Chips ahoy: the new CPUs. (central processing units; includes related article and glossary) (Buyers Guide)
by Steven J. Vaughan-Nichols
Choosing a chip can seem complicated, but all you have to do is ask yourself what you need.
Once it was easy. When you wanted a new computer with a blazingly hot central processing unit (CPU), you just bought one with Intel's newest record breaker, and all was right. Those days are gone. Now, you have to face an alphabet soup of chip makers: Advanced Micro Devices (AMD), Cyrix, Texas Instruments, and even IBM. Making matters even more confusing, Intel is producing a bewildering array of chips. And, as if that weren't bad enough, there are all these exotic chips like the PowerPC and the Alpha, which just might represent the future of computing.
What's a buyer to do? Well, fasten your seat belt, because we're off on a full-speed-ahead journey into the land of chips. Once you know what you need out of your chip, all you have to do is look over your choices.
It seems as if all of the PC CPU makers give their chips names that make them sound compatible with Intel chips. But how do you know that the x86 chips from all these vendors will really be compatible with your DOS and Windows programs? I was unable to find solid claims of any major program or operating system failing on any of the non-Intel chips. In my testing experience, I have yet to see even one oddity. When it comes to compatibility, you can rest easy; any x86 chip you buy today will be able to run your software.
A much harder question to answer is how fast these chips are when compared to one another. This becomes especially perplexing given the names that chips go by these days. For example, a Cyrix Cx486SLC has 486 in its name, but it acts like a 32-bit 486 only when the data is actually inside the chip. For data input and output, it relies on a 16-bit bus. So why bother with it? The Cx486SLC is priced and marketed to compete not against i486s, but i386SXs. Cyrix makes no bones about this. The SLC is meant to be today's low-end chip, not a middle-of-the-road 486 competitor. The newer Cx486SLC2 chips have speeds up to 50 MHz and a 32-bit external bus.
Another concern many users share is that their chip vendors might go up in smoke one day because of lawsuits. While it's true that lawsuits among chip makers are becoming more common, it's unlikely that any of the major vendors will be driven out of business. Most of the judicial battles have been won by the smaller companies, and there's little to show that the trend might reverse.
Cyrix will continue to make Intel-compatible CPUs. Intel, in the face of losing several court decisions to Cyrix, agreed in February to a settlement. While Intel may yet appeal some of the decisions that have gone against it, it now seems clear that Cyrix will be in the x86 chip business for the long haul.
Good Chips, Good Decisions
If you're like most PC buyers, you're probably asking yourself, "How can I pick the right CPU?" There is no right or wrong CPU. The only question is whether the chip you select will deliver the performance you require. But before we talk about chips, you have to realize that the CPU is only part of the story. Overall system performance should concern you more than CPU performance.
Before you go out and buy a system, get all the information you can on its overall track record. Remember that a computer can perform only as fast as its slowest component. The components that can make a computer look like either a snail or a lightning bolt are the video card and the hard disk; their speeds vary the most.
A standard benchmark for video card performance is the WinMark. It's possible to cheat on the WinMark (some video card manufacturers are alleged to have built in circuitry that checks to see whether WinMark is running and, if it is, generates impressive results), so the numbers generated by the test aren't foolproof. Also, there are different versions of WinMark available, which might generate different ratings. But generally, a rating above 12 million would indicate a fast video card, and a rating below 8 million would indicate a card with dismal performance. More and more, manufacturers are installing slow video to keep system prices low. Watch out for that practice.
You don't want to buy a hard disk with an access time of 40 milliseconds or above. When you buy a system, get the fastest hard disk available and install caches to speed it up even more. Generally, disk compression doesn't appreciably change a hard disk's performance.
You will want to buy a system with an EISA expansion bus and probably a VL-bus (VESA local bus) or PCI bus (Intel local bus) unless your work doesn't depend on a high-speed bus.
