In the twelve months that made up the year 2000, there were a total of fifteen retail AMD Athlon processors that were announced, released to the public and shipped.  Of those fifteen Athlon processors, four were based on the old K75 core while the rest were based on the Thunderbird core announced in June of that year.  Three of the eleven Thunderbirds were Athlon-C processors, meaning they featured lower clock multipliers and used the 133MHz DDR FSB.  If you haven't gotten the hint by now, AMD released quite a few Athlons over the course of last year alone.  Interestingly enough, we haven't seen a single Athlon from AMD since October of 2000, just before Intel's Pentium 4 launch.

That fact isn't too surprising actually, seeing that the Pentium 4 was one of the most frequently criticized CPUs we had ever come across.  We were even pretty harsh on it upon its release, simply because the Athlon offered greater performance at a much lower cost while still making use of PC133 SDRAM.  It didn't help Intel's plight that the majority of the public views Rambus, their business partners with the Pentium 4 launch, as one of the more hated companies. 

AMD's initiation process into the mainstream and performance desktop PC market segments is complete.  No longer do OEMs and system integrators have to worry about AMD delaying their processor releases by 9 months nor do they have to worry too much that the yields on AMD CPUs are going to be low enough that they may cause problems for their customers.  The AMD of 2001 is a much more elegant, competitive, and obviously powerful force when compared to what it was just three years ago.  But in many ways, the AMD of 2001 has an entirely new set of worries today than they did three years ago.

Intel has been preaching that their Pentium 4 and its NetBurst Architecture provides a platform for the future, and that the performance of today's applications and games is a secondary concern in comparison to the performance the platform will offer further down the road and in many ways Intel is right.  History has shown us that as PCs get more popular, the software run on them become increasingly more demanding in areas such as memory bandwidth for example.  And as we move further down the road of PC evolution, you can expect there to be some actual use of the Pentium 4's 3.2GB/s of memory bandwidth, more than what currently exists today. 

Which brings up the argument in favor of AMD, and the reason that the most popular system configuration among hardware enthusiasts "in the know" happens to be an Athlon using no more than regular PC133 SDRAM.  Two or three years down the line, an Athlon with "only" 1.06GB/s of memory bandwidth will definitely be bandwidth limited, just like the Celerons with PC66 SDRAM became memory bandwidth limited as applications became more demanding.  When push comes to shove however, how many enthusiasts keep their system configuration static over the course of 2 - 3 years?  That is the question you've got to ask yourself when making this CPU and platform decision moving forward, all we can do is aid you in the process. 

With that said and done, today, AMD cranks up the release factory yet again from its five-month hiatus and graces us with the release of two new Athlons: the 1.33GHz and 1.30GHz parts based on the Thunderbird core; that's right, no Palomino just yet.

Palomino where are you?

The codename Palomino was first introduced to us at Fall Comdex 2000 which we later passed on to all of you in our AMD CPU & Chipset Roadmap for November 2000.  The Palomino core is the successor to AMD's 9-month-old Thunderbird core and, according to AMD, offers no major improvements to the Thunderbird design other than it is a cooler running core.  This cooler operation will be met with much enthusiasm because of the fact that the Athlon's Thunderbird core, operating at 1GHz, generates more heat than a Pentium 4 running at 1.5GHz. 

Fast-forwarding to the present day, the two new CPU's being released here are both being clocked at speeds approximately 33% higher than the 1GHz Athlon we just used as an example.  If you thought the current Athlons were hot, you're about to be in for a "warm" treat. 

The 1.33GHz Athlon can draw a maximum of 42 amps of current (MAX I_cc) while the 1.30GHz Athlon's MAX I_cc is 40 amps.  In terms of heat dissipated, the 1.33GHz and 1.30GHz parts will radiate more than 64W of heat on average and up to ~72W. 

Power & Thermal Specification Comparison
CPU	Core Voltage	Maximum Current Draw	Maximum Thermal Dissipation
AMD Athlon 1.33GHz	1.75v	42A	73W
AMD Athlon 1.30GHz	1.75v	41A	71W
AMD Athlon 1.0GHz	1.75v	31A	54W
Intel Pentium 4 1.5GHz	1.70v	43A	52W

According to AMD's roadmap that was presented at Comdex, the Palomino was originally set to debut in mobile parts at the end of 2000 with the desktop version following in the first quarter of this year. 

