AMD has truly made 2000 into a banner year for them.  Amidst skepticism as to whether or not they would be able to overshadow their numerous shortcomings and actually make the Athlon a successful alternative to Intel’s flagship processors, AMD has gone above and beyond the expectations of virtually everyone in this industry.  If you had talked about AMD as being a performance leader back at the release of their K6-III processor last year you would’ve been ridiculed.  However, with the most recent statistics showing that AMD has almost 40% of the performance desktop market, 24% of the market belonging to Athlon CPUs alone, it is obvious that things have turned around for AMD. 

However, when looking at the performance advantage the Athlon holds over the Pentium III in most applications and games, you quickly realize that AMD hasn’t earned its piece of the pie based on sheer performance.  In many cases, the performance difference between an Athlon and a Pentium III came down to a few percent, and in some cases the Pentium III even came out on top.  So why has AMD had such a wonderful year while we started out our article on Intel’s future by saying the exact opposite?  It truly comes down to the issue of execution and their roadmap. 

For the first time in quite a while AMD had executed with near perfect precision.  They were the first to the 0.18-micron process and the first to release a number of clock frequencies including the landmark 1GHz.  Just less than a month ago AMD launched their 760 chipset which brought Double Data Rate (DDR) SDRAM down to the performance PC level, and it won’t take long, as you’ll soon see, to bring that down to even lower cost price points. 

Three days ago we brought you a look at Intel’s plans for the remainder of this year and their roadmap for 2001.  And just yesterday we gave the keystone to their roadmap, the Pentium 4 processor, a very thorough look.  With almost all of Intel’s cards on the table, it’s time to take a look at AMD’s hand and see what they’re holding. 

Before we dive into AMD’s roadmap let’s point out a few very important topics to keep in mind regarding the future of both Intel and AMD:

• Intel’s roadmap is heavily dependent on two things: the execution of the Pentium 4 and the ramping of its clock speed.  If either one of these things fails to come through, then Intel will be in a very unfortunate situation. 
• As we discovered in our Pentium 4 Review, the processor itself requires quite a bit of attention for its true potential to be seen.  SSE2 optimizations are almost a requirement to see the Pentium 4 excel in quite a few applications.
• On the AMD side of things, the Athlon is running wonderfully, and when combined with a DDR platform its performance at 1.2GHz puts the Pentium 4 to shame.  Unfortunately the Athlon processors of today are simply getting too hot, and at 1.2GHz the Athlon already puts out more heat than the Pentium 4 at 1.5GHz. 
• The Duron is still not selling as well as it should be in North America.  Apparently the sales in Europe are incredible, however the Duron is still in need of a truly low cost platform to run on. 

With that in mind, let’s take a look at how AMD’s future stacks up.

The buck stops here: 1200MHz

As we alluded to in our 1.2GHz Thunderbird Review, the AMD Athlon won’t be surpassing the 1.2GHz mark before the end of this year and at Comdex AMD confirmed this assumption as well.  For them, as you can see by the performance of the 1.5GHz Pentium 4, it doesn’t make sense to get into a clock speed battle with Intel. 

We have always said that there is a large percentage of the PC buying population that buys based on clock speed alone, fortunately for AMD, once you take into account that not everyone can afford the most expensive 1.5GHz Pentium 4 systems, clock speed quickly loses ground to price.  This is what AMD is banking on for the rest of this year. 

According to AMD, if they were to release a higher clocked Athlon this late in the game it wouldn’t do much for availability in systems as no major manufacturers would be able to turn around entire system lines that quickly.  It seems very obvious that AMD is in a much different position than they were last year, since this same reasoning didn’t stop them from releasing an 800MHz Athlon on December 20^th, just 11 days before the end of the year. 

Indeed AMD is in a very different position this time around, they don’t have to fight with Intel as vigorously to gain ground, instead, they’ve got to fight to maintain their momentum.  How do they plan on doing it?  Let’s have a look at the Athlon and its future.

