June 2000 CPU Buying/Upgrade Guide: Part I: General Principles

I’m writing this to gather together in one place all the information I’ve mentioned in earlier articles, and update where needed.

Parts II and III will talk about specific equipment and what to keep an eye on.

However, I think it’s a good idea to first look at the general picture, whether you should or shouldn’t update and what you can expect for your money.

Ambivalency abounds

A long time ago, like last year, buying a new CPU/motherboard used to be simple. 🙂

If you didn’t mind spending money, you bought a PII. If you did mind, you bought a Celeron. In either case, you bought a BX board. If you were committed to AMD for some reason, you bought a K6-X and a super 7 board.

Now it’s not so easy.

On the Intel side, you have two types of Coppermines to confuse you, different steppings with much different potentials floating around simultaneously, and the Celerons aren’t quite as good.

The motherboard situation is a mess, especially if you overclock. Either your BX boards fights it out with your components to see who’s boss, or you go with a fairly funky Via board. With Solano, it’s beginning to look like we’re going from two lousy choices to three.

On the AMD side, AMD has caught up with the Athlon, but it’s been practically Motherboard of the Month and processors haven’t been much better. Overclockers have to play with Golden Fingers and esoteric software because their motherboard may say “133,” but they sure don’t run at 133.

Things will settle down a bit for a couple months, but fall brings the real Memory Wars, along with the processors associated with each side.

So what’s a poor overclocker to do?

I get a lot of letters asking me whether or not to buy and what. (I also get letters telling me to stop doing that.) I wish there were an ideal situation for everyone, but there isn’t.

Different people have different circumstances. Different people have different levels of patience.

My general views on buying computer equipment are:

  • There are better and worse times to buy computer equipment. You tend to spend too much money and not get an optimal product early in the development cycle; you get a shorter useful life if you wait towards the end.
  • Cost/benefit is a big factor for me. If you have to spend $500 for a 25% improvement, but $120 for a 15% improvement, I’m very inclined to spend the $120.
  • Buying a computer should not be an impulse buy. These are not toasters. You should plan and study a bit before buying one. If you don’t, you are likely to not get what you really wanted.
  • Ask before you buy, not the other way around. The worst part of what I do is telling people who ask that their newly purchased equipment won’t do what they want. I hate telling them that, and they hate hearing it even more. That’s why I sound like a nanny at times.
  • There is a lot of hyping going on, and everyone gets swept up in it one time or another. I see my job as countering the hype, so you don’t get disappointed when you don’t get what you had been led to believe.
  • I find price pretty important, but time is also. In most cases, there comes a point where waiting a few months longer is only going to save you $20-30, and it usually isn’t worth the $20-30 you save for the time you go without the upgrade.
  • Your circumstances can make my observations inapplicable. In that case, they aren’t for you. For instance, if you are earning $100 or more an hour doing computationally intense work where 10% more speed means you can do close to 10% more work; the cost of the computer equipment is practically irrelevant.

  • What I say are general guidelines, not exact rules. Don’t take them personally unless I put your name on them.:) Take whatever is good for you from them to make your own decisions.

Upgrades: How Much Pain, How Much Gain?

Each of you have different computers bought at different periods of time for different needs and desires. You all have your own upgrade clock. The person who has a 200MMX system is in a different boat than someone with a Celeron 550.

Each of you have your own expectations as to what improvement you expect from a computer purchase or upgrade. If you expect an immediately noticeable change in everything you do with a computer, it usually takes two-three years for computers to advance far enough for you to get that impression. If you need that kind of improvement to justify your purchase, don’t upgrade yearly, you’re going to get disappointed.

If, on the other hand, you are satisfied with less than a stellar improvement, then it makes sense for you to upgrade more often. If all you want is an increase of 10fps, you can get that fairly often (though you should ask yourself if that 10fps is really going to make a difference).

Let’s go back to those 200 and 550Mhz systems.

The guy with the 200MMX is going to see an immediately noticeable change if he buys a computer that he gets to run at 800Mhz or better. If he’s doing any kind of work or gaming he takes seriously; he’s going to see a real change, so he is a prime candidate for an upgrade.

