Over the summer we were asked to visit AMD in Austin to have a look at some of their future products. When it came time to talk about the future of the Athlon processor AMD began their usual spiel about how most hardware enthusiasts and early adopters understand that clock speed isn't everything, but the average Joe doesn't. They continued to tell us about how it was time for AMD to take a leading role in educating the end user; to help them understand that there's much more to the performance equation than clock speed. This continued for a few more slides of an unfinished presentation, and it honestly sounded as if AMD was going to invest time and money into educating the masses.

So much for wishful thinking; today AMD is introducing the long awaited desktop version of the Palomino core that debuted in June. Alongside the new processor AMD is reaching deep into their bag of mistakes from the past and is bringing back the dreaded performance rating nomenclature. You heard it right the first time; this new line of Athlon processors, dubbed the Athlon XP, will not be referenced according to clock speed rather they will be assigned "Model Numbers."

These "Model Numbers" are supposed to correspond to the real world performance of the Athlon XP CPUs when compared to higher clocked competitors. While AMD will argue that the Model Numbers are used to compare the Athlon XP to an equivalently clocked Thunderbird, it's clear that the ratings are used to somehow bridge the clock speed gap between the Athlon and the Pentium 4. In the eyes of the end user that isn't well informed, the rating system may serve this purpose. We have already seen examples of Athlon XP processors being referred to by their model number as a clock speed, which is exactly what AMD wants. They want the end user to see an Athlon XP 1800+ and think 1.8GHz, thus making the comparison to a Pentium 4 1.8GHz processor. Hopefully, as well educated AnandTech readers you're able to differentiate between clock speed and real world performance.

Introducing the Modeling System

We've already mentioned one of the Athlon XP models, the 1800+ but here is a full breakdown of the clock speeds and the corresponding model numbers for AMD's Athlon XP line:

AMD doesn't arbitrarily select model numbers according to how they feel performance stacks up to the competition, rather they use a collection of 14 widely used independent benchmarking applications and games (including SYSMark 2001, but more on that later) and determine the rating based on overall performance in those benchmarks.

In order to explain in simple but catchy terms why the Athlon XP is able to operate at a lower clock frequency yet obtain these high model numbers, AMD has coined the term "QuantiSpeed Architecture." In reality there is nothing new about the Athlon XP's "QuantiSpeed Architecture"; it's merely a way of saying that the Athlon XP can do more work in a single clock (higher IPC) than the Pentium 4.

The Athlon XPs also feature updated pricing:

Behind the Name

Getting past the marketing strategy of the Athlon XP, what we have is a higher clocked version of the Palomino based Athlon that was introduced in June.

From left to right: Athlon-C (Thunderbird), Athlon MP (Palomino) and Athlon XP (Palomino)

Click to Enlarge

This core remains unchanged from the core that is in the Mobile Athlon 4 and Athlon MP processors, meaning it features the same enhancements over the previous Thunderbird core:

1) Full support for Intel's SSE instructions
2) An improved hardware data prefetch mechanism
3) Enhancements to the Athlon's Translation Look-aside Buffers
4) Lower power consumption & on-die thermal diode

We have already covered all three of those improvements in great depth so rather than regurgitating them here we'll ask that you read our original overview of the improvements.

In spite of the fact that the core is the same Palomino core we saw earlier this year, just by looking at the chip you can tell that there are some other differences between the Athlon XP and the Athlon 4/MP.

As you can tell from the above picture of the Athlon XP 1.53GHz (1800+) the processor features a new packaging type. More specifically, the Athlon XP showcases AMD's first organic based packaging. Prior to the Athlon XP all AMD CPUs had used a very dated ceramic based packaging technology. The move to this organic package will allow AMD to scale the clock speed of the Athlon XP as well as its FSB if need be. It also allows AMD to mount capacitors closer to the core of the CPU on the underside of the package; this aids in improving power delivery to the core and filtering out noise. With AMD scheduled to shrink the Athlon XP down to a 0.13-micron circuit size next year, these capacitors will be even more useful as the core voltage of the processor is dropped yet again. For more information on the importance of packaging technology and its future, take a look at our latest article on it.