Click to Enlarge

We later provided an update to that roadmap that said not to expect the desktop Palomino until much later.  To quote our CPUs in 2001 article, we said the following:

"In February/March the Athlon should gain another clock step or two taking it up to 1.33GHz and possibly up to 1.4GHz.  Both of these CPUs will still be based on the current Thunderbird core and they should generate a considerable amount of heat.  Luckily the 1.4GHz Athlon should be the last Thunderbird based processor for AMD, paving the way for the 1.5GHz Athlon based on the cooler running Palomino core to be released sometime in the May - June timeframe."

Recent information seems to indicate that the Palomino could be released even later than that, it seems like Palomino could turn out to be more than just a cooler running core. Let's hope so.

On the mobile front, the Palomino could be witness to a delayed introduction there as well since there seems to be a lack of mobile chipsets for the part. 

The Chips 

The reason for launching two CPUs today is simple; since 133MHz DDR FSB platforms, such as the AMD760 and the VIA KT133A, are not too widespread yet AMD has to release both a 133MHz FSB processor and a 100MHz FSB part so that they can offer an even faster performing platform as well as a 100MHz FSB part for compatibility. 

If you'll remember, this is actually the opposite of what Intel did when they were ramping up their first 133MHz FSB parts.  Instead of continuing to produce 100MHz FSB Pentium IIIs, Intel decided to stop making 100MHz FSB Pentium IIIs faster than 850MHz and focus entirely on 133MHz FSB parts.  Later on, after the majority of the market had transitioned to 133MHz platforms, Intel released 900MHz and 1GHz parts that used the 100MHz FSB for late adopters. 

It would be advisable for AMD to mimic Intel in this respect if they plan on eventually moving to the 133MHz DDR FSB for all of their Athlon processors, however because AMD does not have a widely available chipset of their own (AMD760 isn't a mass-produced chipset intended for the entire market) that supports the 133MHz FSB this isn't really an option.  AMD's continued dependence on ALi, SiS and VIA for the majority of their platforms may end up hurting them in the long run if they ever do decide to take on a more active role as a technology leader in the processor market. 

CPU Specification Comparison
	AMD Duron	AMD Athlon	Intel Pentium 4	Intel Pentium III	Celeron
Core	Spitfire	Thunderbird	Willamette	Coppermine	Coppermine 128
Clock Speed	600 - 850MHz	750 - 1333MHz	1.3GHz - 1.5GHz	500 - 1000MHz	533 - 800MHz
L1 Cache	128KB		8KB Data	32KB
L2 Cache	64KB	256KB	256KB	256KB	128KB
L2 Cache speed	Core Clock
L2 Cache bus	64-bit		256-bit
System Bus	100MHz DDR EV6	100/133MHz DDR EV6	100MHz quad-pumped GTL+	100/133MHz GTL+
Interface	Socket-462 (Socket-A)		Socket-423	Slot-1/Socket-370	Socket-370
Manufacturing Process	0.18-micron
Die Size	100mm^2	120mm^2	217mm^2	106mm^2	106mm^2
Transistor Count	25 million	37 million	42 million	28 million	28 million

The 1.33GHz part runs at 1333MHz using a 10.0x clock multiplier and a 133MHz DDR FSB.  The 1.30GHz part runs at 1300MHz using a 13.0x clock multiplier and a 100MHz DDR FSB.

The first thing we noticed about the 1.33GHz part is that the chip has actually changed slightly when compared to previous Athlons.  If you'll remember, the Thunderbird core has 256KB of on-die L2 cache compared to the 64KB L2 on the Duron.  This makes the Thunderbird core noticeably larger than the Duron core; the difference in size is further illustrated by a pattern of dots to the right of the core.  The above picture provides the comparison.

The Thunderbird core itself does have a smaller pattern of dots to its right, which originally led us to believe that a future version of the Athlon could have a larger L2 cache.  AMD's roadmap at the time confirmed this with a processor that carried the codename Mustang, a larger cache version of the Athlon.  Since then however AMD has cancelled the Mustang, and if you'll take a look at the new 1.33GHz part we have below, you'll see that the pattern of dots has disappeared as well. 

"old" Athlon 1.1GHz

"new" Athlon 1.33GHz

It looks like the Athlon is not going to get any more than 256KB of L2 cache until the next die-shrink, which will be to a 0.13-micron process in the first half of 2002. 

It's time for new Tests

In order to make this CPU review more interesting, we've added a few new tests into the mix that will allow us to get a broader picture of the performance of all of the CPUs.  In addition to our usual Winstone, SYSMark and Viewperf numbers we are introducing a few new tests. 

