Logo by Tom Purves

Microprocessors (PC compatible)
A little outdated now...recommedations need revising!

Common fallacies

Using MHz between different processor architectures or processor generations to measure performance

This is probably one of the oldest and most common mistakes a person can make. The simple fact is that while one chip may runs at a higher clock speed, it is entirely possible that the other processor may be able to get more "work" done in a given amount of clock cycles. Most people know enough not to make this kind of mistake any more.

Concluding that Processor B's architecture is inferior to Processor A's because:

1. Processor A runs at a slower clock speed than Processor B
2. Processor A performs equal to or better than Processor B

Many people are making this mistake, including people who like to fancy themselves knowledgeable about processors. Processor architectural design is not always about efficiency...it usually involves yield and scalability considerations as well. For example, a lot of people are currently making a comparison between the Cyrix 686MX and the Pentium II. "Oh, the clock speed of the Cyrix is only 187.5 MHz, yet it still manages to perform about the same as a 233 MHz Pentium II!" "Why should we pay for Intel's inferior architectural design?" Well, excuse me, but let me know when Cyrix scales their chip up to 400 MHz like Intel.

Remember, processor complexity and the MHz rating ARE related.  There is no "one is better than the other."  There is only the price and the value you get for that price.

Primary issues with buying CPU's

There are typically two sides to CPU's that you need to pay attention to. These two sides are integer and floating point performance. Floating point performance is commonly referred to as FPU performance, where FPU stands for floating point unit.

Integer performance is the benchmark of processor performance for most people. Word processing, databases, spreadsheets (mostly), and presentation programs all make extensive, if not exclusive, use of integer instructions.

Floating point performance has been less mainstream up to now, but I would argue that that is changing. Floating point performance, in my experience, affects mostly applications which try to model or deal with the real world. To break it down even further, it affects programs which make heavy use of mathematical algorithms and need to manipulate a lot of data. This would include applications like 3D modeling, rendering, video and audio compression/decompression, and other similar tasks. You can really only word process so quickly, but the types of applications which I have just mentioned use floating point power, and a whole lot of it; you literally can't get enough. And yes, this is really starting to include games. The bottom line is that in FPU intensive applications, you need all the CPU power you can get, and the particular kinds of applications I just mentioned are becoming increasingly popular.

I like to think of the whole thing like this. I'm sure you know that integers are numbers like 1, 2, 3, and so on. Floating point represents numbers out to a certain precision, e.g. 2.00000000, but also 3.14159265 or 1.41421356. Well, in an application like a word processor or a database, the computer simply has to follow a complex set of instructions. Each instruction, like, say, underline this word, can be represented by an integer. Using a number like 3.14159265 to represent underlining a word would be stupid! In mathematically intensive applications, however, you can't avoid encountering big numbers like that. Square root this, sine that, and it's easy to see how often numbers like that pop up. An FPU is optimized to handle irrational numbers and perform transcendental operations, among other things.

486 class processors and below

These are completely obsolete, both price and performance wise. You should not even consider buying one of these. The only case in which you might that I can think of is if you have an existing 486 system which can be upgraded via chip and you absolutely don't have the time to go for a Pentium-class processor.

AMD's K5

A pretty decent competitor for the Pentium which looked promising a while back but which was unfortunately a little late out of the gate for AMD. These chips run at slower clock speeds than Intel's Pentium equivalent, but AMD's chips get more done per clock cycle overall. So while the AMD K5-133 performs about the same as a Pentium 133, it's actually only running at 100 MHz. Note that this equivalence applies only to integer performance, which, as I mentioned above, is what 90% of the market cares about. FPU performance is rather bad compared to the equivalent Pentium. K5's also run hot and don't overclock easily. But they are a great value right now and if you really have to save every penny, this is the chip to look at.

Just a little bit of trivia...the AMD K5-90 and 100 actually ran at 90 and 100 MHz. But so do the K5-120 and K5-133, respectively. You can see that AMD did a little (or a lot) of tweaking.

From what I understand (which means you should read up on this a little more if you plan on overclocking), the K5-133 and 166 only have two clock multipliers, 1.5 and 1.75. If you set the clock muliplier to 1.5 or 2.0 on your motherboard, the K5 will use the 1.5 multiplier. Set it to 2.5 and you'll get the 1.75 multiplier. Supposedly the logic behind this was so that people who didn't understand the K5-133 only ran at 100 MHz wouldn't accidentally overclock their processor when installing it.

Cyrix PR166+/PR200+

Everything I wrote about the K5 pretty much goes for the Cyrix chips too. They run at a slower speed than the equivalent Pentium but get the same amount of work done, they run hot and don't really overclock easily, and their FPU performance stinks. I don't really like Cyrix because in the past they've had some weird compatibility problems. This seems to be much less of a problem with this particular generation of chips, but in rare circumstances I've still heard of certain incompatibilities.

Cyrix released "L" versions of these chips a few months ago which use a split-voltage scheme like the Pentium MMX chips. Hence, these chips run cooler and probably overclock a little better too.