The next thing you'll notice is that all of the Athlon XP CPUs run off of a 133MHz FSB (albeit double data rate effectively offering the bandwidth of a 266MHz FSB). You will not see any Athlon XP processors with a 100MHz FSB since there's no real reason to use the lower frequency FSB anymore; the Athlon XP, especially at the higher clock speeds, will definitely make use of the 133MHz FSB.

For you overclockers out there, the process has not changed on the Athlon XP. We had no problem connecting the L1 bridges of our processors and adjusting the multiplier from within the BIOS of our EPoX 8KHA+ test platform.

Athlon XP Platforms: High performing and maturing

You are blessed to be building an Athlon XP system today simply because of the plethora of platform options you have at your fingertips. Motherboards are cheaper than ever, and the performance of the Athlon line has increased over 10% just because of advancements in chipsets. At the forefront of the Socket-A chipset market we have VIA whose KT266A is finally starting to appear on motherboards in the channel. For this review we tested with the first final KT266A motherboard we have been able to get our hands on: the EPoX EP-8KHA+.

In spite of AMD's requests, EPoX's 8KHA+ appropriately reports the "real clock" of the CPU at POST

The board performed just fine and was actually on par with the KT266A reference board we previewed not too long ago. We ran into a few problems with the board not POSTing after changing CPUs without resetting the BIOS but luckily no major issues that couldn't be fixed with a BIOS update.

Click to Enlarge

The 8KHA+ features an external temperature sensor located in the center of the CPU socket meaning that it probably does not take advantage of the Athlon XP's integrated thermal diode. This was to be expected though since it took motherboard manufacturers an incredibly long time to take advantage of Intel's integrated thermal diodes; it really wasn't until the Pentium 4 that it became standard to see motherboards measure the temperature of Intel CPUs according to the thermal diode of the processor.

While the current batch of Socket-A motherboards are not nearly as flawless (in regards to overall stability/quality) as the i845 boards we rounded up not too long ago, they are approaching maturity as time goes on. Especially with the incredible effort NVIDIA is putting into drivers for their nForce platform, we may see things improve dramatically for the platform in the next few months.

SYSMark 2001: The Benchmarking Controversy

Before we get to the actual performance tests there is a point we must address regarding a benchmark we have used heavily since its introduction: SYSMark 2001. It's actually quite funny about how much controversy surrounds SYSMark 2001 as many people will claim that it is unfairly biased towards Intel processors and thus should not be used to measure performance of AMD CPUs. It's funny because if you ask AMD what they think about the benchmark, they will tell you quite the opposite. In the same meeting we had with AMD over the summer the topic of SYSMark 2001 being used as a valid benchmark came up, and other than one issue they had with Windows Media Encoder, AMD had no problems with the benchmark and actually felt it was a good measurement of performance. In fact, AMD even went as far as to tell us that they would probably join BAPCo but not at this point; joining BAPCo (the makers of SYSMark 2001) requires a constant resource commitment which AMD isn't ready to make just yet.

With that said, it turns out that there is a problem with the way SYSMark 2001 reports performance. In the past we always assumed that because a large portion of the Internet Content Creation (ICC) score was derived from Windows Media Encoder performance, the Pentium 4 did very well in that suite courtesy of its incredible FSB & memory bandwidth. However as we discovered in our P4X266 Review, Windows Media Encoder isn't nearly as bandwidth dependent as we originally thought. Then why did the Pentium 4 do so well in ICC SYSMark 2001 while the Athlon clearly lagged behind?

It couldn't have been SSE2 optimizations since there are no SSE2 optimizations in the applications that SYSMark 2001 uses; and it couldn't have been SSE optimizations because otherwise the Athlon MP (which has SSE support) should have done much better in the benchmark.

The conclusion that the Pentium 4 is simply a faster content creation processor than the Athlon would be valid if it wasn't for one problem: what if SSE wasn't properly being used on the Palomino based CPUs?