This is the second time we are using CSA Research's Benchmark Studio, this time using Beta 2.0 of the suite.  From our ServerWorks III HEsl Review, where we first debuted the Benchmark Studio suite, here is a quick description of what makes it stand out:

"The package is called Benchmark Studio and its key component is called Office Bench. The beauty of the way Office Bench works is that it not only performs the normal tasks any Business Winstone-like benchmark would (working with MS Word, Excel and Power Point) but it also can work with Benchmark Studio to simulate other types of load. The types of load that the Benchmark Studio can simulate range from accessing databases to checking email and streaming video. Using the Benchmark Studio interface you can completely customize how many instances of each type of load you'd like to create and when they loop."

Beta 2.0 of Benchmark Studio is simply a more stable build of the suite; the metrics and load simulators remain the same.  We only used one suite configuration; with a total of 13 load simulations running which the Office Bench script ran.  The score that was outputted was in seconds, the lower the better.

The Benchmark Studio scores should be an indication of how users in the IT world as well as power users at home would benefit from the individual processors and platforms. 

MP3 Encoding

In spite of the recent injunction against Napster, MP3s are still as popular as ever and they will only continue to grow in popularity.  Going along with this, we have included two timed audio encoding tests in the comparison.  Using the MusicMatch Jukebox 6.0, an audio encoder that uses the Fraunhoffer codec.  We used a 100MB WAV file and timed its conversion to a Variable Bit Rate (VBR) MP3, with the quality/compression setting at 75%.  We also used the same 100MB WAV file and timed its conversion to a 128-kbps Windows Media Audio (WMA) file. 

MP3 encoding is a good test of FPU and cache performance, since the process is not as system memory bandwidth focused as video encoding.

New Tests (continued)

WebMark 2001

When BAPCo released WebMark 2001 we were highly skeptical of its usefulness, how often do you complain about how slow your CPU is in rendering a web page?  It turns out that WebMark 2001 is much more than a series of timed web page loads, it's actually more of a technology benchmark.  The suite takes into account performance under commonly used applications such as Acrobat Reader, Quicktime, Real Player, Windows Media Player as well as performance in running Java applications, XML/VML rendering and much more.  The benchmark even simulates real world behavior by having chat windows open while typing in a Word document, while browsing the Web among other things.

The benchmark is much more complicated to setup than SYSMark or any other benchmark for that matter.  It requires a Server and a Client machine; the Server must be running either Windows NT Server or Windows 2000 Server.  Luckily, we were in the middle of building a fifth server for the AnandTech Forums and adapted that server, which runs Windows 2000 Server, for use with the WebMark 2001 benchmark. 

The performance of the server does not impact the performance of the client machine in this test, provided that the server is fast enough.  In this case, the 1GHz Athlon that was used in the server with its 512MB of PC133 SDRAM and 100Mbit Ethernet cards on the 100Mbit switched LAN in the lab was more than enough for the test.  If you're curious as to what is necessary to setup the server portion of the benchmark, check out this page that explains the process.

The client is then loaded with the WebMark 2001 client-software as well as copies of Adobe Acrobat, Microsoft Media Encoder, Microsoft Agent, Macromedia Flash, Apple Quicktime, Real Networks Real Player, Microsoft Media Player, and Cycore Cult 3D (3D web visualization software).  These technologies combined with Java, HTML/DHTML, ASP, Perl, SSL, VML and XML are used in a total of 7 mockup web sites and are used to measure the web performance of the client system.

A description of the 7 web sites taken from BAPCo's WebMark 2001 Whitepaper:

The scores generated from the benchmark include an overall score that is then split into three separate categories: Business to Business (B2B) transactions, Business to Consumer (B2C) transactions, and Business transactions (B) such as those you'd find occurring over a corporate Intranet. 

The WebMark 2001 Technology run is another benchmark contained within the WebMark suite and it simply offers performance figures in the following categories:

·         Loads in Average Load Time
·         XML processing in XML Operations per Second
·         Video playback measured as CPU Utilization (smaller is better)
·         Java applications in Java Operations per Second
·         VML in Average Response Time
·         3D visualization in Frames per Second
·         Flash animations Frames per Second
·         Flash applications in Average Response Time

While the benchmark does paint an interesting perspective on processor performance, there is one major flaw, and that being that systems are still generally bottlenecked by their connection to the Internet before they are CPU limited in their web performance.  At the same time, very few pages are as intensive as those exemplified in WebMark 2001.  If anything, allow WebMark 2001 to provide an insight into what may be, not what currently is.

