[ The PC Guide | Systems and Components Reference Guide | The Processor | Processor Families | Fifth Generation Processors ]

AMD K5 ("K5" / "5k86")

AMD's entry in the fifth generation processor sweepstakes is called the K5. This processor was eagerly awaited and it was hoped that it would provide a viable alternative to the Pentium early in the Pentium's life cycle. Unfortunately, AMD delivered the processor over a year late and at much lower clock speeds than had been originally anticipated. As a result instead of being the "Pentium killer" AMD had hoped for, the K5 was positioned as a low-cost Pentium alternative, much like the Cyrix 6x86.

The K5 is, internally, a very advanced processor, the most advanced of the fifth-generation chips. Internally it is more comparable to the Pentium Pro. It is an x86 translation/emulation processor, decoding x86 instructions into RISC-like microinstructions and executing them on a 6-pipeline internal core. This allows the K5 to achieve higher performance than a Pentium of the same speed. In many ways, the K5 is sixth-generation, but its performance level is held back to fifth-generation levels due its low clock speeds.

The K5 is not a Pentium "clone" in the traditional sense of the word, because it is a totally different internal design. It is also not 100% Pentium compatible because it is not an exact clone. However, it is a very close approximation, and suffers from fewer compatibility problems than the 6x86.

Due to its advanced internal architecture, the K5 (like the 6x86) is labeled according to its "P rating", which attempts to label the processor according to what speed Pentium the chip's performance approximates. This is a good way for AMD to position its products in a marketplace that overvalues the clock speed number. The K5 is available in P rating speeds from 75 to 166 and goes in a Socket 5 or Socket 7 motherboard. It is produced on a 0.35 micron CMOS process, until recently state of the art. AMD manufactures its own chips.

Compared to the Pentium (classic), the K5 has the following architectural improvements that contribute to its greater performance to clock speed ratio (it also has all the advantages over the 486 that are mentioned in the section discussing the Pentium):

• RISC-Based Internal Architecture: The K5 is internally a highly-parallel RISC processor using an x86 decoding front-end.
• Six Execution Units: The K5 has five integer units and one floating point unit for processing micro-instructions, while the Pentium has only two integer units.
• Out of Order Completion: Instructions flowing down the two integer pipelines can complete out of order.
• Superior Branch Prediction Unit: The branch target buffer is four times the size of the Pentium's (although K5 claims a lower accuracy with it than Intel does with the Pentium. Seems odd to me.)
• Larger Primary Cache: The K5 has a 16 KB instruction cache instead of the Pentium's 8 KB. The data cache is unchanged at 8 KB.
• Improved Cache Mapping: The primary cache in the K5 is 4-way set associative instead of the Pentium's 2-way.
• Register Renaming: This feature improves parallel performance of the pipelines.
• Speculative Execution: The K5 uses speculative execution to reduce pipeline stall time in its RISC core.

While these improvements are impressive, they are the reason that the K5 has P ratings well above its internal clock speeds; if it didn't have the higher P ratings the K5 could not compete with the Pentium at all. Like the Cyrix 6x86, the K5's main's attraction is significantly lower price than the Pentium. Unlike the 6x86, the K5 has a reputation for being reasonably compatible and has far fewer problems than the Cyrix chip does. There aren't the heat and motherboard compatibility problems of the 6x86 associated with the K5. However, it (arguably) does not offer as high a level of performance as the 6x86, and there is no K5 chip near the power level of the 6x86-PR200+.

The K5 can have some compatibility problems due to the fact that it is not a Pentium. Some Pentium-specific software can cause problems with the K5, but these programs are relatively rare. The "CPUID" problem associated with the 6x86 under Windows 95 is not an issue for the K5.

While the K5 exceeds or beats a Pentium of its "P rating" at integer operations, it lags behind in floating-point performance. The K5 falls somewhere between the 6x86 and the Pentium in terms of its FPU, with performance roughly equal to a Pentium 25% slower than the K5's P rating. This really isn't much of an issue for most people since few applications make extensive use of the FPU (although Quake is one high-profile game that does.)

The 6x86 suffers from the problem where it is occasionally overclocked by accident because users set the motherboard to the clock speed of the P rating instead of the actual clock speed of the chip itself. The K5 avoids this by making two design decisions. First, all of the K5 chips run on the same memory bus speed as the equivalent Pentium of its P rating. So a K5-PR133 runs on a 66 MHz bus speed, like the Pentium 133, and unlike the 6x86-PR133+. Second, all K5 chips are designed so that if you configure them with the multiplier that an equivalent-rated Pentium would use, the chip will run with the correct internal multiplier. The combination of these two characteristics means that to set up a K5 you jumper your motherboard as if it were a Pentium and it will work. This avoids a lot of confusion.

The K5 also has an undocumented, generally unused multipler setting: if you set it to 3x, it runs at 2x. The reason for this is generally believed to be that AMD intended to release a K5-PR200, but never did. This chip was supposed to run at 133 MHz and have performance equivalent to a Pentium 200. Of course the Pentium 200 has a multiplier of 3x, so this would have been consistent with the other K5s' ability to be jumpered as if they were a Pentium. The full table of AMD K5 multipliers is:

As you can see above, the K5 has its own confusions to compensate for its better handling of jumper settings. :^) The main one is that the PR90 and PR120 both run at 90 MHz processor speed, and the PR100 and PR133 both run at 100 MHz. How is it possible for two processors with the same design to have different performance ratings running at the same speed? It isn't--and that's the confusion, because the PR75, PR90 and PR100, have a different internal design than the PR120, PR133 and PR166. The first three chips were created earlier, and were actually at one point called the "5k86".

The K5 is a good alternative to the classic Pentium, with performance equal to a mid-level Pentium. It does not have many of the problems that the Cyrix 6x86 does, but it also doesn't offer performance as high as the top-end 6x86s. However, it is now as obsolete as the 6x86, and for that matter the classic Pentium that both compete with. For the home builder on a very tight budget the K5-PR166 is a decent choice, but for the power user it is already too weak to be considered. The price of the K6 has come down so far, and the K6 is so much better than the k5,  that the latter is really only a good choice for a motherboard that does not support split voltage.

Note: The K5 sometimes scores poorly on the older processor benchmarks such as Norton SI 8.0. In fact, the PR120 scores lower than the PR100. This reflects more on the inaccuracy of the old benchmarks (when applied to a modern processor like the K5) than any weakness of the K5, and the fact that the internal core of the K5 was redesigned between the 100 and the 120. Compare to the 6x86, which has similar performance, or the SI32 scores for the K5 itself.

Look here for an explanation of the categories in the processor summary table below, including links to more detailed explanations.

Next: Sixth Generation Processors

Home  -  Search  -  Topics  -  Up