The AMD Athlon: Too hot for PC

As it stands, the Athlon is quite an attractive processor, especially in the retail market.  The perfect example of this is the fact that you can currently pick up a 1GHz Athlon, using the Thunderbird core, for under $300.  Because of its extremely competitive price, retail boxes feature the CPU in systems that range from $1,400 - $1,600 in overall system price (not including a monitor).  As far as the competition is concerned, a 1GHz Pentium III is almost $200 more expensive and can only be found in the most expensive retail systems, priced at $2,000 and above. 

The introduction of the AMD 760 chipset also brought about the introduction of three new Athlon processors, all of which support the 266MHz (133MHz DDR) FSB frequency.  These are the new 1GHz, 1.13GHz and 1.2GHz processors that are specifically made for use on AMD 760 boards or other upcoming boards that feature 266MHz FSB compliant chipsets such as the VIA KT133A and the ALi MAGiK-1.  As it stands now, these 266MHz FSB parts will be AMD’s performance market segment offerings, and have made their way into systems priced at $1,800 and above. 

The current Athlon CPUs, as you see them today, will carry AMD for the next couple of months into the start of next year.

We mentioned that AMD sees no need to increase the clock speed of the Athlon just yet, unfortunately there is another reason.  With the 1.2GHz Athlon already radiating 54W of heat, and a theoretical 1.4GHz Athlon producing an average of 72W of heat, you can see where AMD could run into a pretty major problem by ramping up clock speed too quickly. 

We’ve known for a while that the Athlon is a very hot running CPU, and with this heat being dissipated comes an increased amount of current draw, which puts a greater stress on the motherboard and the power supply.  While Intel was in a position to adopt the new ATX 2.03 standard and the accompanying ATX12V power supply specification which directly addresses the issue of supplying enough power to certain hotspots on the motherboard, AMD can’t possibly expect to do that with the simple introduction of a higher clocked Athlon.  This paints a picture of two problems with the Athlon, it’s running too hot, and it’s drawing too much power. 

Luckily AMD realizes this issue and has taken into account the need for a remedy.  Thus the majority of what will be going on in 2001 with their Athlon line will be surrounding the release of the successor to the Thunderbird core: the Palomino.

The Palomino core is ready, as we saw a demo of it in AMD’s suite at this year’s Fall Comdex running at no less than 1.5GHz (air cooled) and it will begin sampling by the end of this year with full production shipments commencing in the first quarter of 2001. 

Click to Enlarge

Provided that AMD sticks to this schedule, we should see a 1.33GHz (1333MHz – 266MHz FSB) Athlon based on the Palomino core debuting early next year.  The real question is, what does the Palomino offer in terms of enhancements over the current Thunderbird core?

Unfortunately, according to AMD, very little.  The biggest rumor that has been floating around, unconfirmed, was that AMD was going to be lengthening the pipeline of the Athlon core in order to allow for higher clock speeds.  While we mentioned that this is one method to paving the way for higher clock speeds in our Pentium 4 review, this isn’t by any means the only way to achieve a higher clock speed.  Needless to say, AMD won’t be making any changes to the Athlon’s pipeline for the Palomino core. 

As far as any other major architectural changes are concerned, AMD seemed very adamant in their statement that the only real changes to the Palomino would be related to power consumption and heat production. 

If this is indeed AMD’s position on the future of the Athlon they could be in a bit of trouble if Intel can execute and gain enough support for the Pentium 4.  A > 2GHz Pentium 4 with applications that can take full advantage of SSE2 at its disposal may be all that Intel needs to take advantage of this situation, provided that AMD does leave the Athlon’s architecture as-is.  But as you’re soon to see, AMD may have other things in mind for the future of their line instead of spending much time with tailoring the Athlon to compete with tomorrow’s Pentium 4. 

Before we get to that, let’s talk about the real issue here.  If the Palomino core, still based on AMD’s 0.18-micron process, is going to offer one advantage over the Thunderbird, its lower power consumption, how much lower are we talking?  According to the representatives we talked to, they were expecting at least a 20% reduction in power consumption and heat production.  If we take a look at the specs for a theoretical 1.33GHz Athlon based on the current Thunderbird core, and assume only 80% of its heat production, we can expect that a 1.33GHz Palomino would be dissipating around 55W of heat, around the same amount as a current generation 1.2GHz Athlon. 