Why are we using it?  Because it provides us with another method of looking at performance and extrapolating more information regarding the architecture of the CPUs we are comparing here today.

The Test

Windows 98SE / 2000 Test System
Hardware
CPU(s)	Intel Pentium III 1.0GHz		Intel Pentium 4 1.5GHz Intel Pentium 4 1.0GHz	AMD Athlon "Thunderbird" 1.33GHz AMD Athlon "Thunderbird" 1.0GHz
Motherboard(s)	ASUS CUSL2		ABIT TH7-RAID	ASUS A7M266/A7V133/A7V
Memory	256MB PC133 Corsair SDRAM (Micron -7E CAS2) 256MB PC800 Samsung RDRAM
Hard Drive	IBM Deskstar 30GB 75GXP 7200 RPM Ultra ATA/100
CDROM	Phillips 48X
Video Card(s)	NVIDIA GeForce2 Ultra 64MB DDR (default clock - 250/230 DDR)
Ethernet	Linksys LNE100TX 100Mbit PCI Ethernet Adapter
Software
Operating System	Windows 98 SE Windows 2000 Professional SP1
Video Drivers	NVIDIA Detonator3 v6.50 @ 1024 x 768 x 16 @ 75Hz NVIDIA Detonator3 v6.50 @ 1280 x 1024 x 32 (SPECviewperf) @ 75Hz VIA 4-in-1 4.29V was used for all VIA based boards
Benchmarking Applications
Gaming	Unreal Tournament 4.32 Reverend's Thunder.dem Quake III Arena v1.27g demo127.dm3 Mercedes-Benz Truck Racing Timedemo Serious Sam Public Test 2 (Coop Party 04)
Productivity	BAPCo SYSMark 2000 BAPCo WebMark 2001 Ziff Davis Media Content Creation Winstone 2001 Ziff Davis Media Business Winstone 2001 Benchmark Studio Beta 2.0 MusicMatch Jukebox 6.0 (MP3 Encoder Tests) SPECviewperf 6.1.2

SYSMark 2000 goes about benchmarking in a different manner than the two previous Winstone tests.  Instead of multitasking between various business or content creation applications, SYSMark simply runs one application at a time but features a wide variety of programs. 

Although a few programs that compose the suite are particularly memory bandwidth intensive, the same cannot be said about the whole.  For this reason, combined with the fact that there is no multitasking, the AMD760’s DDR SDRAM does not offer any incredible performance advantages here.  As we discovered in our KT133A review, the Athlon isn’t as memory bandwidth hungry of a processor as the Pentium 4 is, so DDR isn’t given much of a chance to succeed here at all. 

Even the regular Pentium III at 1GHz is fine for this type of usage.  If you don’t find yourself running more than a single application at once, you don’t need to have a blazingly fast machine, even a regular Pentium III will do.

Here is where things get interesting.  We just finished looking at a benchmark that 1) wasn’t memory bandwidth intensive as a whole, 2) ran a single application at a time and 3) resulted in a recommendation for a regular Pentium III for most users. 

Taking a look at Benchmark Studio’s performance figures, the picture changes considerably.  Remember that Benchmark Studio is more of an IT/Power User’s benchmark, there is quite a bit of multitasking going on, and while no particularly strenuous tasks are taking place (e.g. no 3D rendering, code compiling, etc…), the systems are quickly brought to their knees by simulating various types of load from accessing databases, watching videos and even checking email all while running normal office applications and browsing the web. 

The first thing you’ll notice is that the Pentium III is just too slow for this benchmark which is a complete reversal from what we saw in SYSMark (it just goes to show you that there’s no one perfect processor for everyone’s needs, although some come very close).  The Pentium III is crippled here by the lowest bandwidth FSB out of the bunch, and a 6-year-old architecture behind it.  The Pentium III’s days are numbered and it isn’t shocking that Intel will stop manufacturing Pentium III processors after this year is over with.  It’s time to jump on a faster bus guys.

When comparing the Athlon 1.33GHz processor on a KT133A platform to a Pentium 4 1.5GHz, the competition is fierce and very close.  It is so close in fact that you can pretty much say that the two are on par with one another. 

However when you throw the AMD760 platform into the fray, the 1.33GHz processor pulls a fair distance away from the competition, even that of its identical twin on the KT133A test bed. 