The case with the guy with the Celeron 550 is not so clear cut. If he upgrades to an 800 or 900Mhz system; he’s not going to see anywhere near that kind of improvement. He may not notice any in a lot of the things he does. If noticing a big change is important to him, then he probably shouldn’t lay out a whole lot of money to do so.

I’ve seen a lot of people buy a PIII. Then they try to run at 133Mhz or better. Then they find the memory doesn’t work, or the video card, or the motherboard. Or some combination of the three. So they find themselves spending a lot more money than they initially planned to justify the earlier purchases for a general improvement of maybe 25%-30. That’s about what you’ll get in general going from that Celeron 550 to a PIII 800

Maybe they would have been better off leaving the system alone and spending money on something else in their life, or save up for a major upgrade next year. Maybe they would be better off buying a Celeron II to extend the life of the current system for a bit. They’d get a bit less performance for a lot less investment, and again use or save the money.

If spending money that way doesn’t personally bother you, good. My message isn’t for you, but only for those it might bother.

New boxes: What Do You Need?

When people ask me about the kind of computer they should buy or build, I rarely hear initially what should be the first question you should ask (or ask yourself):

What are you going to use it for?

The answer to that question alone can answer everything else.

I have a friend who wants a new computer. Until recently, his processing needs were modest, so I was going to go with some older equipment that met those needs.

Then he told me his flight instructor thought it would be a good idea to supplement his flying lessons with some flight-simulation.

I took a look at some websites dedicated to that, and almost dropped dead seeing the kind of firepower needed to get frame rates up to what would be considered lousy by Quake timedemo standards.

So now we’re looking at overclocking a PIII 650/700 to 900-1000Mhz, lots of RAM and good 3-D video rather than a PII 400 and 128Mb and decent, older 2-D video.

One little change; entirely different setup.

You have a college student who needs to write term papers and hit the Internet, you don’t need much at all. If that college student likes to play genocide generators, now you do.

One little change; entirely different setup.

But I Want A 1Ghz Computer!!!


You may have a legitimate reason like my friend. Or you may not.

Let me ask you this: Wouldn’t a 980Mhz computer be just as good? If your answer is “No,” then you don’t want a fast computer, you want a status symbol.

A computer is a lousy status symbol. Even the most expensive computer is cheap compared to a Mercedes or a mansion. The average member of the female sex will be less than awed and would much rather you spent money on much better things, like her. Computers get outclassed very quickly, and the geeks who would be the most impressed in the first place will be the first to know when something faster comes out, anyway.

There’s nothing wrong with wanting a fast computer, and if you can get one fairly cheaply, great. Just don’t spend a lot of money on a fleeting ego boost.:)

Next, a look at Intel’s processors.

Email Ed

Note: I’m indicating what is very likely to work for people. A few will do better, a few worse. If you did better, good for you, but that doesn’t mean everyone else will. I know there are people out there with older Coppermines that are running wonderfully with BX boards at 155Mhz, but everyone can’t count on that.


This is pretty simple. If you have a good BX board setup, you should buy a Celeron if either 1) you don’t want to spend a lot of money on an upgrade; 2) you’d rather spend your money on a
major upgrade towards the end of this year or early next; or 3) not have to go through all the booby traps of Coppermine upgrading. The performance isn’t as close to the Coppermine as the initial Celeron’s was to the PII, but it’s not that awful, and if it’s a choice between that or buying a Coppermine and memory, most would probably be better off with the Celeron.

If you are building a new system, particularly a back-to-schooler, you should take a very long look at the Duron. I’ll talk about that in the next section.


We’ve now reached an optimal time to buy a Coppermine. The cB0 stepping will usually get you to over 900Mhz, and may get you close to or just over 1Ghz with good air-cooling.

Here is a chart to tell you what you should be looking for:

100Mhz Processors

FC-PGA Processors  
S SpecSpeedRetail/OEM
SECC2 Processors  

An S-Spec is a code which indicates a particular type of processor. If you buy a retail model, the sspec is found at the very end of the product code found on the box. In all cases, it is printed on the CPU.

However, this is not an optimal time to find the right Coppermine.

The problem we face now is that there are still plenty of the older cA2 stepping Coppermines being sold out there. They were still being made until late April. This doesn’t matter too much if all you are looking for is 800Mhz, the older chips should be able to do that.

