The date?  April 2, 1997.   The place?  Advanced Micro Devices' Homepage.  The occasion?  The release of the AMD K6 Microprocessor.  The K6 was AMD's last chance at redemption, the delays the K5 blessed the market with combined with the too-little too-late performance the K5/166+ boasted left AMD with one and only one chance to get back into the game.  Slowly losing their ground against the Pentium II, which was released just about a month later, AMD went back to work on a revision to the K6.  Originally expected for a much more practical release in November '97, AMD planned on releasing a 266MHz

version of the K6 followed by a 300MHz part before the arrival of the new year.

Once again delays, fabrication problems, and much more delayed the release of the 266MHz addition to the K6 line.  Now, almost a year later, the first K6/266 parts are finally being made available to the public and not just OEMs that had been granted the privileges for the past few months.  This new processor doesn't come without its drawbacks, the main downside to going with the K6/266 would be that you're funneling money into a dying world, Socket-7.  The phrase "would be" is used because a few steps by some bold manufacturers have prevented the Socket-7 market from fading into oblivion. 

AMD has announced yet another addition to the Socket-7 K6 line, the K6-3D 300.  The K6-3D will not only be the first Socket-7 Processor to officially support the 100MHz Front Side Bus Frequency but it will breathe more life into the existence of the Socket-7 Motherboard by setting a new standard known commonly as the Super7 Motherboard Standard.  

In order to be classified as a Super7 Motherboard, a board must meet these three requirements:

• AGP Support

• 100MHz Front Side Bus Support with Support for 33MHz PCI & 66MHz AGP Bus Speeds

• Support for all Super7 processors, particularly the AMD K6-3D (Core Voltage - 2.2v, clock multipliers up to 4.5x)

What is the point of mentioning the K6-3D now?   The K6-3D will essentially be the K6/266 that is available for purchase today, with the addition of the AMD 3D Instruction Set, designed to improve 3D Geometry and Floating Point Calculations.  If you are looking for a processor now, and simply can't wait for AMD to crank out the K6-3D, then the K6/266 is your obvious choice.  So how about a closer look at the K6/266?

What makes the K6/266 different from the original members of the K6 line that made their debut on April 2 last year?   First of all, the die of the K6/266 (the chip itself, not the external casing) is manufactured using a 0.25 micron circuit size, compared to the 0.35 micron die size of the previous K6 chips.  Not only does this allow the K6/266 to run at higher clock speeds, opening the world of Socket-7 Overclocking to an entirely new universe, but this die shrink virtually eliminated the heat problems the original K6's gave users, especially when overclocking.

This die shrink, walking hand in hand with the newly lowered core voltage of the K6/266 (2.2v compared to the 2.9v setting for the 166/200MHz K6's and the 3.2v setting for the 233MHz K6's) make the 266 everything the original K6 should have been...and then some.

The K6/266 accepts clock multipliers from 2.0x up to 5.0x, with some 266's being reported as even supporting the 5.5x clock multiplier.   Using the 66MHz bus speed, and the 2.2v core voltage setting the 266 can be taken as high as 300MHz...everything else becomes a little more complicated, so let's eliminate some of those complications.

Overclocking the K6/266 deviates from the conventional overclocking methods for Socket-7 processors.  With the old K6's or the Pentium MMX you could simply take the core voltage up to about 0.6v above the recommended levels and achieve greater stability, provided that you could cool the chip properly.   This is definitely not the case with the K6/266.  All voltages above 2.4v - 2.5v should be considered unusable with the K6/266, and you must keep in mind that just because a motherboard states that it has a 2.2v voltage setting it doesn't necessarily feed the processor 2.2v, rather something *around* that number.  This is where the quality of a motherboard can really come into play, your decision to skimp on your motherboard to save a few bucks may just come back to haunt you.

300MHz can be achieved without increasing the core voltage of your CPU much (if at all), at most you will probably have to take the CPU up to 2.3v but no more.  Booting at 333MHz (66 x 5.0) is a bit of a trick, it requires a core voltage of a little more than 2.3v, therefore the key to running your K6/266 at 333MHz is the use of a 2.4v core voltage setting.  All of these settings take for granted that you don't have the 75, 83.3, or 100MHz bus speeds at your disposal...if you do...then things start to get interesting.

The 66MHz Bus Speed is nearing its complete death, the rest of the PC Hardware World is waiting at the cemetery for the Intel and AMD funeral cars to arrive, and they're finally appearing over the horizon.   The K6/266 will probably be the last 66MHz-only Socket-7 CPU from AMD, making the transition from 66MHz to 100MHz couldn't have been made any easier than the K6/266 makes it. 

Officially not supporting any bus speeds above 66MHz it would be silly not to try the K6/266 at some of the more conventional overclocked bus speeds.  Clocking the K6/266 at 75 x 4.0 (300MHz) yields a performance increase of about 3 - 5% over running your system at 66 x 4.5 due to the increased bus speed.   This may not seem like much, but the processor's stability at the 75MHz bus speed makes you feel like you aren't even overclocking the processor, especially if you make use of a high quality

Socket-7 motherboard.  75 x 4.5 (338MHz) requires a slightly heavier touch than 300MHz, a 2.3/2.4v core voltage setting should do the trick however.   No real problems here.