The performance advantage here is actually quite largely in favor of the AMD760 platform and its DDR SDRAM, which is something that we haven’t seen too much of.  In fact, the 1.33GHz Athlon on the AMD760 with PC2100 DDR SDRAM is able to complete the tests in approximately 80% of the time of the same CPU on the KT133A platform.  Giving it a 20% performance advantage due to the platform/memory bus. 

We’ve already explained the reasons, this is a particularly high memory bandwidth benchmark and it does happen to be how many power users conduct business on their computers, although a tad on the extreme side with this particular test featuring over 13 stress modules up and running. 

The 1.33GHz Athlon is able to approach the Pentium 4’s performance under Quake III Arena, however it is still unable to dethrone it as the Pentium 4 has dominated in Quake III ever since its introduction.  

It is still unclear as to exactly why Q3A does so well with the Pentium 4; when asked, John Carmack the creator of the Q3A engine, has even said he is unsure as to why it performs so well on the Pentium 4.

Unlike Quake III Arena, UnrealTournament provides a completely different picture yet the game is in the same genre as Q3A.  All of the Athlons come out on top in this test, with the 1.0GHz part offering performance identical to that of the 1.5GHz Pentium 4.

Having recently gone into production, Serious Sam is a brand new game and although we are still using the public beta, the performance should be representative of the final game.  The standings here are identical to that of UnrealTournament.

The picture under Mercedes-Benz Truck Racing changes dramatically, as the Pentium 4 achieves an even greater performance advantage over the Athlon than it did under Quake III Arena.

So we have provided you with a total of four game benchmarks, two of which are relatively new games, the remaining two being older engines.  In half of the games the Pentium 4 was faster but in the other half, the Athlon was faster.  This decision is getting tough isn’t it?

The MP3 conversion process isn’t particularly bandwidth intensive, but it is very FPU intensive.  Without SSE2 instructions to really make the Pentium 4’s FPU shine (even then it is questionable how useful they will be in a situation like this), the Athlon easily takes the performance lead here.

The two systems that completed the test in 20 seconds were the two 1.33GHz Athlon platforms, confirming that extra memory bandwidth and slight differences in memory latency don’t seem to effect performance. 

Because the test isn’t bottlenecked by memory bandwidth, the CPU scaling factor plays quite nicely.  The 1.33GHz Athlon is clocked 33% higher than the 1GHz Athlon and is 20% faster in this test. 

Again, the Pentium III is falling far behind the rest of the competition. 

Converting WAV files to Windows Media Audio files does take noticeably longer than converting to Variable Bit Rate MP3s, but the file size is also smaller at 128kbps.  The standings don’t change much here from what we saw with the MP3 encoding tests, the two 1.33GHz test beds still score evenly indicating that the Athlon’s L2 cache along with the available bandwidth courtesy of the KT133A’s PC133 memory subsystem is enough for this test. 

Most of today’s applications fall into this category as well.  It is all too often that when people hear the word encoding they immediately think memory bandwidth limited, and that is obviously not the case.

In the interest of time we have omitted the Pentium III and KT133A from the WebMark 2001 comparison.

Without much explanation behind the numbers you can draw a few conclusions right off of the bat.  The 1.33GHz Athlon seems to be equivalent in performance to the 1.3GHz Pentium 4.  The only other times we have seen that occur have been in some games but not in any of the Windows 2000 benchmarks we have been running. 

What does WebMark 2001 do differently than the rest of the benchmarks?  What part of the Pentium 4 architecture does it stress that gives the Pentium 4 such an advantage here over the fastest Athlon?  Let’s find out as we dig a little deeper into the benchmark.

In the B2B portion of the WebMark 2001 benchmark, 37% of the score is derived from the system’s performance when handling 3D Product Visualization tasks that are mostly FPU intensive.  We have seen proof (e.g. MP3 encoding tests) that the Pentium 4’s FPU is strong, but can be weaker than the Athlon’s FPU, which would lead us to believe that the 3D Product Visualization portion of this benchmark would favor the Athlon’s stronger FPU.

A total of 27% of the score comes from SSL & non-SSL web-page loads, which are mostly integer functions.  We have also commonly viewed the Pentium 4 as a relatively poor Integer performer because of its noticeable branch mis-predict penalty. 