However, if you plan on buying a 650 or 700 expecting to get around 900Mhz or better out of it, it’s probably the difference between success and failure because if you get an older chip, you probably won’t get 900Mhz out of it.

No, you can’t just order a cB0 chip from any retailer. Very few Internet resellers are specifically selling cB0 processor. PC Nut PC Nut so far is the only one that states such on Pricewatch. Proton Computers is another. There’s a couple others, but they are
either brand new, or have had a lot of negative feedback at Reseller Ratings. If I’m not comfortable buying from them, I can hardly recommend you do.

Yes, you can go to the local store and visually inspect the processor to guarantee you’ll get what you want. Yes, you’ll pay more than you would from an Internet reseller.

If you call an Internet reseller who doesn’t specifically sell cB0 processors, no, the salesperson will probably not know what you are talking about.

In most cases, the reseller ships out of a warehouse, so, no, the salesperson couldn’t check the CPU for you even if he or she wanted to.

Yes, salespeople lie to get a sale.

Unless you get it in writing from the company that you are being sold a cB0 chip, if you don’t get one, it’s probably tough luck. There’s no record of any salesperson’s promise, and they sold and delivered the processor they promised. You’ll probably have to pay shipping and a restocking fee just to play Russian Roulette again.

Yes, this is a hassle, but how much of a hassle is it being stuck with a processor that won’t run as fast as you want if you don’t protect yourself?

E or EB?

If you want to overclock, you buy the E because you can overclock a lot more with the E than you can with the EB

Let’s show why the E is a better choice for overclocking:

The speed you’ll get from either CPU is easy to figure (if not asterisked, assume success rate at that speed is 80% or better, presuming proper equipment along with the CPU):

ProcessorMultiplierMhz at 133Mhz FSBMhz at 140Mhz FSBMhz at 150Mhz FSB

**Most people won’t make this speed with an older cA2 stepping chip, so if that’s what you are counting on, you
must get yourself a cB0.

***Don’t count on this speed even with a cB0.

****Probably out of realistic range for cB0 stepping with air-cooling.

As you can see, you’ll always have a much higher potential speed with a 100Mhz “E” chip than with a 133Mhz “EB” chip. You’ll also notice I didn’t list the 750 or 800E. I didn’t because it looks like the cB0
tops out at about 1Ghz with aircooling. A 750 or 800 is unlikely to do any better than a 700, and cost a good deal more.

Tsk, Tsk, Tom’s

Seems simple, right? Unfortunately, the birthplace of overclocking, Tom’s Hardware, doesn’t seem to realize that. Take a look here. It says 133Mhz Coppermines are ideal overclockers. Wrong.

To be fair, the article was written before cB0 processors became generally available, but the author (not Tom Pabst but Patrick Schmid) ignores HIS OWN BENCHMARKING which showed that a 600E running at 800 beats even a more expensive 667EB chips running at a 150Mhz FSB most of the time (and 150Mhz is no gimme on either BX or Via boards). With the cB0 stepping, there’s even more of a spread.

I only found out about when someone questioned me on my preference for 100Mhz processors, saying that Tom’s had recommended 133Mhz processors. I had to tell so many people who bought 133Mhz processors that they weren’t going to get much out of them, and I really wondered why I was getting so many. I guess that was part of the reason.


SECC2 (Slot 1)


  1. You won’t have to buy a slotket.


  1. If Intel’s heatsink/fan isn’t good enough for the retail version, you’ll have to do things like remove the casing.
  2. Future motherboards are likely to come in FC-PGA (though slot1 motherboards should be around for quite a while).

  3. Probably will become harder to find as the year progresses.



  1. Heatsink fans are a bit cheaper than for SECC2 models.You’ll buy and attach the best heatsink/fan without having to strip the casing.
  2. If you buy the retail version, you can discard Intel’s heatsink/fan without having to strip the casing.
  3. Intel plans to convert most Coppermine production over to FC-PGA shortly, so new motherboards will likely come in FC-PGA.


  1. You’ll have to pay extra for a slotket for it to fit on a slot 1 board.
  2. Might be harder to find/more expensive right now.

What motherboard?

You have three choices right now: use your BX board and overclock it a lot, buy a VIA board, or wait for a Solano2 board.