Normally how you're supposed to detect whether or not a processor supports SSE is by looking at the SSE bit in the standard features register. If the SSE bit is set to true, then the processor you're dealing with supports SSE; if it's set to false, then your processor doesn't support SSE. The Athlon has this bit set to false since it doesn't have a full SSE implementation, however the Mobile Athlon 4, Athlon MP and the Athlon XP all have this bit set to true. The problem with SYSMark isn't really a problem with SYSMark, it's a problem with Windows Media Encoder 7. Windows Media Encoder 7 doesn't look at this bit to determine whether to use SSE or not, instead it reads the manufacturer code in the CPUID string of the processor. If the manufacturer code reads 'GenuineIntel', as all Intel processors report, and the SSE bit is enabled then WME7 uses SSE. This wasn't a problem until recently since AMD processors didn't have SSE support, but now with the Palomino core they do have SSE support but according to the WME7 SSE detection algorithm they will never run WME7 with SSE enabled. The reason behind this being that the manufacturer string outputted by all AMD processor is 'AuthenticAMD' which doesn't pass WME7's test for SSE compliance. Microsoft is aware of the problem and will fix it in the next version of Windows Media Player, but that doesn't change the issue with SYSMark since it relies heavily on WME7 for determining performance.

So how do you make WME7 detect the Athlon XP as a SSE enabled processor? The solution is actually quite simple; If you run a profile on Windows Media Encoder during an encoding task you'll quickly come to realize that most of the work is being done in a library located in the \Windows\System32\ directory called 'wmvdmoe.dll'. This is the encoding library used by WME hence a great portion of the encoding process is spent dealing with this file. By modifying this file you can set what manufacturer string to look for in order to enable SSE. In order to accomplish the goal of enabling SSE on the Athlon XP you'd want to look for 'AuthenticAMD' instead of 'GenuineIntel'. Those that have experience with editing dlls will find that this workaround is quite easy to implement; for everyone else, AMD has released a patch that will do this for you and even make a backup copy of the old wmvdmoe.dll file for you to compare to.

As a quick comparison, here is the performance of an Athlon XP 1.53GHz (1800+) processor with SSE enabled and disabled under SYSMark 2001:

You can clearly see that enabling SSE paints a much brighter picture for the Athlon XP; arguably a much truer picture of its performance. This brings up the question, how should we test the Athlon XP? With SSE enabled or disabled?

Obviously BAPco and Intel have both expressed concern over testing with this "patch" since it could set a dangerous precedent for CPU manufacturers releasing "patches" to "fix" performance issues in benchmarks. In this case, our opinion is that since the modification doesn't change the workload of the benchmark at all, rather it fixes a bug with the detection of a processor's features it is ok to use the patch. This is akin to using drivers that properly enable the features of a video card or a chipset, however in order for us to continue to use this workaround Microsoft must implement an officially sanctioned fix into Windows Media Encoder. From what we have been told, they have recognized the issue and they will fix the problem. It is our duty to present you all with the most thorough and complete performance data for the hardware we evaluate, and in doing so we must test the product with all features enabled.

Because we actually were aware of the problem and fully understood the issue before AMD released the patch, we feel comfortable in using the workaround. But, as a warning to AMD and any other manufacturer that finds themselves in a similar situation: work with the software developers to implement officially sanctioned patches; don't attempt to take benchmarking matters into your own hands. Had we not fully been aware of the problem, understood the issue and applied the workaround ourselves we would not have used such a patch from AMD or any other manufacturer.

If you don't feel as if the performance with SSE enabled under SYSMark 2001 is representative of the Athlon XP's performance, then feel free to mentally knock 8% off all Athlon XP scores but remember that the Athlon XP does have fully functional SSE support that works fine in all other applications outside of WME7.

With that out of the way, it's performance time.

The Test

This review marks our first endeavor into benchmarking under Windows XP. We disabled System Restore and followed Microsoft's directions for benchmarking under Windows XP to obtain repeatable results. We enabled all visual options including anti-aliased fonts.

Windows XP Test System
Hardware
CPU(s)	AMD Athlon XP 1.53GHz (1800+) AMD Athlon XP 1.47GHz (1700+) AMD Athlon XP 1.40GHz (1600+) AMD Athlon XP 1.33GHz (1500+) AMD Athlon-C 1.40GHz	Intel Pentium 4 2.0GHz Intel Pentium 4 1.9GHz Intel Pentium 4 1.8GHz Intel Pentium 4 1.7GHz Intel Pentium 4 1.6GHz Intel Pentium 4 1.5GHz
Motherboard(s)	EPoX EP-8KHA+	ABIT TH7-II RAID (Intel 850)
Memory	256MB PC800 Mushkin RDRAM 256MB DDR266 Crucial (CAS2) DDR SDRAM
Hard Drive	IBM Deskstar 30GB 75GXP 7200 RPM Ultra ATA/100
CDROM	Phillips 48X
Video Card(s)	NVIDIA GeForce3 64MB DDR
Ethernet	Linksys LNE100TX 100Mbit PCI Ethernet Adapter
Software
Operating System	Windows XP
Video Drivers	NVIDIA Detonator 4 v21.80