The 83.3MHz bus speed adds a little bit of difficulty to overclocking with the K6/266, for all settings above 266MHz (83.3 x 3.5, 83.3 x 4.0...) it is almost a requirement that you increase the core voltage to at least 2.3v.  Some signs that your processor isn't receiving enough voltage are:

• Doesn't complete memory count, all other components are known to work properly

• Freezes at "Starting Windows 95..." prompt

• Random crashes during normal Windows operation without overclocking the bus speed, and all other components are known to work properly.

Some motherboards falsely identify voltage settings as 2.1/2.2v settings, when in actuality they are more likely to be around 2.5v or 2.7v.  The first MTech R581-A I tested wouldn't boot with a K6/266 installed, however a newer revision (B3) worked perfectly fine and supplied the necessary 2.2v to the CPU.  

If you happen to be good with your hands you may want to consider this approach a FIC PA-2007 user proposed on the Anand Tech BBS: (warning: Anand Tech assumes no responsibility for damage caused to your system by following the below procedure, do so at your own risk):

-> I was really upset that my PA-2007 at 2.1V would probably not support the K6-266 so I spent some time today investigating further. I found out that the PA-2007 could get 2.2V please read further and find out how:

FIC PA-2007

First a few Warnings:

Remember you do everything at your own risk. If you don’t know exactly what you are doing, ask someone with experience for advise. I have done my best to avoid any mistakes in the following documentation and I cannot guarantee you will get the same results I have.

CPU Core Voltages

The core voltage can be set between 2.1V and 3.3V in 0.1V increments without soldering on your motherboard.

Motherboard Layout

This the FIC PA-2007 rev 1.2 component layout for the VR jumper block.

VR
3.5V 1 O---O 2
3.3V 3 O---O 4
3.2V 5 O---O 6
2.9V 7 O---O 8
2.8V 9 O---O 10
2.1V 11 O---O 12

I measured the core voltage settings at the R267 resistor on the side closest to the CPU socket. The core voltages can only be measured with a split voltage CPU in the CPU socket and in my case I used a Pentium MMX processor. First, I made sure that when I shorted the VR jumper position 9-10 the voltage measured 2.8V and then I shorted the VR jumper position 11-12 and confirmed the 2.1V setting. I verified all voltage settings with a digital multimeter. Then I began by doing multiple jumper shorts and I tried various combinations while measuring the core voltage output at each point and documenting it all. Following are the results:

VR jumper positions

To achieve some of the voltages it is necessary to solder a resistor onto a little connector ( I found that the turbo switch connector in my PC case would fit on the VR jumpers so I cut it off and soldered the resistor to it - it worked just fine.). I admit there is some work involved in soldering the resistor onto the connector but it will be worth to get at the 2.1V to 3.3V core voltages in steps of 0.1V

Core voltage Measured voltage Jumper positions

2.1V - 2.123V:  11-12 short
2.2V - 2.196V:  11-12 shorted with a 47K ohm resistor
2.2V - 2.251V:  11-12 shorted with a 100K ohm resistor
2.3V - 2.340V:  11-12 short & 9-10 short
2.4V - 2.414V:  11-12 short & 7-8 short
2.5V - 2.501V:  11-12 shorted with a 100K ohm resistor & 9-10 short
2.6V - 2.625V:  11-12 short & 5-6 short
2.7V - 2.735V:  11-12 short & 3-4 short
2.8V - 2.791V:   9-10 short
2.9V - 2.899V:   7-8 short
3.0V - 3.072V:  11-12 shorted with 100K ohm resistor & 9-10 short & 5-6 short
3.1V - 3.134V:  11-12 short & 9-10 short & 5-6 short
3.2V - 3.205V:   5-6 short
3.3V - 3.301V:   3-4 short

As you can see the nominated core voltage setting is not always the exact voltage i.e. 2.1V is not exactly 2.100V but is actually measured to be 2.123V. <-

Antonio Cataldo, the person that proposed the above changes made an interesting point, if our motherboards don't support the settings we need...why not make them?  Using the above fix, the PA-2007 Mr. Cataldo purchased a few months back, gets a few more months of living.  This is a topic that will be addressed in greater depth later on, looks like your webmaster will be making a trip down to Radio Shack ;)

Here's the big question, which CPU should you buy, a Pentium II - 266 or a K6/266? 

Cost

If you're thinking about cost, the Pentium II - 266 is about $20 - $50 more than a K6/266, as far as motherboards are concerned expect to pay at least $30 more for a Slot-1 LX motherboard when compared to an average priced Socket-7 AGP board.  The final difference in price you'll see when buying a Pentium II - 266 over a K6/266 will probably be, at most, $100...and depending on how much you shop around, that difference could be taken down to as low as $60.  If you already have a motherboard that does support the K6/266 completely, then the obvious choice is a K6/266.  However if you either don't have a K6/266 compliant motherboard, or are making a fresh upgrade the choice becomes a bit more difficult.

