admitted there’s been a few problems associated with that.

    The projected XP numbers approach or perhaps will even exceed the Hammer number. They certainly are close enough so that people would think twice about junking their socket A systems.

    Unless . . . .

    There’s two possibilities here, an optimistic and the pessimistic one:

    The optimistic view is that whatever breakthroughs/fixes AMD came up with for Athlon will be just as applicable to Hammer, and that the initial Hammer will come in higher than the 3400+ promised. You don’t get fratricide with a 2800+ or 3000+ Athlon when you have a 4000+ Hammer.

    The pessimistic view is that Hammer is really having problems (and the prime suspect in that case would be the memory controller). AMD doesn’t see a near-term fix, so it’s falling back on Athlon to keep the company going.

    I can argue either side about as well, so I might as well do something unusual, just shut up already. 🙂


    This was something AMD needed to do to keep its head above water the next few months. Does this the equivalent of the cavalry coming in and making everything all better? No. Better enough? Maybe.

    It’s certainly good news, in some ways, very good news, but it just doesn’t wipe the slate clean.

    More on this Thursday.


  • Be the first to comment

    Leave a Reply