Skip to main content

AMD Athlon using DDR SDRAM

Review - we investigate the supposed performance benefits of DDR memory, using Crucial-branded memory samples

Dark blue icons of video game controllers on a light blue background
Image credit: Eurogamer

- AMD, memory provided by CrucialPrice - £Variable

DDR

We've spoken quite a lot about DDR memory in the last few months, but this is the first opportunity we've had to test the new technology for ourselves. With the falling prices of Athlons and memory, the only outstanding question is how stable the DDR platform is for gamers, but we'll save that for our appraisal of the various motherboard chipsets another time. For the moment, the performance freak inside of us needs a workout. First thing's first though, lets go over the history of the famous DDR acronym. The first time AMD applied those three magical letters was two summers ago with the original Athlon, which transferred data back and forth to the rest of the system twice for every clock cycle, doubling the theoretical bandwidth at its disposal. Since then the processor has retained that throughout further generations, receiving its second injection courtesy of DDR-SDRAM or as we will refer to it today, simply DDR memory, towards the end of the Thunderbird's days. Following NVIDIA's lead with the GeForce, AMD created a system architecture that truly took advantage of 200 and 233MHz front side buses. Now, the thing that has confused a number of people is the way JEDEC, the standardization body for this sort of technology amongst others, have classified DDR SDRAM standards. Unlike the previous days of PC66, PC100 and PC133, the denotation here does not focus on maximum clock speed, but the theoretical memory bandwidth, hence PC1600 and PC2100. The large number indicates that the memory can handle X number of Megabytes data per second under ideal conditions. PC1600 deals with 1600Mb/s at 200MHz, while PC2100 deals with 2100Mb/s at 266MHz.

The Flipside

Of course, the Athlon's closest rival, Intel's Pentium 4 (which is bound by agreement to use RAMBUS RDRAM), is effectively quadrupling its data rate, with a 400MHz front side bus, but in practice the Athlon holds its own more than adequately. Lets examine that phenomenon just briefly. The first thing to point out is that although the Athlon's DDR front side bus has a theoretical upper limit of 2100Mb/s, the Pentium 4 beats it in memory performance every time, because it has a theoretical high of 3200Mb/s. In fact, if you look closely at the benchmarks (they're not exciting enough to warrant a space-filling image, so take our word for it), then you can clearly see that the Athlon system falls far short of the target. Switching the Pentium 4 RDRAM for a lower speed variant, reducing its theoretical high to below that of the Athlon DDR system, the result remains in the Pentium 4's favour. Now, it would be nice to have some way of explaining this away, but in reality it simply looks as though the Athlon can't quite take advantage of its own available bandwidth just yet. Even so, as remarked just a paragraph ago, in virtually all real world benchmarks, clock for clock the Athlon tramples all over the Pentium 4. The reason for that, some argue, is that SSE2, which the P4 thrives upon, isn't implemented in any real world programs yet. Suffice to say, it's this factor which will keep AMD ahead of Intel even when the latter's clock speeds eclipse those of the former by several hundred MHz.

Regrouping

So what do we know? We know that clock for clock Intel is not a competitor, but we also know that AMD's chips don't fully take advantage of the memory bandwidth on offer. This is slightly worrying, because it means that the introduction of DDR memory into an Athlon system may not provide the mesmerizing benchmarks one would hope for, while with the increasing clock speeds of the Pentium 4 and its likely move to DDR SDRAM within the year may well do so. The only way to prove anything is to do some testing, so we built a DDR system using some of the best components we could find and gave it the run around. Equipped with an AMD 761-based ASUS A7M266 motherboard, a 1.3GHz (200MHz FSB) or 1.33GHz (266MHz FSB) Athlon and a GeForce 3, we tried our system's hand at some nasty benchmarks, alternating between 256Mb sticks of Crucial-branded PC1600 and PC2100 memory. For comparison, we chose an Athlon system using the KT133A platform (ABit KT7A-RAID) alternating between 256Mb sticks of Crucial-branded PC100 and PC133 memory. The first test is the obligatory Quake III workout using the recent demo127.dm3 timedemo benchmark.

