Pros: Faster, cheaper than the Pentium III.
Cons: Not all SSE instructions supported; no dual-processor support yet; AMD's spotty delivery history.
Bottom line: A real alternative--if AMD can keep its act together.
AMD made a name for itself with its K6 family of processors, low-cost alternatives to the Intel Pentium II and Celeron that helped create the sub-$1,000 PC market. AMD's latest offering, the Athlon (known as the K7 while in development) changes the game entirely. Although the exact advantage depends on the test, the bottom line is clear: AMD is no longer just a supplier of inexpensive alternative processors but also the source of the fastest x86 processor you can get.
The Athlon is based on an entirely new processor core. Like the Pentium II and Pentium III, the Athlon is mounted in a module with separate SRAM chips for the L2 cache, and it has a special bus that connects the processor to the L2 cache chips. Also as with Intel's current processors, the L2 cache is 512K, and the interface runs at one-half of the processor's clock speed.
The Athlon design, however, has a number of advantages over the Pentium III's. It has four times as much total L1 cache as the Pentium III. The chip can decode any three x86 instructions at a time, whereas Intel's chip can decode three instructions only if two of the three are simple instructions that translate to a single internal operation. The Athlon can dispatch up to nine internal instructions per clock cycle to its execution units, compared with the Pentium III's five. And unlike previous AMD processors, the Athlon is no slouch when it comes to floating-point and MMX instructions. Pentium III can execute a floating-point multiply every other cycle or a floating-point add every cycle; the Athlon can issue both a floating-point multiply and a floating-point add every cycle.
These microarchitectural advantages mean higher performance for integer-based business application and server processing, and floating-point-intensive workstation and gaming applications. The Athlon's advantages showed themselves well on our benchmark tests. The 600-MHz Athlon was 4 percent faster than a 600-MHz Pentium III on ZD Business Winstone 99; 27 percent faster on ZD CPUmark99 (integer CPU calculations); 8 percent faster on ZD FPU WinMark 99; and 2 percent faster on ZD 3D WinBench 99.
The Athlon/600 also beat the Intel Pentium III Xeon/550 with both processors running Windows NT. The Athlon was 16 percent faster on ZD High-End Winstone 99, 14 percent faster on Business Winstone, and 26 percent faster on ZD CPUmark 99. Thus, the Athlon is a strong offering for workstations and entry-level servers as well as for PCs.
One major source of concern about the Athlon is its lack of SSE. To counter such concerns, AMD's new Enhanced 3DNow! technology provides equivalent functions and AMD also added a few new instructions to speed up digital signal processing tasks, such as MP3 audio decoding. We expect to see many games optimized for both environments, but in the meantime, users will have to check their favorite applications to see which scheme is supported.
Another concern, for PC makers at least, is the immediate availability of Athlon-ready chip sets and motherboards. The Athlon requires entirely new chip sets and motherboards because the Athlon bus runs at 200 MHz and uses a totally different design. At 200 MHz, the Athlon bus is twice as fast as today's Pentium III bus and 50 percent faster than the 133-MHz bus Intel will launch this fall. While the performance benefit of the faster bus is minimal in the initial Athlon systems since a typical memory system today runs at 100 MHz, the fast bus gives AMD lots of headroom. Future high-end chip sets could support two banks of SDRAM memory, or they could use Direct Rambus memory, either of which will be able deliver data faster than Intel's 133-MHz bus can transfer it.
Most of the early Athlon systems will use AMD chip sets. ALi, SiS, and VIA, however have all signed up to produce Athlon chip sets. In addition, because Athlon uses the same bus as the Alpha 21264 processor, multiprocessor chip sets being developed by Alpha Processor will also support the Athlon. Furthermore, a start-up called HotRail is developing a multiprocessor chip set for the Athlon.
Pros: Fast, scalable.
Cons: Expensive.
Bottom line: The right choice for performance oriented users skittish about AMD.