Internet Content Creation & Office Productivity Performance

Let's start off by doing a Model Number to clock speed comparison, shall we? AMD's Athlon XP running at 1.33GHz boasts a 1500+ model number, indicating that it should perform like a Pentium 4 1.5GHz, if not a bit faster. Considering that the 1.33GHz part performs somewhere in between a 1.7GHz and 1.8GHz Pentium 4, we'll just assume that the Athlon XP has a habit of being modest.

This modesty continues as the model 1600+ part (clocked at 1.40GHz) comes in as being 9% faster than the Pentium 4 1.6GHz; and the trend continues. You can also see the positive effects of SSE and the other Palomino core enhancements as the Athlon XP clocked at 1.40GHz is just over 21% faster than the regular Athlon-C running at the same frequency; most of that performance increase is due to SSE since it is used so heavily in Windows Media Encoder, a large portion of this benchmark.

With SSE enabled, the Athlon XP 1.53GHz (1800+) CPU is just as fast as Intel's Pentium 4 2.0GHz part. In another month however we expect to see a 2.2GHz CPU from Intel that will shift standings yet again.

Even before the introduction of the Palomino core the Athlon did extremely well in the Office Productivity portion of SYSMark 2001. You'll notice that the improvements of the Palomino core have no positive (or negative) effect in this benchmark as the Athlon XP is just as fast as the Athlon on a clock for clock basis. This isn't necessarily a bad thing since the Athlon already lead in this benchmark at 1.4GHz. With the boost up to 1.53GHz, the Athlon XP is able to continue to offer class leading performance. The 7% performance advantage over the 2GHz Pentium 4 isn't too noticeable but it's approaching that 10% figure where performance numbers actually gain meaning.

Again we see that the model numbers are meaningless as they don't convey the true performance potential of the Athlon XP CPUs.

Remember that strong performance under this test is actually very important to many AnandTech readers; don't let the title fool you, there's much more to Office Productivity SYSMark 2001 than just running Word and Excel. The benchmark is very multitasking intensive combining tasks such as voice recognition and real time virus scanning with common email and document manipulation tasks. If you're anything like the majority of the AnandTech staff, you always have an Outlook (or other email program) window open, a few web browsers, a copy of MS Word and maybe Excel with your virus scanner running in the background. This type of usage model is exemplified by these tests.

Here you can see exactly how important the Palomino core enhancements are to the Athlon XP. The Athlon XP running at 1.4GHz is able to encode our MPEG-1 source file 10% faster than the previous generation Athlon, also at 1.4GHz. Even the 1.33GHz Athlon XP is able to outperform its 1.4GHz brother. This performance increase is mostly due to the Athlon XP's data prefetch which comes in handy with data intensive applications such as MPEG-4 encoding.

The Athlon XP 1.53GHz (1800+) gives the recently announced 2GHz Pentium 4 some hefty competition; it, too holding almost a 10% lead over Intel's flagship.

The standings don't change much when looking at MPEG-2 video encoding performance. The only real difference being that since MPEG-2 encoding isn't as CPU bound with today's processors, the performance differences between the CPUs is vastly reduced. For example, the close to 10% difference between the Athlon XP 1.53GHz (1800+) and Pentium 4 2.0GHz is cut in half in this benchmark.

For our MP3 encoding test we used version 3.89 of the Win32 LAME encoder binaries. We took a 170MB wav file and encoded it using the following commandline options: -v -V 0. This created a variable bit rate MP3, varying the bit rates between 160 kbps and 320 kbps. The end result was a 27MB MP3 file that took between 2 and 4 minutes to encode.

The Athlon XP isn't able to dominate as clearly in this benchmark as the Pentium 4 remains quite competitive with all of the new processors. The Palomino core enhancements help the Athlon XP out as the Athlon 1.4GHz processor is 6% slower than the 1.4GHz Athlon XP.