Overclocking

With the price factors virtually eliminated, we must now consider reliability/stability of the K6/266 in comparison to the Pentium II - 266.  You can almost bet on buying a clock-locked Pentium II - 266 CPU now, with the highest achievable clock multiplier setting on the processor being 4.0x.  AMD's K6/266 on the other hand has yet to be clock-locked, with the highest reported clock multiplier ever used was 5.5x.  However a clock multiplier means nothing if you cannot put it to good use.  For example, chances are that you won't be able to clock a K6/266 at 66MHz x 5.5 too easily.

We have already established that the K6/266 can make it up to 333/338MHz fairly easily on most motherboards, and provided you have decent cooling (boxed heatsink/fan would do just fine) the Pentium II - 266 can make it up to 338MHz without any problems.  With the Pentium II you will probably have an easier time since the processor's voltage settings aren't controlled by the user, and don't need to be adjusted for overclocking purposes.  The K6/266 may give you a few problems here and there with overclocking, but once you achieve that point of stability the processor should be fine.

100MHz Bus

The Pentium II - 266 and the K6/266 are both able to run at the 100MHz bus speed, albeit an unofficially supported "feature" of the two processors.  The Pentium II - 266 can be used in conjunction with the ~92MHz Bus Speed setting on the ABIT LX6 to provide you with stellar performance, unfortunately the LX6 doesn't provide the user with an option for a 1/3 PCI Clock Divider, meaning your PCI Bus is put under tremendous stress as you increase your Memory Bus Speed.  Newer Socket-7 and the upcoming Super7 boards on the other hand do support an asynchronous PCI/AGP Bus Clock (~32/64MHz respectively), making the 100MHz bus speed much easier to use with one of these motherboards. 

It has been proven that Socket-7 systems receive a much larger performance boost from the 100MHz bus speed when compared to Slot-1 systems, simply because the L2 cache of Socket-7 systems receives a fairly large performance boost as a result of the 100MHz operating frequency.  Slot-1 systems don't derive the L2 cache speed from the system bus speed, rather the processor's internal clock, making the 100MHz bus irrelevant to L2 cache speed in Pentium II systems.

Can the K6/266 operate at a 100MHz bus frequency?  Short answer, yes, it can, but can it in all cases?  Nope.  The K6/266 wasn't designed for use at the 100MHz bus frequency, although it does work at the bus speed much more reliably than any other Socket-7 processor.  Using the MTech R581-A, the K6/266 managed to boot at 100MHz x 2.5 and 100MHz x 3.0, unfortunately Windows 95 would not operate with enough stability in order to consider it a reliable setting.  There have been success stories across the net with the K6/266 running at the 100MHz bus speed, as well as benchmarks, which prove that the K6/266 can operate at the 100MHz bus speed, however you must remember that not all K6/266's are created equal...neither are all motherboards.   What does this mean?  Basically, it's a risk, don't expect the K6/266 to work flawlessly at the 100MHz bus speed, then again don't expect it to crap out at 100MHz.   At the 90MHz setting on the MTech board Windows was much more stable, unfortunately the test system was plagued by Windows 95 registry corruption during normal operation.   Again, there have been success stories with the K6/266 and the R581-A at both the 100MHz & 90MHz bus speed settings, and in the tests conducted here the K6/266 did manage to run Winstone at least once however for the most part the setting was unusable for even light operation (basic Windows tasks, copying, deleting, etc...). 

Intel or AMD?

As mentioned above, if you currently have a motherboard that does support the K6/266, then the choice is clear, go with the AMD chip.   In the tests Anand Tech performed, the K6/266 came in about 6 - 7% slower than an equivalently configured Pentium II - 266 LX system in Business Winstone tests, and about 10 - 12% slower than the Pentium II in High End Winstone tests.  At 300MHz (overclocked) the K6 still fell behind the Pentium II in most tests, however it did manage to cut the gap by a few percentage points. 

If you want the best possible Quake 2 performance, the K6/266 still doesn't reach Pentium II performance levels, even with a Voodoo², so if you're a Quake 2 fanatic then the Pentium II - 266 should be your choice.  Of course, if you're a die hard AMD fan (or prefer not to buy Intel processors) then the K6/266 is an ideal choice, in most cases however, the Pentium II - 266 is a better buy.  Provided you buy a current revision of the Pentium II - 266, chances are you'll able to clock it at 100 x 3.0 on a BX Board come April 15 of this year.   Considering that the Pentium II - 266 is a chip that's almost begging to be overclocked with a higher bus speed, and the K6/266 isn't as overclockable (in terms of the percentage of success rates) the Pentium II - 266 is a better choice in this case.

Conclusion

The best way to phrase the overall quality K6/266 is this: A processor that would've been great had it been released 5 months ago, unfortunately the performance and physical enhancements are just too little too late on AMD's part.  Hopefully the K6-3D won't fit this mold quite as well...we can only wait and see.

Performance & Compatibility List