You'll want to buy a system with the fastest RAM available that you can afford. Even very fast RAM might have trouble keeping up with the hottest CPUs available, and for that reason, you should also look for systems with secondary caches of very fast memory--a cache of 256K is a minimum.
A good system is like a good team: It must be made up of good individual performers. A hot CPU will chill out if it's part of a low-performing team. But once you've made sure your system has what it takes, it's time to choose a CPU, and that's what this feature is about. The broad range of chips can generally be broken down into low-end, middle-of-the-road, and hot performers. We'll cover each of these in turn.
The Low End
A 486 may be nice, but when your bank account is more important than your megaflops (a measure of CPU speed expressed in millions of floating-point operations per second), a 386 computer can still fill the bill. Here, your choices come down to Intel, AMD, and a pair of low-end Cyrix chips.
Intel still offers its 386DX and 386SX chips, but the real winner here is AMD. AMD's 40-MHz Am386DX-40 and Am386DXL-40 are the fastest 386 chips on the market. You'll find the Am386DX-40 in desktop machines, while the DXL, which uses only 3.3 volts of electricity, is usually found in laptops.
You might also want to consider systems with a Cyrix Cx486SLC. It comes in speeds of up to 33 MHz. In some ways, this chip is an odd duck. It's a perfect fit for an i386SX socket, but internally, it works like an i486SX. Unfortunately, although manufacturers find building a system around the Cx486SLC simple because of its 386SX compatibility, these systems are stuck with the 386SX's slow 16-bit external data bus.
Inside the Cx486SLC (and its lowpowered laptop brother, the Cx486SLCe), there's a 32-bit data path and a 1K internal cache. It's a tiny cache, but it does improve the Cx486SLC's performance. The chip includes the entire i486SX instruction set. Nor is the Cx486SLC a mere i486SX copy internally. It goes beyond the i486SX by including its own caching and power management instructions.
While the Cx486SLC doesn't compare well against the 486SX in terms of performance, it's another story when comparing it to the 386. Here, the top-of-the-line 33-MHz Cx486SLC fares well against the fastest Intel and AMD 386 chips.
The next Cyrix chip family, the Cx486DLC, zips past the 386 chips. Although it's pin compatible with a 386DX socket, the Cx486DLC comes with an i486SX instruction set with 32-bit internal and external data paths and a 1K cache. Like the i486SX, the Cx486DLC doesn't have a math coprocessor. While the Cx486DLC blows the doors off 386DX chips, its more of a dead heat between the Cx486DLC and the 486SX family.
The bottom line on these low-end chips is that the DLC demands your attention for fast 386-level performance. If cost is a serious issue, the AMD 386 or the Cx486SLC should be your first choice.
The Middle of the Road
Did you ever think you'd see the day when an Intel 33-MHz 486DX would be thought of as an average chip? That day is here. Still, even power users shouldn't overlook the i486 family. With the exception of such advanced operating systems as NextStep and Windows NT, almost no software seriously pushes the limits of a fast 486.
Any discussion of 486s must start with the Intel family. These chips, with their internal and external 32-bit data buses and 8K caches, still define today's computing. The i486DX CPUs range from 25 MHz to 50 MHz. The 33-MHz i486DX is by far the most popular member of the family. This is because of the perception that the 25-MHz i486DX is too slow for the jobs requiring a 486. The 50-MHz model suffers from overheating to this day. In addition, Intel is swinging its emphasis for high-end computing from the 50-MHz chips to the Pentium.
Edging in on Intel's territory, AMD and Cyrix are now serious contenders in the 486 wars as well.
AMD led the assault on the i486. For years, Intel and AMD have slugged it out in court. One legal challenge has to do with whether AMD can use Intel microcode; the other challenge is whether AMD can produce 486s using clean-room techniques. In clean-room engineering, engineers produce a chip that's similar to the original but not based on the original design. Rather than cloning the chip, the clean-room engineers try to create a chip that reproduces the original chip's actions and reactions. Intel's contention is that AMD doesn't have the right to do either.