Even its developers have indicated that Max Payne is unusually dependent on CPU power which makes it the perfect gaming benchmark. Every since AMD processor in this comparison is at the top of the charts, closely followed by the 2GHz Pentium 4. At 1.53GHz the Athlon XP is a good 11% faster than the fastest Pentium 4.

Finally we have Serious Sam and the performance here is no different than in any of the other 3D gaming tests. Here the Athlon XP has the largest lead over the Pentium 4 that it has had throughout this entire review.

3D Rendering & Animation Performance

While we normally publish the entire results of the SPECviewperf 6.1.2 run, for this review we picked the three most CPU dependent viewset tests and presented you with those. The tests absent from this review are AWadvs, MedMCAD and ProCDRS which were largely video card bound to the point where there was no more than a 1% span of performance figures between all of the CPUs in this review.

These standings should be very familiar to you by now; the Pentium 4 has a difficult time keeping up with the Athlon XP. The Palominco enhancements only offer approximately a 3% performance boost for the Athlon XP here.

The Pentium 4 is much more competitive in this viewset but more importantly is the huge performance increase that is courtesy of the Palomino core enhancements. You can see that the Athlon 1.4GHz is clearly the worst performer in this benchmark, but the Palomino core enhancements give the Athlon XP 1.4GHz processor a 34% advantage over its older brother.

For the first time we see dominance exhibited by the Pentium 4 line in the Light-04 viewset. It is important to note that there is little difference (percentage wise) between the top five or six contenders in this test.

Final Words

While there was some debate as to which was faster, the Pentium 4 2.0GHz or Athlon-C 1.4GHz, the Athlon XP clearly establishes a performance lead over the competing Pentium 4 solutions. At 1.53GHz, the Athlon XP is able to consistently outperform the Pentium 4, partially due to the incredible performance of the retail KT266A motherboard that we paired it with. Especially in games such as Wolfenstein and Serious Sam, the idea of the Pentium 4 being the better gaming processor is clearly put to rest.

In most cases, the Pentium 4 is outperformed on the order of 10% which is exactly the performance enhancement the 512KB on-die L2 cache of Northwood is supposed to offer. If the first 2.2GHz Northwood based Pentium 4s are indeed delivered next month as we have been expecting, then AMD's reign as king with the Athlon XP may be short lived; at least until they can answer back with a higher clocked processor, but remember it's much more difficult for AMD to ramp in clock speed than it is for Intel because of architectural differences.

This flip flopping of performance leaders was quite uncommon between AMD and Intel in the past. If you'll remember back to a couple of years ago, the performance leader was consistently Intel, especially in gaming and professional level benchmarks while the value leader was AMD. One of the most honest quotes we ever received from an Intel employee (although not an official quote for obvious reasons) was in response to the question "Will Intel ever make a return to the level of dominance that existed a few years ago?" His response was on the order of, 'not as long as AMD continues to stay on the ball'. With the Athlon processor AMD finally had a powerful part on their hands and unless they seriously screw things up, you can expect to see this sort of heated competition between AMD and Intel for some time to come. You've got to admit it's definitely been good for the end users as we've never been blessed with such high performance CPUs at such very low prices.

Now we get to the issue of AMD's new performance rating system. If you haven't gathered already, the performance ratings do little justice to the Athlon XP platform. We've already proven that they don't correspond to real world performance (the CPUs considerably exceed the real world performance implied by the ratings) and those that are targeted by the ratings could potentially become more confused because of the ratings. It's very clear that AMD's Athlon XP can stand on its own merits without requiring a silly numbering system to "educate" the uninformed. In reality it's misleading the uninformed. How hard would it have been to present the results of those 14 benchmarking applications rather than devise a modeling system that confuses more than it helps? And what happens if AMD is able to attain a clock speed parity with Intel, using the upcoming ClawHammer CPU for example? Will we see even higher ratings on CPUs, defeating the purpose of using this rating system to level the clock speed playing field between AMD and Intel? The unfortunate answer is probably yes.

We understand what AMD is trying to accomplish with the rating system, but from our perspective, the Athlon XP is a strong enough product on its own to survive and excel without the use of such a system. There are much better ways to educate the masses.