DDR400 SDRAM with Athlon XP Platform: Performance and Future Potential

by Ilya Gavrichenkov
03/31/2003 | 08:52 PM

We have already told you a number of times on our site that very soon Intel is going to introduce a new 800MHz Quad Pumped Bus in systems based on Pentium 4 processor family. This is scheduled to happen in the middle of April and will be accompanied by the launch of new chipsets (Springdale and Canterwood) that support such system bus frequency and 200MHz FSB frequency. The boost of FSB bus bandwidth will require faster memory to use up the bigger processor bus bandwidth. While the new DDR II standard is far from entering the mass production phase yet, Intel suggests that we use dual-channel DDR400 SDRAM in new Pentium 4 based platforms. This fact means that DDR400 SDRAM is going to give up its reputation of the “memory for overclockers” and become a mainstream product. This is the way Intel views the near future of the memory market:

But what about the Intel’s good old rival, AMD? Right now, Athlon XP family processors use 333MHz system bus and 166MHz FSB frequencies. Considering that most chipsets for Athlon XP quite naturally show their best performance in case the system and memory buses work synchronously, DDR333 SDRAM seems to be a preferable memory type for modern Socket A platforms. Thus, after the arrival of Pentium 4 with the 800MHz bus, AMD can find itself lagging behind as concerns the employed memory type. Of course, it’s theoretically possible to use DDR400 SDRAM with Athlon XP using 333MHz bus. But numerous tests prove that this brings no perceptible advantage from the performance point of view, or even slows the system down.

That’s why the transition of Athlon XP processors to a 400MHz bus is vital today. A faster bus would allow using the advantages of the upcoming DDR400 industry standard. In the end, this should increase the performance of Socket A processors without raising their clock frequencies. Such an opportunity must be most appealing to AMD: Thoroughbred and Barton CPU cores, which are available by AMD today, have practically reached their frequency maximum. So, it would be a nice thing for AMD to raise the performance of their processors in some other way than by mere overclocking.

Meanwhile, AMD keeps silent about its plans on making a faster bus, and only says that company engineers are considering this possibility. The Alpha EV6 bus, currently used in Athlon XP CPUs, can in theory work at 400MHz, but there can arise some problems with chipsets and mainboards. Manufacturers just didn’t intend their products to be used at 200MHz FSB. This may lead to lower stability of Socket A systems and AMD just can’t allow this to happen. As a result, the release of Athlon XP with a 400MHz bus may require the development of new chipsets and mainboards (or new revisions). But the expediency of putting new effort and resources into making up new products for Athlon XP is questionable as the Socket A CPU line from AMD is going to stop its evolution soon.

Nevertheless, in spite of all those problems, we are inclined to think that the next Athlon XP 3200+ processor may be released in two variants: for the 333MHz and 400MHz bus. But even if AMD doesn’t bring itself to such a move, the 400MHz bus is sure to become widely used among overclockers. As soon as they get DDR400, they wouldn’t miss the chance of getting some speed-up through overclocking the FSB.

We are going to answer a few important questions in this article. First, we will estimate the performance growth provided by the use of the 400MHz bus, considering DDR400 has poorer timings than DDR333 and DDR266 modules. Second, we will check the capabilities of available DDR400 memory modules. Third, we will see how memory timings affect the overall system performance.

Testbed and Methods

After we checked all DDR400 SDRAM memory modules currently available in the market, we found they boasted strikingly different parameters. The timings of memory modules we considered varied from 2-3-3-5 (CAS Latency – RAS# to CAS# Delay – RAS# Precharge – Active to Precharge Delay) by best specimens from OCZ Technology and Corsair to 3-5-5-10 in relatively low-cost no-name modules that had flooded the stores. That’s why we benchmarked the system with the 400MHz bus and DDR400 SDRAM, setting different timings. This will also help us determine their effect on system performance and prove the necessity of using high-quality DDR400 modules. As for DDR333 and DDR266 memory modules, most manufacturers now offer products capable of working with the lowest 2-2-2-5 timings. So, we used these timings when we tested the systems with the 333MHz and 266MHz memory bus. Thus, we will be able to see whether slower DDR400 memory can outperform the faster DDR333 in practice and also whether it’s reasonable to use DDR400 plus the 400MHz bus instead of DDR333 plus the 333MHz bus.