BX board


  1. If it works, you’ll get a bit better performance than with a VIA board.

  2. You don’t have to buy a new motherboard, but you should check our database and also http://www.deja.com/home_ps.shtml, type in something like {your BX board} {processor you want to buy} and see how people are doing with it.


  1. You probably won’t be able to overclock the CPU as much on a BX board as you can with a VIA board.
  2. Overclocking the AGP transfer rate by 33% or more could be hazardous to the health of your computer.

  3. You are more likely to hit a memory bottleneck.

This all assumes you have a pretty current motherboard handy. If you don’t:

  1. Don’t even think about doing this if your motherboard doesn’t support a PCI divisor of /4 at 124Mhz and above
  2. If your current motherboard doesn’t support Coppermines, you might be able to get it to work with a slotket with a voltage regulator, but don’t try it unless you’re ready to buy a new motherboard, too, if it doesn’t work or work satisfactorily.

VIA Apollo Pro 133+


  1. With a 1/2 AGP speed, you don’t have to worry about your video card or motherboard blowing up, and you’ll probably overclock more than with a BX board.
  2. You are less likely to run into problems due to memory speed with the VIA board, since you can set the memory speed to CPU-33Mhz.


  1. The VIA boards don’t perform as well, clock for clock as BX boards, though it’s usually by only a few percentage points.
  2. They don’t appear to be as reliable as the BX standard.
  3. You’ll have to buy a new motherboard.



  1. Probably will be a little better than a VIA board, may or may not be better than the BX.
  2. Will have an AGP divisor of 1/2, so you won’t have to massively overclock the board.


  1. They’ll be introduced June 19th, but how available they’re going to be is anybody’s guess.
  2. Unless somebody’s figured out how to eliminate the integrated video chips that are supposed to be part of the board, it will probably cost more than BX or Via boards.

  3. Intel hasn’t been too good making chipsets lately; this is a first-generation product that could have lots of its own problems.


Unless you have a VIA board that lets you play games with your memory speed, you need to know how good your memory is if you plan to run a Coppermine at 133Mhz or better. Don’t assume because it is “good” PC100 RAM that it will run at 150Mhz. It probably won’t. Most “good” recently manufactured PC100 RAM will get you to 133Mhz or a little more.
If you have older PC100, don’t even assume that. Take a look at our database and see how many people say, “Have to buy new RAM to get it going faster.”

If you don’t currently have or plan to buy quality PC133 RAM, you probably should be prepared to replace your memory (or buy a VIA motherboard that MAY let you overclock the processor a lot while running your memory slower) if what you have now isn’t fast enough. If you aren’t willing to do either, then you should buy a Celeron.

Video cards

It doesn’t seem to much matter what video card you use. They should work on a VIA board; on a BX board, they’ll either work or not, without much rhyme or reason. Some people can play Quake III all day long at 155Mhz. Others can’t do the same thing with the same video card at 120Mhz with all the tweaks.

Some people get helped by certain tweaks, some don’t.

If you have a BX board and you would rather keep using it, there’s nothing wrong with trying it first. However, if it doesn’t work, what are you going to do? Be prepared to buy a Via (or Solano) board just in case it doesn’t due to video, or just buy a Celeron in the first place.

If you are upgrading your system to play games better, you probably should also consider the purchase of the next generation of video cards coming soon to take advantage of the increased speed. You may see little or no improvement if you don’t.

Tomorrow, we’ll take a look at AMD.

Email Ed

Bottom line:

We have to see if the QDI overclocking circuitry works, and if it does, how far one can overclock a Thunderbird or Duron. The results of that testing
will determine what you should do with any type of Athlon, or whether you should get involved with them at all at this point.

Land of Confusion

We have .25 micron Athlons. We have .18 micron Athlons. We have Irongate motherboards. We have KX-133 motherboards.

We have Durons coming. We have Thunderbirds coming. We have aluminum processors; we’ll have copper processors. We have Slot A processors, maybe they’ll work with some older motherboards, maybe they won’t. We have Socket A processors. We have KT-133 motherboards coming. We’ll have DDR motherboards coming, and probably new versions
of Durons and Thunderbirds to work with them.

What we have is confusion.

It’s been amazing AMD’s been able to keep TRACK of all these changes, let alone make them.

Give AMD credit, though, when the roadmap of what we’ve seen over the last year became apparent, my attitude was “Yeah, right.” But they did it with only a few scattered problems.