Pentium and Pentium MMX chips

These are the standard setters manufactured by Intel. They set the pace for a while, but as of June 1997, they are starting to show their age. The Pentium MMX chip, which was introduced only recently, will exhibit a 10% improvement in speed over the original Pentium (aka Pentium Classic) at an equivalent clock speed due to a doubling in size of the L1 cache from 16K to 32K as well as some other minor optimizations. Pentiums are also capable of SMP, but to date I have not seen a motherboard which supports an individual cache for each chip. This hurts SMP performance a lot, since it is very bad for two chips to have to share the same cache, but two chips is still better than one. Still, if you are serious about SMP, I'd have to go with the Pentium Pro. Both chips have decent integer and FPU performance. However, given that the prices of AMD's K6 have dropped to realistic levels recently, I can't recommend buying a Pentium or Pentium MMX right now unless you are really set on getting an Intel chip. Pentiums overclock pretty easily, although on some chips Intel has disabled certain clock multipliers, meaning you have to increase the bus speed in order to overclock the chip. Pentium 60's and 66's ran at 5V, but they are completely obsolete right now. All other Pentiums run at 3.3V or, in certain cases, 3.6V. Pentium MMX chips use a split-voltage scheme where the Pentium interfaces to the motherboard at 3.3V, but the core of the chip is actually powered at something like 2.9V.

Personally, I consider the Pentium line of processors to be obsolete.  I'd rather buy an AMD K6 than a Pentium MMX.  The only place where these processors are interesting now are in mobile products where low power dissipation is crucial to battery life.

AMD K6

As one of the most eagerly awaited processors of this year, the AMD K6 certainly has a lot going for it. Some background info first: The K6's PR2 rating refers to the equivalent Intel Pentium Pro speed. It also just so happens that the clock speed of the K6 happens to be the same as its PR2 rating. So the K6-PR2/200 is supposed to be equivalent to the Pentium Pro 200 and also runs at 200 MHz. The K6 also supports Intel's MMX instruction set.

Anyhow, let's start with the good points. First of all, its integer performance is in the same class as the Intel Pentium Pro. Second, it fits into Socket 7 motherboards (most Pentium compatible motherboards), meaning in many cases you can simply buy the K6 for an easy upgrade from your existing processor or put together a whole new system very cheaply. Third, it overclocks quite easily. Fourth, it's equally proficient at running 16-bit code as it is at running 32-bit code. And last, but not least, it's cheap!

Now for the downsides. The K6's FPU performance has improved on the K5's to approximately Pentium-class performance. It is unfortunately nowhere near the Pentium Pro's...an unfortunate thing for those of us who were hoping for a real Pentium Pro killer. The K6 does not support Intel standard APIC for SMP because that's patented...instead, it supports something called OpenPIC. Unfortunately, no motherboard chipset currently in existence has OpenPIC implemented. That means the K6 is essentially useless for SMP as of now, although that may change in the future if a chipset supporting OpenPIC is released.

Finally, although it's a thing of the past now, the prices for the K6 were initially much much higher than they should have been. I really feel that a lot of the people who were buying the K6-200 at $450 a pop (and that's being generous...I saw them going for up to $600) were simply not thinking straight. I'd even go so far as to say those people either had money burning a hole in their pocket or were just plain stupid. However, at a price of only $250 for the K6-200, you should buy this chip rather than a Pentium or Pentium MMX.

The K6 runs at different voltages depending on the processor speed rating. The K6-166 & K6-200 run at a core voltage of 2.9V, and the K6-233 runs at 3.2V.

Cyrix 686MX

Orignially codename the M2, this newest chip from Cyrix bears some similarities to the K6. Although raw CPU benchmarks seem to show it falls behind other processors such as the K6 or Pentium Pro, in real world benchmarks it seems to hold its own. Like the K6, it also supports Intel's MMX instruction set. And, unfortunately, its FPU is even weaker than that of the K6, making it particularly unsuitable for high-end applications, multimedia development, or many kinds of gaming. At this point, it's not possible to say whether this chip will overclock easily. However, it is expected for the M2 to come in at a price lower than its equivalent competitors. I would still buy the K6 instead, though.

Intel's Pentium Pro

Intel's successor to the Pentium drew a lot of fire near its introduction because of its somewhat lackluster performance with 16-bit code. Intel had predicted that the PC industry would have made a transition to a 32-bit architecture more quickly than it actually did...a prediction that proved to be incorrect. At any rate, the Pentium Pro 200 remained the king of hill processor for an amazingly long time, and its price has really only varied from $700 last summer to around $480 today.

While some people may try to point out the 16-bit code deficiencies in the Pentium Pro, the reality is that in today's world, that weakness isn't an issue. Assuming you can even notice the difference between the Pentium and a Pentium Pro while drawing up a presentation or writing a paper, it won't be a very big one. Furthermore, in the applications which really need the extra power, like Photoshop, 3D Studio, Layer III playback, games, and so on, you WILL feel the extra power of the Pentium Pro. That goes for either Windows 95 or NT.

