On Tuesday, our visit to AMD proved to be mostly a review of information we already knew, but as usual, it was nice to meet up with the guys and gals behind the Athlon. For those of you that haven’t been following the roadmaps, AMD has quite a future planned out for the Athlon as well as the Super7 platform. Let’s first start out with the future of Super7.

It seems as if the K6-III is going to finally go away, the rumors of a 500MHz part were reduced to just that -- rumors, and AMD is finding that the K6-III has no role in the future of the Super7 platform. The K6-2, however, will continue to be the "low-end" Super7 part and will even gain a clock speed increase towards the end of this year. The upcoming K6-2 part will be no different from the previous K6-2 chips we’ve seen (it will still be a 0.25-micron chip), and will feature a 95MHz FSB (with the last K6-2 being a 500, you can pretty much guess what this upcoming chip will be clocked at). Other than that, there is nothing new in store for the K6-2 in 1999.

The chip we’ve all been hearing rumors about was shown to us behind closed doors. The K6-2+ is the chip we’re talking about and it will soon take the place of the K6-III as the "high-end" Super7 CPU. While AMD is not willing to release the specifics of the processor, we do know that it is a 0.18-micron Super7 part, and features an on-die L2 cache. The size of the L2 cache will be less than that of the K6-III making 128KB the most likely option for the on-die L2. Other than that, there shouldn’t be many changes from the K6-2 featured in the K6-2+. The L1 cache is most likely going to stay at 64KB.

The K6-2+ we saw was used in a notebook setup in order to demonstrate one of its new features called Gemini. The Gemini technology is a lot like Intel’s Geyserville (Speed Step) in that it allows the frequency and core voltage of the CPU to be dynamically adjusted depending on the amount of battery power available (and whether or not the notebook is plugged into its AC adapter). The demonstration we saw showed the K6-2+ running at a 2.0v core voltage at its maximum frequency and at 1.5v when its battery power was limited. These may not be final core voltage figures, but it will give you an idea of what kind of motherboard support will be necessary for the K6-2+. According to AMD, a BIOS update is probably the only requirement for compatibility on current Super7 motherboards, but that has yet to be confirmed.

The chip we’re all wondering about is, of course, the Athlon. AMD is hoping to deliver 1 million Athlons this quarter, which is a high but reasonable goal. The future of the Athlon is a very bright one, and if AMD can deliver as they plan to, then the chip will most likely receive continued success throughout the first half of next year.

In order to compete on a clock for clock basis with Intel, AMD’s roadmap features an accelerated release of the 750MHz Athlon. The Athlon 750 will be based on the K75 core which differs from the original K7 core in that it is based on a new 0.18-micron fabrication process. Keep in mind that this new 0.18-micron core is not Copper based and still uses Aluminum interconnects. According to AMD, the move to 0.18-micron is going very well and they should have no problem producing enough chips for the market to consume. Also on the day of the announcement of this upcoming processor, systems will be available for sale based on it. While this may not seem like much, it has been something that AMD hasn’t been able to accomplish in the past few years.

In the first quarter of next year we can expect a follow-up to the 750MHz Athlon in the form of you guessed it, a 800MHz CPU based on the same K75 core. A 900MHz CPU should follow shortly thereafter. If you are skeptical about AMD’s ability to release an 800MHz or 900MHz Athlon, here is some food for thought. Remember just two months ago when the only 800MHz Athlon available was the Cool Athlon 800 by Kryotech? We managed to take a look at an air-cooled Athlon 800 and 900 running with just a standard heatsink and fan (although the fan was a fairly large unit). Using the 0.18-micron process, it is very possible for AMD to push these high frequencies, and don’t be surprised if they do even more once the yields on 0.18-micron CPUs improve.

Still in the first half of 2000, we will see the introduction of AMD’s Thunderbird CPU. The Thunderbird will be based on the same 0.18-micron K75 core from above and will feature a full speed, on-die L2 cache. We were afraid that AMD would decrease the size of the L2 cache of the Athlon when moving it on-die but apparently the size will most likely stay at 512KB. This will dramatically increase the transistor count of the already complex Athlon CPU and will definitely provide for a nice boost in performance across the board and will improve the performance boost you get at higher clock speeds (L2 cache speed scales directly with CPU speed).

The introduction of the Thunderbird will also mark the branching off of the Athlon into two distinct form factors. The first form being the standard 242-pin Slot-A that we were first introduced to the Athlon with, and the second a 426-pin Socket-A interface that will help to remove some of the cost of the already affordable CPU. AMD is following in Intel’s footsteps here by taking advantage of the fact that by moving the L2 cache entirely on-die, there is no longer a need for the Slot-A PCB. During the first half of 2000 the two CPUs (Slot-A & Socket-A) won’t differ in frequency or performance at all(they’ll be using the same CPU, just different physical interfaces).

The next CPU introduction will mark the Athlon’s entry into the low-cost market with the Spitfire. The Spitfire will be available only in the 426-pin Socket-A form factor and will feature a smaller on-die L2 cache than the Thunderbird (most likely 256KB). Other than that key difference (and cost), the Spitfire will remain identical to the 0.18-micron Thunderbird.

In the second half of 2000 we will see the introduction of the Mustang. The Mustang will feature, according to AMD, some "mobile features" which will most likely include the Gemini technology discussed above. It will feature a full speed on-die L2 cache of up to 2MB in size, making the CPU ideal for high end workstations and servers. It will be available in both a Slot-A and a Socket-A format and is still based on the 0.18-micron core, that will be introduced later this year. AMD also made it a point to mention that the CPU will feature an "Enhanced Core" but we are not certain about the specifics of those "enhancements."

A "value" Mustang will be available, only in a Socket-A format, and it will most likely have a smaller L2 cache than the high end version.

AMD briefly mentioned the SledgeHammer CPU which is their upcoming x86-64 CPU. There wasn’t much that they were willing to disclose, but the data we were given is summarized in the slide below:

This is AMD's entire roadmap presentation