Overclocking the nForce2: the Dual Channel Edge, a Brief Comparison

[RaynorWolfcastle]

Many nForce2 owners wonder whether or not the dual channel memory controller on their board should be used when buying a new system. This brief review has the specific objective of comparing a fast dual channel setup with a faster single channel setup on an nForce2-based motherboard.

In short, I aim to answer the question: ?Should I buy one stick of memory or two, for my new nForce2 set up??

This review has only come about because I happened to have 3 512 MB sticks of high performance RAM, so the variables are not as controlled as I would have liked them to be; because of this, I classify this as a ?comparison? instead of a ?review? in which you might expect a more rigorously controlled environment. If you are looking for more formal nForce2 reviews there are plenty of other sites that have already been written them.

Also note that this comparison is aimed at the enthusiast as both the motherboard and CPU are overclocked significantly. Be warned, overclocking is strictly at your own risk.

Test system:

Common components
Motherboard: Epox 8RDA+, rev. 1.1 ? nForce2 IGP (A2 stepping) & nForce2 MCP-T
CPU: AMD Athlon XP 1700+ (Thoroughbred A)
Video: Radeon 8500LE @275/275 (AGP locked to 66 MHz)
Network: nForce2 MCP-T, set to ?CPU? optimization
Hard disk: Maxtor 80 GB 7200 RPM, 740DX
DVD-ROM: Pioneer 16x DVD-116
Sound: Onboard nForce2 SoundStorm sound

OS: Windows XP Pro, SP1
BIOS: May 07, 2003
Platform drivers: nVidia nForce2 2.41 drivers
Video Drivers: Ati Catalyst 3.4


Dual channel specific components:
1 x 512 Corsair PC3200C2 set to ?Aggressive? 5-2-2, CAS 2 timings
1 x 512 Kingston HyperX PC3000 set to ?Aggressive? 5-2-2, CAS 2 timings
1:1 FSB-to-memory ratio

Single channel specific components:
1 x 512 Corsair PC3200C2 set to ?Aggressive? 5-2-2, CAS 2 timings
1:1 FSB-to-memory ratio

Index:
200MHz SC: Single channel, 200 MHz FSB, 1704 MHz CPU
213MHz SC: Single channel, 213 MHz FSB, 1704 MHz CPU
200MHz DC: Dual channel, 200 MHz FSB, 1704 MHz CPU

*note that the HyperX memory could not run faster than 200 MHz at these timings, which is why the Dual channel results are limited to 200 MHz FSB


The Results:

SiSoft Sandra 2003 (build 2003.3.9.44) ? Higher is better, in MB/s
Setup Read/Write
200MHz SC: 2907/2620
213 MHz SC: 3071/2760
200MHz DC: 2921/2673

Aida32 (build 3.50) ? Higher is better, in MB/s
Setup Read/Write
200MHz SC: 2926/1129
213MHz SC: 3072/1206
200MHz DC: 2958/1164

Stream ? Higher is better, in MB/s
Setup Copy/Scale/Add/Triad
200MHz SC: 1244/1147/1317/1289
213 MHz SC: 1196/1147/1295/1281
200MHz DC: 1123/1084/1232/1228

Cachemem (Version 2.6) ? Higher is better, in MB/s
Setup Read/Write
200MHz SC: 1785/1199
213 MHz SC: 1849/1198
200MHz DC: 1750/1140

Cachemem (Version 2.6) ? Lower is better, in cycles
Setup Latency (4K pointer movement)
200MHz SC: 173
213 MHz SC: 147
200MHz DC: 173

ScienceMark 1.1 ? Higher is better
Setup Read/Write (MB/s) Score
200MHz SC: 1183/1115 155.64
213MHz SC: 1891/1263 154.22
200MHz DC: 1770/1182 156.30