The Pentium Pro boasts some general architectural improvements over the Pentium. Integer performance is moderately improved, and FPU performance is much stronger than the Pentium. As you should have realized from the above information, the Pentium Pro's forte is 32-bit code, which means it's much more at home running an OS like Windows NT or various flavors of UNIX rather than Windows 95, or, god forbid, Windows 3.1. Although the L1 cache of the Pentium Pro is only 16K, like the Pentium, the L2 cache of the Pentium Pro is actually bonded to the chip during the manufacturing process and runs at the chip's speed! This largely eliminates the major memory bottleneck created by the disparity in chip and bus speed in today's Pentium systems.

The bonded L2 cache also leads into another one of the Pentium Pro's strengths. SMP. The Pentium Pro and its supporting chipsets easily allow for up to 4-way SMP. Each Pentium Pro has its own L2 cache, which is critical to good SMP performance. From this you might gather that the Pentium Pro performs significantly better than the Pentium in SMP environments, and that is indeed the case.

All Pentium Pro motherboards should come with a VRM (Voltage Regulator Module) which brings the voltage to the chip to the proper level. If you don't have one, something is really wrong. Very old Pentium Pro's need to have the VRM adjusted manually via jumper settings to modify the voltage output of the VRM. I personally have never seen one of these. The newer chips program the VRM automatically to the correct voltage, so no user intervention is needed to set the voltage.

Pentium Pro 150's run at 3.1V and, to my understanding, are manufactured on a different process than the other PPro's. That's why the 150's, as far as I know, only overclock to 180 with a reasonable success rate. I haven't heard of any making it to 200, but almost all of them get to 166.

The rest of the Pentium Pro's, 166, 180 and 200, run at 3.3V and are manufactured on the same process. Buy one of these chips and you have a reasonable chance of getting to 233 MHz. Nice, isn't it? Anyhow, the Pentium Pro 150, 180 and 200 all come with 256K of L2 cache. The 200 also comes in a 512K cache model. The 166 is exclusively manufactured with 512K L2 cache. Intel charges approximately double what you would expect for the 512K cache chips, and the larger cache doesn't really affect performance noticeably outside of server applications.

Intel's Pentium II

The Pentium II is Intel's newest flagship processor. The processor and cache come in a cartridge style packaging which plugs into the motherboard. The cartridge and slot are patented by Intel and are called Slot 1. The patent prevents other competitors like AMD and Cyrix from producing their own chips (or cartridges, it seems) which are compatible with Slot 1. Quite frankly, there doesn't seem to be any advantage to the Slot 1 architecture over the socketed chips, so this really seems like more of an anti-competitive move than anything else.

The Pentium II is extremely similar to the Pentium Pro. The four major differences are:

1. The addition of a segment descriptor cache, which now allows the Pentium II to execute 16-bit code at a very decent speed, unlike the Pentium Pro. The Pentium Pro often had to flush its entire pipeline while executing 16-bit code (I don't remember exactly under what conditions, I'll probably add the reason later). As I mentioned before, however, 16-bit code is almsot a moot issue at this point.  Nobody I know runs 16-bit code any more.

2. The L2 cache is built onto the Slot 1 cartridge, but runs at only half the processor speed. Intel was having yield problems with the Pentium Pro because bonding the L2 cache and the chip together could only take place before testing. Hence, failures meant the whole chip was bad. This new design makes chip yield much higher since the cache is separate from the chip and only has to run at half the chip speed, but obviously slows things down a little.  Additionally, because Intel doesn't have to waste fab space making cache, they can use the extra capacity to make processors, which is what makes money for Intel.

3. The L1 cache was doubled from 16K total to 32K. This speeds things up and probably was done largely to compensate for the halving in speed of the L2 cache.

4. MMX instruction support was added. Whether or not this is really important depends mostly only what you like to do with your computer. In general, I'd say MMX is more hype than substance at this point.

In general, the Pentium Pro actually performs equivalently to the Pentium II in almost all applications at the same clock speed. In fact, the Pro often performs better because of its faster L2 cache. The big advantage the Pentium II has is simply the fact that Intel plans to scale the Pentium II's clock speed up to much higher speeds. It's hard to beat that higher clock speed. Currently, the lower end Pentium II's, such as the P2-233, are good deals.  Even the new P2-333 is an OK deal because of its incredible overclockability.

The Pentium II core voltage is only 2.8V for the 233-300 MHz line...this helps a lot in keeping the chip from overheating too much at its higher clock speeds.  At 333 MHz and above, the current core voltage is 2.0V, due to a process shrink.  This is what makes the P2-333 so good at overclocking.  The Pentium II 300 features ECC L2 cache. I suppose this could help processor reliability, but it really seems useful in server mission-critical situations.

Most of the Pentium II 233-300 MHz processors are now locked to a maximum clock multiplier for their speed rating.  That means that you cannot overclock these processors without increasing the bus speed.  The 333 MHz line is not restricted at this time.