This does mean that the Pentium 4 will continue to be the cooler running processor unless AMD can surpass their initial estimates which is quite possible, however greater than a 30% reduction in heat production seems unlikely without some major reworks of the core.  

By the second quarter of 2001 the Athlon, based on the Palomino core should be at speeds of 1.5GHz and greater, and by this time the entry level mainstream systems will be able to carry 1.1GHz Athlons (200MHz FSB) in the $1,100 - $1,300 price range. 

The end of 2001 will see the Palomino core reach speeds of 1.7GHz and above, potentially going as high as 2GHz, however this is very dependent on how Intel chooses to play their cards as well.  Also, sometime in the second half of 2001 AMD expects their entire performance/mainstream line to be composed of 266MHz FSB Athlon CPUs, with the lowest speed grade being 1.2GHz.  This means that DDR SDRAM should be very common by that time, and for most AnandTech readers, even sooner.

The Palomino will also be the first Athlon core that will be partially targeted at the single/dual processor server and workstation markets.  Although many have already been using the Athlons in such environments, the Palomino is a step in the right direction for AMD as the server/workstation market could definitely use a higher performance alternative to the Pentium III Xeon. 

The first quarter of 2002 will be met with the debut of the fifth Athlon core since the processor’s introduction late last year.  For those of you that can’t remember, the Athlon’s first core was the K7 (0.25-micron, off die L2 cache), followed by the K75 (0.18-micron, otherwise identical to the K7), then came the Thunderbird (0.18-micron, on-die L2), we just finished discussing the Palomino (lower power version of the Thunderbird) and next is AMD’s Thoroughbred core. 

Thoroughbred, quite possibly the weirdest core name we have ever heard that hasn’t come from VIA/Cyrix, is the next logical step for the Athlon.  It will use AMD’s upcoming 0.13-micron fabrication process, a die shrink which will decrease power consumption yet again while allowing for even higher clock speeds.  There is very little that is known about the Thoroughbred and quite a bit about it could change between now and its official release in over a year, we’ll be sure to keep you posted on anything we discover. 

AMD’s Athlon hits the road

With all of this focus on decreasing power consumption and die shrinks over the next year, it looks like we will finally see notebooks from AMD based on something other than the K6-X line of processors. 

Intel’s mobile Celeron and Pentium III processors have been trampling all over AMD’s K6-X line of mobile CPUs.  You definitely don’t see the level of acceptance of AMD’s mobile CPUs as you do of their desktop CPUs.

Luckily, along the same time as the Palomino’s release, a mobile version of the Athlon based on the Palomino core will also hit the streets.  Mass production of the mobile Athlon based on the Palomino will occur sometime before the second quarter of 2001. 

The chip will remain relatively unchanged from its desktop counterpart; it will just run at lower clock speeds and will feature PowerNow! support.  Don’t expect to see any 266MHz FSB mobile Athlon parts though. 

The launch of the Thoroughbred core will also mark the launch of the mobile version of a 0.13-micron Athlon based on the same core.  Just like the mobile Palomino’s relationship to its desktop brother, the mobile Thoroughbred will simply run at lower clock speeds and have PowerNow! support.

With ATI and NVIDIA both bridging the gap between desktop and mobile 3D graphics acceleration and AMD, Intel and Transmeta working on improving battery life along with performance it won’t be long before traveling will be much more entertaining experience. 

The Mustang rides off into the sunset

As you all have probably heard by now, AMD has cancelled their Mustang which was supposed to be another Athlon core that is specifically aimed at the workstation/server market.  The Mustang core is still on AMD’s roadmap in spirit as the Palomino is a derivative of Mustang, but instead of having two separate cores for the performance desktop and server/workstation market, AMD will just be sticking to the Palomino. 

The only advantage the Mustang would offer over Palomino would be a larger L2 cache, but AMD has something else planned for the niche market that demands that.

In the first half of 2002 AMD will be debuting their x86-64 architecture, which we explained in our article entitled AMD’s Voyage into the 64-bit Arena: x86-64 Revealed.  The x86-64 processor we area all familiar with is the SledgeHammer, designed mainly to compete with the Itanium by offering the benefits of a 64-bit architecture while offering the same level of 32-bit application performance as the Athlon. 

