The industry happens to be at one of those points in its existence where the old saying "the decisions you make today will affect you for the rest of your life" applies quite nicely.  Manufacturers are getting off their most recent two-year high in which the market was booming and sales couldn't have been better, and are now facing the reality of not being able to rest on their laurels indefinitely.  It is during times like this that products like the Athlon and the Pentium 4 are designed, and the users are left with little more than the same-old, same-old to keep them occupied until the next major push for an upgrade.

If you think about it, it has honestly been a while since we've recommended a major upgrade for you all.  When the AMD760 chipset was released we said wait until DDR prices come down; and when the Pentium 4 was released we told you to wait for the 0.13-micron version. 

Unfortunately there is very little on the radar for the majority of 2001 that will be of the magnitude of the introduction of the very first Voodoo cards, or the first Pentium IIs or even the first Athlons.  The server/he-workstation market will see a few interesting products with the new Intel Xeon and AMD760MP, however for the majority of desktop users there aren't any major upgrades coming over the next few months. 

That isn't to say that there aren't users out there that will want to upgrade, in fact, there are quite a few users that will want to upgrade.  But without a major push for a full system upgrade, for many, this will be the time for an incremental upgrade.  Those users that may still be on overclocked Celerons, older Pentium IIIs or maybe even the first 500MHz Athlons may find their systems responding a little slower than they're used to.  At the same time, with a new Palomino core in the works and a larger cache version of the Pentium 4 on a 0.13-micron process due out later this year, you don't really want to spend all too much on an upgrade right now only to find out that you're longing for another fix in just 9 months. 

The idea of an incremental upgrade is an interesting one.  For starters, you want something faster than your current setup, but at the same time you don't want to spend a lot and would ideally like to reuse as much as possible.  At the same time, one man's dream is another's reality, so what we're referring to as an "incremental" upgrade is also a great way to put together an entry-level system. 

The heart of this type of an upgrade or entry-level system has often been a low-cost processor from either Intel or AMD's line, and lately our recommendation has fallen in the realm of AMD's Duron.  Today AMD provides our choice for an entry-level processor with a boost; the Duron gets its incremental upgrade from 850MHz to 900MHz.

An updated Roadmap

As you all have probably already heard as well as noticed, the new Palomino and Morgan cores for the Athlon and Duron respectively are not available.  In fact, they won't be available anytime this quarter.

At CeBIT and WinHEC AMD has gone around informing the public of their new, revised roadmap.  We provide a look at that below:

Click to see the Full Roadmap through 2002

The first thing to notice is that both Palomino and Morgan, upgraded versions of the Thunderbird core, are going to be out for desktop platforms in the second half of this year. 

Interestingly enough, the Palomino core itself will actually make its debut in the mobile and 1-2 processor server/workstation markets in this quarter.  This actually makes perfect sense, and it's an approach Intel has been using for a little while now.  The MP Server/Workstation market and the mobile market are both much lower volume markets than the performance, mainstream and value desktop market segments.  Because of this fact of life, it makes the most sense to introduce a new processor core into these two relatively low-volume segments as it will help to ramp up production and tweak yields, thus preparing the part for a high-volume desktop launch. 

The fact that AMD is employing this technique for the launch of the Palomino indicates that the Palomino core is significantly different from the Thunderbird core to the point that such a staggered launch is necessary. 

At the same time, the Morgan and Palomino launches appear to be concurrent, or close enough that it can be argued that there won't be any significant differences between those two which is what we already knew from the first time we were introduced to Morgan. 

Whatever ends up making its way into the Palomino/Morgan cores should be interesting, since they are still going to be 0.18-micron parts.  There is enough out there to believe that AMD would want the Palomino/Morgan to be more than just a cooler running Thunderbird/Spitfire core.  It could very well be that in the process of optimizing the current cores for more efficient operation AMD was able to unlock a significant amount of performance that was worthy of delaying the core's introduction until Q3. 

Let's hope that is so because by the end of this year, the Pentium 4 will be cheaper, smaller, faster, and have a larger cache; in other words, it will be much more competitive.

The baby Duron hits big brother speeds

The Duron has been slowly but surely increasing in clock speed over time, and it has finally hit 900MHz.  No matter how you try and present it, the Duron is much more powerful than any other entry-level CPU that we have come across. 