Although beaten by its DDR-driven rival, the KT133A platform demonstrates its worth. The AMD761 equipped DDR system beats it by a mere 6%. 3DMark 2001 is a benchmark we have been itching to put to good use for a while, so lets try that next.

Although both DDR systems pull ahead slightly, the difference in performance is even smaller than the Quake III figures. As the person who sat there and watched all four runs of 3DMark, I didn't feel any was terribly lacking in terms of performance.

Real World

How about some real world benchmarks to test general performance though? One a lot of you probably want to know about is MP3 ripping performance. So as to eliminate the CD-Rom drive's capabilities, we converted from WAV to MP3 using the same 50Mb file on all four systems, then measured the time, as reported by Audiograbber. The encoding speed varies so much that it seemed largely irrelevant. We used version 1.88 of the "lame" MP3 encoding DLL.

Stalemate. The conversion process is very FPU intensive, rather than bandwidth intensive, so some might argue that it has no bearing on our conclusions. As an everyday user of Windows though, I'd argue that any purchasing decision I make has to be based on a number of factors, and if MP3 encoding speed is one of those, it matters a darn sight more than how many numbers it scores in situations I will never be in. The best real world benchmark of a new system is how it feels to use from day to day. As a long time user of the KT133 and KT133A platforms, I know that they can handle a lot of multitasking but tend to drop off a bit on bandwidth-heavy tasks. Using the DDR system feels, sadly, almost the same. In fact, moving from KT133A to AMD761 was pretty unnoticeable. The only difference came in the ever-so-slight gaming benchmark improvements and during heavy multi-tasking, when things seemed a little less hectic.

Overclocking

Although our DDR memory is rated to 100MHz, 133MHz and CAS 2.5, we thought it would be fun to push it as far as it would go using an unlocked Athlon. The highest speed our memory achieved was 160MHz at CAS3 (both sticks managed this, but the PC2100 POSTed slightly higher than the PC1600 would go). At CAS2 our memory was as stable as we would expect, falling over quite a lot at 150MHz, but seemingly fine at 133MHz. Using CAS2 didn't make a world of difference to our benchmarks - no more so than changing the CAS timing with regular SDRAM. If you want to learn more about the different settings and their meanings, have a look at our Performance Memory article from last month. It's also worth noting that we achieved slightly higher speeds with our 1.33GHz "AXIA Y" Athlon using the KT133A platform. That said, the ASUS A7M266 and the ABit KT7A-RAID are vastly different motherboards, so it's difficult to tell what is at fault. ASUS boards are renowned for their stability over their performance, after all.

Conclusion

When DDR memory first became available, a lot of people dismissed it as an expensive fad because of the relatively minor performance increase. However, if you're in the market for a whole new PC, upgrading from anything but high performance KT133A / PC133 equipment, the market is in such a state that opting for DDR SDRAM is actually just as cost effective as choosing a regular SDRAM-powered machine. All things being equal (and judging by current memory prices which put DDR at a mere £5 more than SDRAM per 256Mb stick, they pretty much are), you have no excuse as a new PC buyer for choosing less. In the future, one would argue that your DDR SDRAM will be a more important thing to have than your SDRAM… The only thing preventing me from issuing a whole-hearted recommendation of the Athlon DDR platform is stability. We haven't had the chance to play around with our AMD761-based motherboard for long yet, and although we have noticed relatively few quirks compared to some KT133A boards, there are far more people complaining about DDR motherboards and chipsets on the Internet at the moment than there are clamouring at the KT133A's door. Our advice is to wait and see how things pan out. With hobbyist favourite ABit still to produce an Athlon DDR motherboard, things are not yet set in stone.

-

Performance Memory Showdown

April's Marching Motherboards

KT133A: Performance for Less

8 / 10

Read this next