The Difficulties With Recommending AMD Now

There’s been some specific problems, but the perverse problem for me has been that the Athlon hasn’t quite grown-up yet. Ironically, that’s no knock on AMD.

Looking at Coppermine vs. Athlon is like looking at two kids. The Coppermine kid is five feet tall; and the AMD kid is four foot ten. The Coppermine kid has parents that are both five foot two; the Athlon kid has parents who are six foot eight.

The Coppermine kid may be a little bigger now, but you know the Athlon kid is probably going to end up a lot bigger. However, CPUs don’t grow in your computer like kids. So if you have to grab the taller kid right now, you grab the Coppermine kid, but you know you’d grab the Athlon kid down the road.

We’ve seen about all we’re going to see out of the Coppermine. Intel mumbles about 1200Mhz; but they are going to have some real problems doing that in .18 micron. So grabbing something now makes sense because you’ll probably wait quite a while to get much over 1Ghz, if it even happens.

On the other hand, we know the Thunderbirds, due to their design, have a lot more legroom. We expect well over 1Ghz from them in the next six months. However, we also know that 266Mhz DDR is coming in a few months, which should boost performance by roughly 10-15% all by itself.

So we’re at a dead end with Coppermines, but we’re not with Athlons, which hurts current sales because you know something better’s coming soon.

(Something you have to keep in mind is that AMD is still not making a whole lot of Athlons. This quarter, roughly 1.8 million, a little more than 5% of world CPU requirements. Next quarter: 3.6 million as Dresden starts kicking in. 4Q: 7.2 million. AMD’s actually having shortages now, so saying “buy later” isn’t exactly a knife in the back.:))

Current overclocking difficulties: one problem fixed, one not

Most people really don’t want to struggle to overclock. Give them the choice of a couple BIOS settings, or hardware modifications, and the vast majority will take the first.

You can call them wimps, but these folks are voting with their wallets. A lot more people have been overclocking Intel chips rather than AMD chips for just that reason.

The current Athlons are a lot harder to overclock than Coppermines. It’s certainly not Congressional Medal of Honor material, but it’s more than most want to do. Opening up an Athlon case and inserting Golden Fingers isn’t that big a deal, nor is using software to reduce the cache ratio. But it’s a lot harder than changing an FSB BIOS setting. On top of that, reducing the cache ratio hurts performance, adding insult to injury.

True, the cache problem’s been fixed with the Thunderbirds and Durons. It’s not like a low-end Thunderbird or Duron is going to be much faster than an “old” Athlon; it isn’t. This is a removal of a handicap, not an enhancement, and this handicap doesn’t hurt too much at 750Mhz.

But compare a Thunderbird at 1Ghz to an old Athlon at the same speed, and then you’ll see a real difference. Even more importantly, a Thunderbird may be only roughly the same as a 1Ghz Coppermine, but there’s going to be 1.1 and 1.2 and 1.3 and 1.4 GHz and more Thunderbirds, and there won’t be equivalent Coppermines.

Why can’t a K*-133 run at 133?

Although the Thunderbird/Duron has taken care of its cache flow, you have to stick them into something, and the motherboards so far have let AMD down.

The old BX board, with its 100Mhz bus, seems quite capable of running at 150Mhz provided you find compatible video partners for it.

The Via boards, with their 133Mhz bus and at least some of the same circuitry that goes into KX-133 boards, seem quite capable of 133Mhz and better.

The Athlon boards, with their 200Mhz bus, have a hell of a time getting over 110Mhz. This is a problem, and until that problem gets fixed, AMD has a real problem, not only for overclockers, but everyone else.

You can’t expect to get great performance out of any processor, no matter how good, if you are running it at a 12X or 14X multiplier. This gets ridiculous.

(However, AMD looks ready to be ridiculous. Their data sheets indicate multipliers up to 12.5X and more. See here, page 62.)

If there were Athlon motherboards capable of running at 133Mhz (and presuming the Thunderbirds and Durons could handle the overclock, which I suspect would not be a problem); I’d be saying, “To hell with Intel, buy AMD instead” right now. They can’t, so I can’t.

AMD has to do something about this in the next few months to compete against Willamette. If I’m not in any big rush to upgrade, I still like the idea of waiting for a Thunderbird/DDR combo. If I have a BX board, maybe I buy a Celeron to tide me over for six months.