The Celeron's day is long gone, even with its 100MHz FSB the peak performance of the Celeron is nowhere near that of the Duron.  Pay attention to our use of the word peak in the previous sentence because as we noticed in previous reviews, when paired up with either the SiS 730S or VIA KM133 chipset, the Duron loses much of its performance advantage over the Celeron.  So if you're buying a pre-built system with integrated video, the [100MHz FSB] Celeron 800 might actually be a better option since its integrated video platform (i810E/815E) is much more mature, however when building your own system we already know what the better option is.

At 900MHz the Duron is relatively unchanged from any previous incarnations of the core.  As usual, below you will find our comparison table of specifications.

The Test

Winstone 2001's focus on a light multitasking load takes into account disk performance, which is always a limiting factor, as well as overall platform performance through the interaction of the FSB and memory buses.  The test is mainly integer oriented which is why the Pentium 4, not shown above, does not do particularly well in the benchmark courtesy of heavy branch mis-prediction penalties. 

Starting off with Content Creation Winstone 2001 gives us a 2.5% increase in performance over the Duron 850 courtesy of the 5.8% increase in clock speed.  The ratio of performance to clock speed increase results in a 43% mark of efficiency, meaning that although the performance is scaling relatively well there are still a few bottlenecks present. 

The most obvious bottleneck in this case remains the size of memory and the speed of the hard drive.  Because we used 256MB of PC133 SDRAM and an IBM 75GXP in the test, it is unlikely that you'd be able to get a larger performance increase out of upgrading any other one component in our test system.

It is important to note the poor performance of the 66MHz FSB Celerons courtesy of their castrated 66MHz FSB.  The processor is only as fast as its weakest link and in this case, only being able to be fed by a 66MHz bus and its 4-way set associative L2 cache is truly holding it back.  The 100MHz FSB Celeron 800 does perform noticeably better, and actually remains relatively competitive. 

The entry-level Duron 900 actually offers 88% of the performance of the high-performance Athlon running at 1GHz in this particular test, which happens to resemble the way in which many users use their PCs. 

The Business Winstone 2001 test is much like the Content Creation test in that the benchmark does do a good job of simulating normal PC usage, and likewise it falls victim to the same bottlenecks. 

Because of this, the performance picture does not change much although the Duron 900 now offers 90% of the performance of the Athlon 1GHz and the Pentium III 1GHz actually pulls ahead of the Athlon by approximately 6%. 

Remember, the Pentium III has a much wider internal bus to its L2 cache so the data can be accessed quicker.  In business applications that are generally highly L2 cache dependent for peak performance, the Pentium III's 256-bit internal bus to its L2 cache does come in handy (the Athlon only has a 64-bit path to its L2 cache).

Not everyone is a multitasking fiend, in which case SYSMark 2000 provides a better overall picture of system performance as the benchmark only runs one application at a time.  The benchmark itself consists of twelve applications that range in function from word processing to image editing and 3D rendering to video production.

The suite is evenly distributed between those applications that focus on integer performance and those that require a strong FPU for good performance.  There is a larger focus on having a high performance L2 cache, and the Pentium III commands a strong lead here partially because of its 4x wide L2 cache bus. 

The Duron 900 offers a 2.4% increase in performance over the 850MHz part, which is similar to the increase we saw in Content Creation Winstone 2001. 

Benchmark Studio has proved itself to be a truly powerful test that should be reserved for the most extreme of users.  CSA Research, the creators of the benchmark, refers to the type of load simulation the test creates as "constant computing," and without a doubt this test put our test bed under a constant load. 

There are no disk limitations here as less than 2% of the benchmark is spent waiting on the disk to respond.  The suite consists of a variable about of stress modules, for this particular test we simulated a total of 13 sources of load using these stress modules, and in combination with those modules running in the background the test executes and times performance doing various tasks in MS Word, Excel, Power Point and Internet Explorer.

The benchmark is very bandwidth intensive, which is why the 66MHz FSB Celerons do so poorly.  Even at 800MHz, the Celeron's 4-way set associative L2 cache holds it back although it is admittedly faster than the Duron 600. 