The second processor that AMD has announced will use their x86-64 architecture is the ClawHammer which will actually precede the SledgeHammer’s release in the first quarter of 2002.  Very little is known about ClawHammer other than it will be aimed at single or dual processor workstation and server systems and a very small portion of the performance desktop market. 

The ClawHammer will offer the same 32-bit performance as the Thoroughbred, but will add 64-bit support courtesy of its x86-64 implementation.  The ClawHammer will debut at speeds of 2GHz, conveniently enough just shortly after Intel is scheduled to surpass the 2GHz mark with their Northwood Pentium 4 which is still a 32-bit only processor.  For most users, the 64-bit capabilities of the ClawHammer will remain unused, however it will be perfect for those individuals that have a need to go down Intel’s Itanium route but are holding off on it because of a continued need for solid 32-bit performance. 

An interesting note about the 0.13-micron ClawHammer is that it will have a die size of under 100 mm^2.  This is smaller than the original K7 core without any on-die L2 cache.  If the ClawHammer is going to be this small you can expect to see the Thoroughbred in a similar size range as well. 

Super7’s Replacement

It was only a matter of time before AMD’s Duron began to threaten the domain once held by the K6-X processors.  The value market was AMD’s where AMD was gaining ground last year, but as we announced at the start of this article their focus has changed as the Athlon is definitely not a value desktop part. 

AMD’s goal for 2001 is also to regain some of that value market share that was lost to Intel’s Celeron.  We have seen time and time again that the Duron is clearly a better processor when compared to the Celeron, and is much faster on a clock for clock basis, however, at least in North America, the Duron hasn’t seen the kind of retail support that Intel’s Celeron continues to enjoy.

Without any major system design wins for the Duron, the Celeron is allowed to dominate in the retail market.  A big factor that has kept the Duron out of many systems is that it is lacking any low cost platform with integrated video to run on.  To most hardware enthusiasts this is actually a positive point about the Duron, since we all know how slowly Quake III will play on the “fastest” integrated video chipsets.  But in the retail market, when a manufacturer is trying to sell a system at the lowest possible price to a user that’s only going to be really surfing the net and typing up documents in Word, video performance isn’t the top priority. 

We’ll address AMD’s chipset roadmap later on to see if the Duron will be paired with a low cost platform, but before we do that let’s have a look at the future of the processor itself.

By the end of this year the K6-X line of Super7 processors will be phased out, and the Duron will be allowed to grow into the value mobile market as well as the value desktop market that it has been developing in as well. 

If you recall back to our initial review of the Duron back in June you’ll remember that the Duron is a direct derivative of the Athlon’s Thunderbird core.  With the Thunderbird due to be replaced by the Palomino core it would make a lot of sense for the Duron to get a new, cooler running core that is based on the Palomino core as well. 

So in the second quarter of 2001 AMD will release the Morgan core, it will feature the same 192KB total on-die cache as the current Duron (128KB L1 + 64KB L2) and will run approximately 20% cooler than the current Durons that are using the Spitfire core. 

The Morgan will also be going mobile around the same time and it will replace the mobile K6-2+ and K6-III+ CPUs in the value mobile market segment.  Just as with the mobile Palomino, the mobile Morgan will be identical to its desktop counterpart while adding PowerNow! support and will ship at lower clock speeds. 

The final step for the Duron on AMD’s current roadmap is the transition to the Appaloosa core, which is the Duron derivative of the Athlon’s Thoroughbred core.  This core will be due out in the second quarter of 2002 and will be using AMD’s 0.13-micron fabrication process just like the Thoroughbred. 

AMD is still not a chipset manufacturer

In spite of the recent release of the AMD 760 chipset, AMD continues to maintain that they are not a chipset manufacturer and they will only produce chipsets when they need to drive technology for their processors. 

This makes quite a bit of sense for AMD, unfortunately it also means that the success of their CPUs is greatly dependent on third party chipset manufacturers, which isn’t a situation that any CPU manufacturer wants to place themselves in.  If you remember, the fact that Intel placed their entire CPU line in the hands of a single memory manufacturer ended up hurting them quite a bit.  While AMD has done this for quite some time, the company may be getting to the point where their execution may be hindered by these third party chipset manufacturers.