True, one motherboard manufacturer (QDI) is claiming that they can modify the multiplier setting on Thunderbirds/ Durons. We’ll have to see if this is so, and whether other mobo manufacturers have or can pick up on this trick shortly.

If true, this certainly helps AMD from an overclocking standpoint (especially for Durons), but it doesn’t solve the underlying problem. If the motherboards could run at a decent FSB, you wouldn’t need such circuitry.

So what should you do now?

How about my favorite four-letter word: Wait?

I know, a few of you think I should put that on my gravestone. 🙂

However, the QDI announcement tosses everything into the air. If this pans out, and if the Thunderbirds and Durons prove highly overclockable, this changes the whole picture.

For instance, if we find that an $80 Duron can run at 1Ghz, or a $300 Thunderbird can run at 1200Mhz; that’s going to change matters dramatically. Maybe I’ll find myself saying, “To hell with Intel, buy AMD now.” But we don’t even know if the QDI setup will even work, not even QDI is guaranteeing that.

Until we find out (and this should be a matter of just a few weeks), it’s better to wait for results, then decide based on them.

Presuming the QDI modification doesn’t result in highly-overclockable, highly performing Thunderbirds and Durons:

If you aren’t going to overclock, the “old Athlons” give more bang for the buck than the PIIIs. Ironically, the real competition is not against Intel’s chips, but AMD’s Duron.

I suspect we’re going to see some big clearance sales on “old” Athlons along with KX-133 motherboards in the next month or two. You may well find some bargains when that happens.

If you want to overclock, we’re just going to have to see about this QDI thingy (or equivalent). If you can get a Duron running at 8.5 or 9 X 100; it should blow away even an overclocked Celeron. That could become the next best bang-for-the-buck
back-to-school special, and a lot more.

The preliminary benchmarks we’ve seen on Duron indicates that it is a few percentage points slower than an “old” Athlon at the same speed, and a few more percentage points slower than a Thunderbird.

It’s probably reasonable to assume that the performance difference between Duron and Thunderbird will increase at higher speeds, but I wouldn’t at all be surprised to see a Duron running at 900Mhz or more beat an “old” high-speed Athlon, maybe by quite a bit.

We’ll see in the next couple weeks. Durons may prove to be the greatest bargains around, or they may not. If you have a current setup you want to upgrade, the price of a new motherboard may make it much less of a bargain for you, and an overclocked Celeron may still be your best bargain upgrade.

However, if you are thinking about a Coppermine upgrade, this might prove to be a real alternative.

Are these new processors DDR compatible?

Here’s my predicament.

Tom’s Hardware published a spec here, indicating that there were 133Mhz, 200Mhz and 266Mhz versions of these chips. From that, I presumed that there was reason to believe that current processors would not run at 266Mhz.

Some of you complained to me about this, pointing out that there never had been a 133Mhz Athlon processor (which up to now has been true).

I looked around a bit and found this, on page 69. This contains much less information than the first.

Now it’s quite possible Tom’s Hardware was able to fill in the details after talking to people at AMD. It’s just as possible that these were just educated guesses. I don’t know.

However, even if we ignore the Tom’s Hardware blurb, having something identified as an “200Mhz processor” raises the question of whether or not it will run at a higher bus speed.

AMD can make a statement today saying they’ll definitely run, and I’ll be happy to revise what I said. Until then, though, it is not responsible or prudent to recommend products when you have doubts they will work in future platforms.

Forget DDR, what about Thunderbird now?

That will depend on how the QDI fix pans out. If we find out a 750Mhz Thunderbird can run at 1100Mhz or more, probably would be pretty hard not to prefer a Thunderbird over a Coppermine if you have to buy now.
We have to see.

What about Willamette?

We don’t really know how it will perform (though there are preliminary indications it won’t be very good at games without SSE2 optimization); we don’t know what it will cost (except it’s hard to see it being cheap in 2000). Too early to talk about it.

Why not the detail you gave to Coppermine, oh biased one?

It’s pretty hard writing up specific recommendations when you don’t know what you’ll recommend for a couple weeks, and when the most likely candidates aren’t even available yet. After we find out about the QDI claims and the results, then I’ll go into detail. 🙂

Email Ed