The system we used for our test session was based on NVIDIA nForce2 chipset. This chipset is most popular today among enthusiastic users thanks to its high performance, dual-channel architecture and advanced networking and sound capabilities. As this chipset works best in the synchronous mode, all the tests were run at identical system and memory bus frequencies.

But when we were thinking of a mainboard for our experiments, we faced a certain problem. It turned out that many nForce2-based mainboards don’t allow raising the FSB frequency synchronously with the memory frequency over 180-190MHz. This is a disappointing fact, which may become a serious obstacle for Athlon XP processors supporting 400MHz bus. Anyway, we did find a mainboard, free from this problem. This was ABIT NF7. Unlike mainboards from other makers, ASUS, Chaintech and EPoX among them, this mainboard allowed increasing the FSB frequency synchronously with the memory frequency up to 225MHz, at least.

ABIT NF7 owes this amazing feature to a well-thought design as well as to the ability to send increased voltage to the North Bridge of the chipset. So, the stability of the chipset grows up considerably. However, other mainboard makers are working on this problem and promise to roll out new revisions of their products that would work in the synchronous mode at 200MHz FSB or higher.

To check the dependence of Athlon XP CPU performance on the system bus frequency, we took a CPU working at exactly 2GHz. This 2GHz clock-rate was set as 10x200MHz, 12x166MHz or 15x133MHz.

Now, let’s welcome our testbed:

• AMD Athlon XP (Thoroughbred) CPU 2GHz;
• ABIT NF7 mainboard (NVIDIA nForce2 chipset, BIOS version 15);
• 512MB DDR SDRAM (2x256MB);
• ATI RADEON 9700 PRO graphics card (Catalyst 3.1 driver);
• Seagate Barracuda ATA IV 80GB HDD.

As said above, we used different memory modules; their timings are specified in test results. All the memory was used in dual-channel mode. The benchmarks ran in Microsoft Windows XP Professional SP1 OS. The BIOS of the mainboard was set to maximum performance.

Performance

First of all, we ran synthetic tests measuring memory subsystem performance. The following table contains the results of Cachemem:

The shift from the 266MHz to 333MHz bus was evidently advantageous: it proved efficient providing higher memory bandwidth and lower latency. However, we can hardly hope to get similar improvements resulting from the shift to 400MHz bus. The problem lies in the memory subsystem latency. Unfortunately, DDR400 SDRAM doesn’t allow setting such aggressive timings as DDR333 SDRAM. That’s why the latency of a system with 333MHz bus and DDR333 memory is often better than that of a system with 400MHz bus and DDR400 memory. Therefore, the transition of Athlon XP to the 400MHz bus may bring no positive effect at all in some cases. But those tasks that require high memory bandwidth are certainly going to benefit from the faster memory and system buses: the bandwidth of CPU-to-memory highway doesn’t depend on timings much.

Now, the results of SiSoft Sandra 2003:

The numbers are one more proof to what has been said above. The CPU-to-memory bus bandwidth is barely affected by the timings and thus goes up along with increased frequencies of system and memory buses. Meanwhile, the transition from the 266MHz to 333MHz bus brings a 22% boost, while the transition from 333MHz to 400MHz bus – only 17%. This once again tells that the influence of the bus speed onto the system performance will become lower along with the growth of its frequency.

We can say the same things again about the results of PCMark2002.

But let’s go over from synthetic tests to real applications.

SYSmark2002 measures system performance in basic office and content-creation applications. The results indicate that in some cases DDR400 SDRAM together with 400MHz bus doesn’t provide any advantage over the 333MHz bus. Office applications were most fastidious to memory timings, while streaming-data applications put more emphasis on the communication speed between the CPU and the memory.

The DDR400 system with “bad” timings is once again no better than the DDR333 system at archiving data with WinRAR utility. Even most aggressive timings of DDR400 SDRAM give us only a 3% gain over the 333MHz bus and DDR333 SDRAM system. Meanwhile, the transition from the 266MHz to 333MHz bus made the system about 13% faster.

Video-stream processing proved to be more dependent on the data transfer rate between the CPU and the memory. DDR400 with any timings is better than DDR333 here.

The final rendering in 3ds max 5 is performed faster on the 400MHz bus and the speed doesn’t practically depend on memory subsystem settings.

Gaming applications come next.

Both versions of the popular 3DMark benchmark yield similar results. The 400MHz bus gives a certain performance growth, but this growth is definitely lower than the improvement we saw on transition from the 266MHz to 333MHz bus, which AMD carried out some time ago.