Despite Intel's opinion, AMD will soon have not one, but two, full lines of 486 chips: one using Intel microcode and the other using code created using clean-room techniques. If Intel wins its suit claiming that AMD has no right to use its microcode in its 486DX chips, AMD will be prepared to switch its 486DX to code developed using clean-room techniques.
Today, there is an AMD alternative to virtually all i486 chips. For the i486DX-33, there's the AM486DX-33; for the i486SX-33, there's the Am486SX-33; and so on. The only Intel CPU for which AMD has yet to develop a direct competitor is the i486DX-50 chip.
In addition to the Intel twins, AMD also makes several chips that don't have Intel counterparts: the 40-MHz Am486 CPU family. In this group, you'll find both SX and DX versions. AMD also makes low-voltage chips that use an SMM compatible with Microsoft APM-aware applications (APM and SMM are explained in the sidebar titled "Chiptionary"). Some of these chips, like the 33-MHz Am486DX-33) can be powered down to a complete stop without losing data. Clearly, these chips will soon be popping up in laptops and green (environmentally friendly) PCs.
AMD is taking the offensive in the 486 wars. Compaq, the second-largest manufacturer of IBM-compatible PCs, is expected to be making systems built around AMD chips by the time you read this.
The Cyrix Cx486S series is pin compatible with the i486SX, and in most ways the Cx486S chips work just like their Intel equivalents. The most important difference, however, is that the Cx486S has only a 2K internal cache (the i486SX has an 8K internal cache). On the plus side, Cyrix's chips can slow down to a dead stop without losing information. This makes the Cx486S a natural for laptop vendors. Cyrix, like AMD, also has a 40-MHz i486SX-style chip for which Intel has no equivalent.
Cyrix also makes 486DX chips. The Cyrix C x 486 D X 50, Cx486DX40, and Cx486DX33 have 32-bit data paths and 8K caches just like their Intel counterparts.
The newest trend in 486s is the clock doubler. This is a chip that transmits and receives data at one speed and processes data internally at another speed that's twice the input/output speed. For example, a 25/50-MHz i486DX2 takes in and sends out data at 25 MHz. Inside the same chip, the data is processed at 50 MHz.
Once more, Intel leads the way in chip progress. The company's i486DX2 comes in models that run at 25/50 MHz and 33/66 MHz. Like their DX brother, these chips come with 8K caches, built-in FPUs (see the sidebar titled "Chiptionary"), and 32-bit data paths. If you want Intel reliability and a strong processor, the 33/66-MHz i486DX2 demands your attention.
By the time you read this, AMD and Cyrix will each have a DX2 chip. This time, though, they're not the only major players. Texas instruments (TI) and IBM are also in this round.
AMD offers both 25/50-and 33/66-MHz processors. AMD is also developing a 486SX-style clock doubler--the Am486SX2-50. With the exception of this last chip, the AMD models look and act like the equivalent Intel chips.
Cyrix also has its clock doublers in 25/50 and 33/66 brands. The Cx486DX2s differ from Intel's and AMD's in two ways. First, the Cyrix chips use 8K write-back caches. In write-back caches, the tiny cache holds data until the system is relatively idle before committing the data to main memory or disk. This results in a small but perceptible performance boost. The other difference is that Cyrix claims its built-in FPU is faster than those used is equivalent 486s.
Texas Instruments, building on Cyrix's technology, has its own clock-doubling 486s: the 486SXLC2s. These chips, like the Cx486SLC, fit into 386SX sockets and use the 486SX instruction set. The important difference here is that the SXLC2 has an 8K cache. This chip comes in 20/40-MHz and 25/50-MHz flavors.
The SXLC is not the end of the TI 486 story. TI's Rio Grande project will produce chips that combine the 25/50-MHz TI486SXLC2's core with a memory controller and a Peripheral Component Interconnect (PCI) localbus controller. Throw in low power consumption and size, and TI should establish the standard high-end chip for notebook and laptop computing for the rest of the year.