The perfect example is the poor adoption of the Duron.  Without an equivalent of the i810 chipset the Duron was kept out of many retail systems where it would’ve enjoyed incredible success since it not only outpaced the Celeron in performance but in clock speed as well.  When Intel launched the Celeron 766 processor we asked them why they hadn’t moved to the 100MHz FSB earlier, and their reasoning was simple.  Even though the Celeron is slower, both in clock speed and in overall performance, they weren’t really losing much sales to the Duron because of the very fact that the Duron still has no reliable integrated chipset to run on.  It’s unfortunate, but it’s the truth for most of the market (remember, tweakers and overclockers only make up a small portion of the overall market but we’re growing in numbers, beware). 

The SiS 730S chipset used in a FlexATX system with a Duron CPU

Click to Enlarge

AMD will remain very quiet in terms of chipset releases in 2001 and on into 2002.  The only chipset they will be releasing is the multiprocessor version of the 760, not unusually named the AMD 760MP chipset. 

Click to Enlarge

The 760MP, which was on display in AMD’s suite at Comdex, will pretty much be a dual processor version of the 760.  The 760MP is physically a different chip than the 760 since AMD’s point-to-point EV6 bus protocol states that each processor shall have its own data path to the North Bridge.  While this means that, in the case of a 266MHz FSB, each CPU in a dual processor AMD system will have a full 2.1GB/s of bandwidth to/from the North Bridge, it also means that motherboards will have to include twice as many traces to the North Bridge making the motherboards very expensive. 

We’ve known for over a year now that Tyan has been a testing site for the 760MP chipset, and word finally got out that they made the 760MP board that AMD was showing off at Comdex.  You can expect a 760MP board from them sometime early next year which should work with all current Socket-A CPUs.  A dual Morgan system should be a killer entry level workstation, not to mention what two Palominos could do for a server. 

The rest of AMD’s chipset roadmap, as you can see from the chart below, is entirely dominated by VIA, SiS and ALi. 

Click to Enlarge

We have ALi’s DDR solution on the way, and we already have SiS’ solution which we have been playing with for a little while now.  The most interesting thing on the roadmap is the fact that DDR SDRAM will become a value option towards the second half of 2001.  That will do wonders for the low end market as you can guess, but once again AMD’s entire chipset lineup is dependent on the proper execution by VIA, ALi and SiS.  The latter two haven’t had the best track records and VIA has historically had some memory performance issues with their controllers, so it will be interesting to see how things turn out next year for AMD. 

Final Words

AMD has had a very strong year, there is no doubt about that, however with the Pentium 4 able to ramp up to some very high clock speeds and already running much cooler than the Athlon at equivalent clock speeds.  Not to mention what could happen if SSE2 gets the quick adoption that Intel is hoping for. 

Click to Enlarge

The Athlon’s potential is quite great, however it’s going to take much more than a cooler running core to allow AMD to continue their growth in the performance desktop market.  We’ve seen the benefits that a very powerful branch prediction unit can give the Pentium 4, and it is widely known that the Athlon’s unit is holding it back in many cases. 

The Duron is on track to be our low cost processor of choice throughout 2001, but for AMD’s sake it needs much better chipset support than what it currently has.  As a mainstream solution the Duron needs little more than the KT133 and AMD 760 platforms that it has at its disposal, however in the sub $1,000 system market, especially when you get down to systems priced at under $600 there is a definite need for a highly integrated platform. 

While two solutions are currently on the way, VIA’s KM133 and the 730S from SiS, the question is how comfortable do you feel with the Savage4 core of the KM133 or SiS’ solution considering that they have never made an Athlon chipset before. 

Just as Intel’s success in 2001 is very dependent on the proper execution of their platforms as well as their CPUs, AMD will find themselves in a similar situation, however now with all of Intel’s cards on the table it will be interesting to see whether the Athlon gets an architectural improvement or not next year in order to continue its dominance.  The wrong decision on AMD’s part, or poorly executed third party chipsets could give Intel the chance to regain a lot of lost ground next year.  At the same time, if AMD repeats the level of success they have attained in the past 9 months alone, Intel could be in some very serious trouble.