Improvement from SC to DC: 1.3%

Divx Encode (DivX Pro 5.05, 4 passes at 1000 KB/s, Swordfish: Scene 2) ? Lower is better
Setup Time (min: sec)
200MHz SC: 11:45
213MHz SC: 11:39
200MHz DC: 11:31

Improvement from SC to DC: 1.99%

Zip Compression (Warcraft 3, using WinRAR, zip format, best compression) ? Lower is better
Setup Time (min: sec)
200MHz SC: 1:31.5
213MHz SC: 1:31
200MHz DC: 1:30

Improvement from SC to DC: 1.67%


Conclusion:
Seeing the results above, we realize that the second memory channel does not improve memory bandwidth in any significant way when the FSB and memory are run synchronously. However, there are small improvements in applications. DivX encoding was the tested application that most benefited from the DC mode and it only gained a meager 2% on its SC counterpart.

Note that these applications were not chosen to be bandwidth limited so there is a good chance that the CPU is playing an important role in the close scores that we see.

Also note that increasing the FSB by 6% significantly closed the gap on the DivX test, reducing it to a mere 1.1%. Moreover, some tests yield better results with a slower FSB than a faster one in SC mode; I can?t really explain why this happens.


Recommendations based on test:

For those buying 1 GB of RAM:
Dual channel is the only way to go as 1 GB modules of high speed DDR RAM are very expensive at the moment. Get 2x512 MB of your favorite manufacturer?s RAM

For those buying 512 MB of RAM:
Unless you live for that very small performance improvement, get a single 512 MB stick of the fastest RAM you can afford. As you can see in the tests, the performance improvement for dual channel RAM is very slim. Buying 1 stick will have you ready however if you ever choose to step up to 1 GB of RAM.


- RaynorWolfcastle

 

[Jeff7181]

Am I reading it right? You used a stick of Corsair paired with a stick of Hyper X in your dual channel test? Shouldn't you have used the same kind?

 

[TenEgg]

If he can run at that speed, why not?

 

[RaynorWolfcastle]

Originally posted by: Jeff7181
Am I reading it right? You used a stick of Corsair paired with a stick of Hyper X in your dual channel test? Shouldn't you have used the same kind?

Ideally yes, but in practice it doesn't make a difference as long as it works. It's a "it works or it doesn't"-type deal since memory is a passive device. That's why there's no point in benchmarking different brands of RAM at the same speed, you're actually measuring the chipset's performance. The chipset issues commands and either the RAM can perform them or it can't. Period. That's why it doesn't matter that the RAM used in dual channel mode is not identical.

Using two matched sticks of RAM however, minimizes the *chance* of incompatibility and also makes it more likely to have two sticks of RAM that have the same maximums. In this case the HyperX RAM can't quite keep up with the Corsair RAM so I have to limit the benches to the FSB speed (200 MHz) where they can both operate. Keep in mind that this can (and does) happen even with two sticks of the same make and brand when overclocking.

I hope that clears up the confusion

 

[Jeff7181]

Originally posted by: TenEgg
If he can run at that speed, why not?

Because he said the dual channel setup was limited to 200 Mhz testing because the Hyper X wouldn't run any faster...

 

[Boogak]

Nice writeup. Even if the HyperX can't run over 200mhz, he did provide benchmarks for 200mhz single channel and 200mhz dual channel w/ the same CAS settings, so it's a valid comparison.

 

[RaynorWolfcastle]

no one likes my comparison? :brokenheart:

 

[mechBgon]

Originally posted by: RaynorWolfcastle
no one likes my comparison? :brokenheart:

I like it! w00t, we got a compression benchie using WinRAR and some other real-world stuff. And what I'm concluding here is that the new nForce2 400 non-Ultra (single-channel-only) boards may be a good buy for some purposes, when they show up.

I read that the non-Ultra nForce2 400 boards can't use the MCP-T southbridge, which would rule out the nVidia APU if it ends up being true. Asus has the specs up for the A7N8X-X and it is indeed a MCP board, not MCP-T. I'll have to read up on that more later.