IBM has the fastest 486-style CPU of all. The company's Blue Lightning chip is a clock-tripling microprocessor. To communicate with the rest of the computer, the Blue Lightning runs at 33 MHz, but internally, it blasts along at 99 MHz. In addition, the Blue Lightning has a 16K cache--twice as large as those of its competitors.
The Blue Lightning has only one disadvantage: it doesn't come with a built-in math coprocessor. Still, if your applications don't need FPU power, a Blue Lightning-equipped PC is the current 486 power champ. Its reign may not last long, though. AMD will soon enter the lists with its own clock-tripling CPU, and Intel plans to release a clock tripler called (confusingly enough) the DX4. The DX4 will be available in 75-, 83-, and 100-MHz speeds.
Not everyone can afford to buy a new computer every time the industry has a new champion. If you're in that situation, you're in luck. Cyrix offers several upgrade chips, and there's always the shimmering promise of Pentium upgradability.
The [CxDRx.sup.2] gives users a single-chip solution for upgrading 386DX computers. This clock-doubled 486 CPU fits into the 386 socket and brings with it 32-bit addressing and a 1K cache. The closest relative to this chip is the Cx486DLC.
These chips can give 16-, 20-, and 25-MHz 386DXs new life. The Cyrix CPUs run internally at twice the speed of the chips they're replacing. Unfortunately, while Cyrix will be bringing out 386DX replacement chips, there are no current plans for 33- or 40-MHz 386SX replacement CPUs.
The replacement chips that most people know about are Intel's OverDrive chips. An OverDrive chip boosts a standard 486SX or 486DX to 486DX2 performance and speeds. Also promised is a next-generation OverDrive chip based on Pentium technology that will boost 486DX and 486DX2 systems to new performance levels.
Also coming from Intel is a Pentium upgrade chip for 486 machines with special Pentium upgrade sockets. The chip is currently called the P24T, and it is expected to run at speeds up to 100 MHz. System buyer beware: Many systems with Pentium upgrade sockets will not be able to use the new chip because of system design. Intel has started an Intel Compatibility Lab to certify systems as P24T-ready. The chip is expected to be a fully functioning Pentium except that it will have a 32-bit data bus instead of the Pentium's standard 64-bit data bus. Is the rumored 486 OverDrive chip the same thing as the chip now known as the P24T? Only time will tell.
The High End
At the high end, the Intel Pentium stands alone, at least for now. The Pentium, with its 60- to 66-MHz clock speed, runs much faster than any other chip in the x86 family. Pentiums with speeds of 90 and 100 MHz should be appearing as you read this.
But pure MHz numbers tell only part of the story. The Pentium's advanced design, with its superscalar architecture, dual 8K caches, and two integer execution units, breaks all former x86 architecture speed records. How good is it? With plain, old software, a 66-MHz Pentium is almost twice as fast as an i486DX2-66. Give the Pentium some software that's been optimized for it, and it will leave the other x86 chips in the dust.
Recent releases from Intel describe the 90- and 100-MHz Pentiums as the most powerful chips on the market. The 66-MHz Pentium is expected to be the midrange CPU in a year, and the clock-tripled DX4 will be the standard for mobile computing. These same releases speak tantalizingly of a P6, a new high-end chip to replace the Pentium.
Roaring up to challenge the Pentium is AMD's clock-tripled 486 and Cyrix's M1 chip (at the time this is being prepared, M1 is the code name for the Cyrix Pentium challenger the chip's name could change by the time it's released on the market). AMD is also working on a direct challenger to the Pentium known (at least for now) as the K5. Details on these chips are sketchy at press time, but you can safely assume that the new CPUs will perform at speeds close to the 60-MHz Pentiums, though the manufacturers like to say, "Pentium performance or better." One stated that a 100-MHz Pentium challenger would appear.