XML makes up 21% of the B2B score and Flash Animation composes another 10%, leaving 5% for Image Processing (Java) and Flash Applications.  However we are unable to find clear explanation for the Pentium 4’s performance here.  Let’s keep on digging.

For proof we went to WebMark’s Technology Performance tests, that run the individual technologies that are benchmarked here and give us performance figures in each category.

If you’re wondering why the Pentium 4 did so well in the B2B tests, it’s because 37% of the score is derived from 3D Visualization Performance, and as you can see from the above graph, the Pentium 4 does come out on top in that category.  We noticed similar performance leads in many 3D games as well, indicating some sort of correlation between the two types of 3D applications.

Only 10% of the B2B score came from the Flash Animation performance which was lead by the 1.33GHz Athlon but closely followed by the Pentium 4.

Many won’t argue that XML is a technology to expect to see more and more on the web as it matures, and you also can’t argue that the Athlon and Pentium 4 are very close in their XML performance.  As we mentioned in our original critique of the WebMark 2001 benchmark, you aren’t very likely to notice this type of a performance difference between two processors.  But it is interesting to note how the processors do perform in these situations.

As if things were odd enough, in the B2C portion of WebMark 2001 the Athlon actually comes out ahead.  In this portion of the benchmark, there is a stronger bias towards the Photo Editing (Java) portion of the test, with 12% being dedicated to SSL/non-SSL page loads, 26% dedicated to Flash Animation and 20% dedicated to FPU intensive 3D visualization. 

Looking at the Java Performance figures we understand why the Athlon comes out ahead.  With more attention being paid to the Athlon’s strengths, it manages to pull ahead in the B2C benchmark.

The “B” portion of the benchmark is actually quite easy to explain since 65% of this score comes from a Microsoft Media Encoder test which easily shows off the Pentium 4’s impressive memory bandwidth figures. 

This particular benchmark scales very well with CPU speed as the 1.33GHz Athlon is around 26% faster than the 1GHz Athlon.

In order to support our claims we took a look at the Encoding Performance numbers, and just as we had thought, the Pentium 4 definitely does come out ahead causing it to lead in the “B” section of the WebMark 2001 suite.

Final Words

Looking at the argument we have provided thus far, and taking into account the benchmarks in favor of the Athlon and the Pentium 4, we can make a few conclusions.

First, the Pentium III is an aging platform. It is perfectly suitable for the majority of the users out there, however once you begin to weigh down its 133MHz FSB with multiple concurrent transactions, the architecture will show its age. Does that mean all you 1GHz Pentium III users should go out and upgrade your machines today? Of course not, but do realize that if you're going to invest money into purchasing a new system or possibly an upgrade, you will most likely want to stay away from the Pentium III as an upgrade path today. A year ago it would've been perfectly acceptable, but simply put, there are better options out there.

The Pentium 4 isn't as bad of a performer as we originally thought. The Linpack and Sandra memory performance graphs we have been showing you all for so long have surfaced in a few real world tests where the Pentium 4's architecture combined with its incredible memory bandwidth end up performing quite well. The unfortunate problem with the Pentium 4 is that it is too expensive and it doesn't offer enough of a performance boost in the majority of today's applications to be worth it. Just as we are waiting for AMD to release Palomino, Intel's Northwood Pentium 4 core may be worth the wait as well. At 0.13-micron, the Northwood core should help to lower prices on the processors and by then the processor will also be using its new Socket-478 interface, so we will revisit the idea of going down the Pentium 4 route at that time. For now, it still doesn't make much sense to upgrade to the Pentium 4. You're better off getting a cheap solution in the interim (possibly a Duron) and making the decision once Palomino and Northwood are out.

Finally, we get to the Athlon, the topic of this review of course. Now at 1.33GHz, the Athlon is running faster than ever. However the most attractive point about the Athlon outside of its performance, is its price to performance ratio. The benefit of the 1.33GHz Athlon right now is that it will help drive the prices of the rest of the Athlon line down even further. If you already hadn't noticed, the 1GHz Athlon we used in this comparison performed quite well, and is currently very affordably priced.

AMD has also been improving yields on their Athlons, we were recently testing an OEM 1.1GHz Athlon that would work perfectly fine at 1.33GHz without even increasing its core voltage. The 1.33GHz part we used for these tests, in spite of its incredible heat dissipation, was able to hit just under 1.5GHz without too much effort. If AMD can get the current Thunderbird core to carry them to 1.5GHz and beyond, they might just save Palomino for a rainy day and hope to hit with something much harder than just a cooler running core.