The CPU test from 3DMark03 shows it even more clearly. By the way, we would like to point out the fact that CAS Latency is no longer the key memory timing. In many cases, you can have a faster system by increasing CAS Latency, but reducing RAS# to CAS# Delay, RAS# Precharge or Active to Precharge Delay.

The performance growth due to the transition from 333MHz bus to 400MHz bus and DDR400 memory makes 6% in Return to Castle Wolfenstein.

The performance growth is lower in Unreal Tournament 2003. Meanwhile, we haven’t seen it in any game that the faster system bus and less aggressive memory timings could negatively tell on the performance. So, if AMD does go over to the 400MHz bus, the gaming community should be pleased. Other users will have to spend some time and money to find and purchase high-quality DDR400 modules.

Best DDR400 Modules

Thus, our test session showed that the best memory modules could function at 400MHz frequency with 2-3-3-5 timings. In this case, the system working with 400MHz bus proved about 3%-7% faster than the system working with DDR333 memory and 333MHz bus. Ordinary DDR400 doesn’t allow setting CAS Latency = 2 at 400MHz and thus brings no such performance gain. So, below we discuss those modules that provided the maximum performance in our benchmarks.

First of all, we would like to single out OCZ EL DDR PC-3700 Enhanced Latency modules.

Theoretically, these modules can be clocked at 466MHz. So, it is no wonder they do excellently at 400MHz. But, according to the official specification, it’s recommended that you set CAS Latency = 2.5 and raise memory voltage to 2.7V for these modules to work at 466MHz. As for 400MHz, they worked without any problems with 2-3-3-5 timings.

We also tried to find the maximum frequency these modules could work at in our testbed with 2-3-3-7 timings and 2.6V voltage. We got 456MHz and it’s the absolutely best result among all memory modules we took for this review.

CMX256A-3500C2 modules from Corsair are second best.

The manufacturer claims these modules can work in mainboards based on VIA KT400 chipset at 434MHz frequency and with 2-3-3-7 timings. In our nForce2 based system they worked stable at 400MHz with 2-3-3-5 timings, just like OCZ EL DDR PC-3700. During overclocking with 2.6V voltage and 2-3-3-7 timings, Corsair CMX256A-3500C2 notched 440MHz.

Many memory makers have started to sell modules in pairs especially for dual-channel chipsets. Such modules are specifically tested for collaborative work. We checked out one such pair: OCZ EL DDR PC-3500 Dual Channel Enhanced Latency Series.

These modules also did well at 400MHz with 2-3-3-5 timings. Their specs say they can work at 434MHz with CAS Latency = 2. And these 434MHz are exactly what we got overclocking these modules at 2.6V voltage and with 2-3-3-7 timings. By the way, OCZ uses copper heatsinks on its modules and permits setting up to 2.8V voltage on memory chips. Of course, the overclocking results would be even higher in this case.

Among brand memory modules, Corsair CMX256A-3200C2 are immensely popular now.

And no wonder they are. These modules, just like their faster mates, could work at 400MHz with 2-3-3-5 timings. Further overclocking showed, however, that they don’t have a big “frequency reserve”. The maximum clock-rate they worked stable at with 2-3-3-7 timings and 2.6V voltage was 420MHz.

As our tests showed that memory timings seriously affect the performance, the use of high-quality memory becomes one more way to improve the performance of your system.

Conclusion

A year ago we showed that it made sense to overclock the Athlon XP system bus from 266 to 333MHz. It resulted in about 7-10% performance growth. In October, AMD used this possibility and rolled out processors intended for the 333MHz system bus.

Will the story repeat? We learned today that the transition of Athlon XP to the 400MHz bus also could increase their performance. This time, however, the performance growth is smaller, about 6-7% in some cases. Moreover, to get even this small growth, you have to use high-quality memory modules with pretty aggressive timings. Nonetheless, the shift to 400MHz bus is the easiest way available now to squeeze some extra speed out of the Athlon XP series.

The disadvantages of the 400MHz bus are obvious: older mainboards and chipsets have to be re-tested for stable and faultless work with it. Moreover, 400MHz-bus processors may have limited compatibility with older platforms and require new mainboards.

It’s hard to tell now, which way AMD will choose. But if you are in no mood to wait, do the transition yourself! You will need a mainboard capable of working at 200MHz FSB and high-quality DDR400 memory modules. You can look up our recommendations on the topic above. We would only like to add that not every DDR400 SDRAM is of help in the noble quest for raising Athlon XP system performance.