Even at 900MHz the Duron isn't close to its 1GHz Athlon brother that is able to complete the rigorous tasks in 70% of the time that the Duron takes.

If SYSMark is an example of "light use" and Winstone is an example of a power user's usage patterns, then Benchmark Studio is representative of what a truly high-end PC should be able of handling.  For those of you that already aren't using your PCs like this, with multiple video streams coming in, multiple database connections and connections to exchange servers open, this is more of an indication of how your systems will perform in the future as applications become more bandwidth intensive.

As we move into gaming performance, the CPU's importance begins to take a secondary role provided that you have a fast enough graphics accelerator.  Generally speaking, almost all of the CPUs in this chart (maybe with the exception of the two 66MHz FSB Celerons) are enough to drive a GeForce2 GTS or GeForce2 Ultra to the point that you're not going to be severely CPU limited. 

If you're upgrading or purchasing a new system for improved gaming performance you're going to have to ask yourself what type of game you're looking to get better performance in and where you are currently limited. 

Luckily games shouldn't be too disk limited, meaning that as long as you have enough memory (128MB - 256MB being the ideal target range) you shouldn't be swapping to disk except for during level loading stages.

Games like Quake III Arena mainly depend on a graphics card that is capable of delivering high fill rate power, which the test GeForce2 Ultra does quite well.  However there are other games such as Flight Simulators and 2D real-time strategy titles that are more dependent on raw CPU power than graphics card fill rate.

UnrealTournament, another first person shooter like Quake III Arena, also focuses mostly on fill rate potential and then CPU power.  Memory and FSB bandwidth do play relatively important roles in performance here as well. 

The engine scales relatively well with CPU speed at this low of a resolution, and with the GeForce2 Ultra actually continues to scale well at higher resolutions too.  The Duron 900 is exactly 2.5% faster than the Duron 800, which is on-par with the improvements we have seen in other benchmarks as well.

Serious Sam is a recently released game based on a completely independent engine developed by croteam.  The game has a relatively high polygon count and deals with a number of outdoor scenes and the complexities that go along with them. 

The standings continue to remain the same, as the Pentium III is able to edge out the Athlon for first place and here the Duron 900 is offering 84% of the performance of the 1GHz Athlon. 

CPU scaling is actually slightly improved under Serious Sam as the Duron 900 is 3.3% faster than the Duron 850, compared to the 2.5% increase we have seen in most of the other benchmarks.

Remember, as games continue to increase in complexity they will become much more demanding on not only your graphics hardware but your CPUs and memory subsystems as well.

Mercedes-Benz Truck Racing is another relatively new game however it doesn't seem to be scaling well with CPU speed as well as previous tests since the Duron 900 isn't even 2% faster than the 850.  At the same time, the Duron 900 offers 90% of the performance of the 1GHz Athlon, indicating that the limitations present once you get high enough in clock speed are either related to memory bandwidth or the graphics card. 

Final Words

The Duron continues to prove itself as a cost effective upgrade for those that are currently looking to revive system performance while at the same time offering the perfect entry-level solution that the market has demanded for some time now. 

The Duron still suffers the same issues that its older brother unfortunately is faced with.  Without AMD's commitment to produce top tier chipsets for the CPU, the processor is completely reliant on third party solutions from ALi, SiS and VIA in order to succeed.  However as we have seen from our performance investigations of the two UMA chipsets that are currently available and targeted at the Duron's market, many of the performance advantages the Duron offers over the competition are negated by the use of these chipsets. 

The reasoning behind it is simple; with only 60% of the total cache of the Athlon, the Duron is obviously more dependent on memory performance.  The data that the CPU cannot find within its L1 or L2 caches must be retrieved from main memory, the longer that retrieval process takes, the longer the CPU must wait and the more performance is lost.  Unfortunately the memory performance of both the SiS and VIA low-cost solutions is noticeably lower than what Intel's similarly targeted chipsets are able to provide. 

For enthusiast buyers that won't be spending any time with any of the UMA solutions with integrated video, the Duron on a KT133/KT133A continues to be a great buy and is now running at close to the frequency of the fastest Pentium III. 

AMD should get used to the Duron vs Pentium III comparison since this will be the comparison made as the Celeron is phased out over time and either replaced with the Pentium III or a more Pentium III-like core.