The big news about Intel's latest high-end chip can be summed up in three letters: SSE. In fact, the only substantive difference between the Pentium III and its Pentium II predecessor (which Intel will discontinue at year's end) are these Streaming SIMD (single instruction, multiple data) Extensions and higher clock speeds.
The "Streaming" in SSE refers to instructions that expedite the transfer of long streams of data to and from memory. Hence, the value of these instructions is not limited to 3-D and multimedia; they can speed up any program that accesses large data sets, such as database managers.
SSE also contains a few instructions, called new media instructions, that are essentially extensions of MMX. These instructions use the same set of registers as the MMX instructions and, like the MMX instructions, perform integer computations. They speed up video processing (such as software-based MPEG-2 encoding) and speech recognition engines.
The SIMD instructions in SSE can significantly improve the performance of 3-D geometry and lighting calculations. This part of the 3-D task is not typically performed by the graphics accelerator chips available today (though that will change next year), so SSE is complementary to--not competitive with--accelerator chips. It's important to stress that, as with MMX, SSE improves performance only when software is modified to use the new instructions.
The value of SSE showed up clearly on some of our tests and not at all on others. On ZD 3-D WinBench, a 450-MHz Pentium III beat a 450-MHz Pentium II by 62 percent on the Transform and Lighting task. Using Adobe Photoshop with the Twirl filter applied, the Pentium III bested its predecessor about 30 percent. By contrast, under ZD Business Winstone 99, which does not benefit from SSE, the Pentium III showed no advantage over the Pentium II.
The Pentium III family also includes the Xeon processor. The main difference: The Xeon has a faster and larger L2 cache (up to 2MB) than the standard Pentium III, which makes it better suited for multiprocessor servers, where the large caches reduce bus usage and provide better multiprocessor scaling; and for systems that use large working sets, such as database engines. Our testing showed little performance advantage in typical PC applications.
This fall, a new version of Pentium III, code-named Coppermine, will debut. Built using Intel's most advanced semiconductor process--an 0.18 micron design--Coppermine should scale to higher clock speeds. Intel is having trouble achieving the speeds it expected, however, and it recently delayed the chip's launch (in the desktop version) from September to November. It is, however, expected to hit speeds of at least 667 MHz (with a 133-MHz bus). In the meantime, Intel has launched a 600-MHz Pentium III based on the original 0.25-micron Pentium III.
Coppermine will include 256K of on-chip L2 cache that will run at the full speed of the processor. Initially, Coppermine probably will be sold on a Slot 1 module, just like today's Pentium III, to maintain motherboard compatibility. In time, however, all of Intel's processors are likely to move to a pin-grid-array socket similar to that used for the Celeron (the so-called Socket 370).
Pros: Affordable; fast enough.
Cons: No SSE (yet).
Bottom line: A good bet for all but the most power-hungry users.
Where home and corporate buyers are concerned, the biggest development in the past year has been Intel's newfound commitment to the low-cost PC market. The first Celeron chip (released in spring 1998) had no L2 cache and was a performance dog. Buyers (especially business users) balked, and Celeron became the chip non grata. It wasn't until the late-1998 launch of the Celeron processor with on-chip L2 cache--code-named Mendocino--that Intel's offerings became competitive with the low-cost processors from AMD and Cyrix.
The Celeron processor takes the same processor core that is in the Pentium II and adds 128K of on-chip L2 cache that runs at full processor speed. By contrast, the Pentium III has a larger 512K cache, but it is off-chip and thus runs at one-half of the processor speed. Consequently, the Celeron's performance is much closer to that of a Pentium III than the four-to-one difference in cache size might suggest.
In fact, for many applications, a Celeron processor is nearly as fast as a Pentium III at the same clock speed. For example, our ZD Business Winstone 99 tests showed that a Celeron/ 500 delivered the same performance as a Pentium III/450, while a Celeron/466 trailed the Pentium by only 3 percent.
Of course, if the L2 cache were the only difference between Celeron and Pen tium III, there would be very little reason for PC makers to pony up as much as $300 more per chip for Pentium III. To maintain greater differentiation, Intel has limited Celeron's bus speed to 66 MHz, whereas Pentium III processors use a 100-MHz bus
Pros: Affordable; fast enough.