Raising the Roof
The x86 family is no longer the only game in town. RISC chips are rushing out from the workstation world to assault the PC's CISC land (RISC and CISC are both explained in the side-bar titled "Chiptionary"). Why, after attempts to enter the PC market that were so feeble that almost no one noticed, are RISC chips now becoming a factor in the PC market? The answer is Windows NT. Windows NT is a PC user-friendly operating system that doesn't run just on PCs.
Leading the way is the Alpha chip from Digital Equipment Company (DEC). The Alpha 21064 can run as fast as 200 MHz (DEC has promised even higher speeds to come). In addition to speed, it offers 64-bit internal and external buses and a 16K internal cache. The Alpha is simply the fastest PC chip in the land. It won't run standard DOS, but a special form of Windows NT has been written for the Alpha, and you can run DOS software under Windows NT, so an Alpha machine could be considered as an alternative to a standard PC (though a very expensive one).
A company named MIPS is also demanding attention with its R4000 series. The 64-bit R4000 chips can run as fast as 150 MHz and have up to a 32K internal cache. You may never have heard of MIPS, but its chips have what it takes to make an impact on at least the network server side of the PC world. Nintendo plans to use MIPS chips in its high-end game machines starting in 1995, so you may encounter them there, if nowhere else.
As fast and powerful as the Alpha and the R4000 are, the chip that really has the Intel-bound contingent of PC users talking is the PowerPC. This chip family has the backing of three of the PC's superpowers: Apple, IBM, and Motorola.
The PowerPC 601, the first of the series, will soon be appearing in shops everywhere as the heart of the new PowerPC Macintoshes. This chip boasts a 32K cache, 64-bit internal and external data buses, and 66-MHz speed. Although comparing CPUs of different types isn't easy (comparing clock speeds alone isn't enough because some chips accomplish more tasks per cycle than others), the 601 seems to improve on the 66-MHz Pentium's performance.
Things will only continue to get faster from there. The big three are already planning faster 601s. The PowerPC 603, with its low power consumption, will bring Pentium-level power to laptops. Past that, the 604 will bring even more power to desktop machines by early 1995. About the same time, the 620 will also appear in servers and multiprocessor systems.
What does it all mean for you? It depends on your needs. There are some things that you can't have. For instance, you can't buy a computer that won't be out-of-date in three years (maybe less). Computers are advancing too fast for anyone to stay on technology's ever-sharper leading edge.
If your aim is to get the least expensive computer you can find that will run Windows, machines packing the AMD 40-MHz Am386DX and the Cyrix Cx486SLC and Cx486DLC are worthy contenders for your dollars. Are your favorite applications programs like WordPerfect 5.1, dBASE IV 2.0, and Lotus 1-2-3 2.2? If so, consider any of the Intel and AMD 386DXs. A machine with a 386SX at its heart is now at the end of its useful lifespan, and it's not a good buy.
If you have a few thousand dollars earmarked for your next computer, consider a machine with a 33-MHz Intel 486DX, a 40-MHz 486DX, or any of the clock-doubled AMD, Intel, or Cyrix chips. You may want to avoid chips that don't include FPUs. Most of the time, you're not likely to use an FPU, but they sure are handy when you do need them. Some high-end design programs make extensive use of the FPU and may slow to a crawl without it.
At the high end, the Pentium almost has to be your choice today. The 66-MHz Pentium is more valuable than the 60-MHz, and its price should be dropping by the time you read this. Even the 100- and 90-MHz Pentiums will probably come down in price in a year or so.
That's the safe solution. If you want to live more dangerously, go with a brand-new 100-MHz Pentium, a Cyrix M1, or an AMD K5. You can be the first in your town to boldly go where no computer user has gone before.
For the truly adventurous, the RISC chips beckon. If looking for a server is the name of your game, both the MIPS and the DEC chips are real possibilities. The PowerPCs, with their ability to run almost any microcomputer operating system in existence today, also look like winners. Macintosh users will have little choice about switching over to RISC, since all future Macs will be PowerPC based. That may not be a bad deal though, and PC users should consider PowerPCs of their own. The future of computing is now.