Cons: MMX and FPU performance are below that of Intel CPUs, and K6-2 (not K6-III) falls short on integer performance as well.
Bottom line: A perfectly acceptable choice, especially for home users. Also a good choice for corporate users, who rarely have any need for MMX or floating-point performance.
AMD has made a huge impact with its K6-2 processor, especially in the U.S. retail market, where AMD's share is neck and neck with Intel's, according to InfoBeads (www.infobeads.com), Ziff-Davis's market-research service. This chip has also been successful in the "white box" business PC market (systems sold by system integrators under their own brand) and is used in some direct-channel PCs as well. The K6-2 also rocketed to a greater than 40 percent share position in the U.S. retail market for notebook computers, though its share of corporate notebooks remains small.
The K6-2 uses the Socket 7 bus--the interface Intel defined for its older Pentium processor--but runs the bus at 100 MHz instead of the 66 MHz Intel used. AMD calls systems using this bus, which typically also have AGP, the Super7 platform.
Unlike Pentium II and Pentium III, the K6-2 does not have a separate interface for the L2 cache, so the cache speed is limited to 100 MHz and must share the bus bandwidth with memory and I/O traffic.
Thus as clock speeds increase, its performance does not rise as steeply as that of the Intel processors. The AMD K6-2/475, for example, scored only 3 percent above the K6-2/400 on our ZD Business Winstone 99 tests, whereas the Intel Celeron/466 beat the Celeron/400 by almost 7 percent. And tick for tick, the Celeron tended to outpace the K6-2 by 5 to 10 percent. The K6-2 processor also has slower floating-point and MMX units than Intel's chips, so its performance fell further behind on many 3-D, image processing, and multimedia applications.
AMD's follow-up, the K6-III, debuted in early 1999. It added an on-chip, 256K L2 cache, providing a significant performance boost. For example, a 450-MHz K6-III was 21 percent faster than a 450-MHz K6-2 on ZD Business Winstone 99. And the K6-III/400 performed about 11 percent better on Business Winstone 99 than a Celeron/400.
Nevertheless, systems based on the K6 III are relatively rare for two reasons: semiconductor economics and retail PC buyer behavior. The on-chip L2 cache makes the chip much larger, which both increases the cost of manufacturing and decreases the number of chips AMD can produce given its capability. At the same time, PC buyers look at clock speed more than benchmark test performance.
Pros: Affordable; adequate for low-end machines.
Cons: Sale of the company may make PC makers wary.
Bottom line: Worth considering for a budget-priced home machine.
The Cyrix M II was quite successful in the low-cost PC market in 1998, especially in the least expensive systems. As Intel became more aggressive with Celeron prices and clock speeds, though, Cyrix's market share plummeted. National Semiconductor, Cyrix's parent company, decided in May to exit the PC processor market and put Cyrix up for sale. As of this writing, the sale of Cyrix to chip set maker VIA Technologies is pending.
Unlike AMD's chips, the M II is rated by its "equivalent performance," not its clock speed; the M II/433 actually runs at 300 MHz. Our tests showed that it didn't quite live up to this rating; on ZD Business Winstone 99, the M II/433 scored about 5 percent lower than its closest competitor, AMD's K6-2/400, and 8 percent lower than Intel's Celeron/400. It fell much farther behind on tasks that make extensive use of floating-point or MMX functions.
The M II design is clearly showing its age, and Cyrix plans to replace it this fall with a chip code-named Gobi. This new device uses an enhanced M II–based CPU core, code-named Cayenne, and adds faster floating-point and MMX units and 3DNow! instructions. In addition, the Gobi will include a 256K on-chip L2 cache, which will run at full processor speed. Gobi also marks a shift from the Socket 7 bus; it will be the first non-Intel chip to use the P6 bus and plug into the same socket as a Celeron.
Adopted from PC Mag. By M. Slater 8/20, 1999