DDR2-800 vs. DDR3-1333.
Does DDR3-1333 Make Sense?

There exists a special group of articles, which bring the following feedback in our e-mail boxes and forums: "Why did you publish this article? Your conclusions are evident!" They just forget that they come to evident conclusions in the process of reading this very article. :)

This article belongs to such category: comments to test results are sometimes so short that yours truly was tempted to skip them. However, we are of the opinion that such tests still make sense: yep, we've found out again that the old rule holds true. But if old rules are not reviewed from time to time, there is a chance to make a big mistake...

So, we're going to compare performance of two processors in combinations with different memory modules: standard DDR2-800 and higher clocked DDR3-1333. The Core 2 Extreme QX9650 processor (1333 MHz FSB) will be tested on one and the same motherboard that supports both memory types (Foxconn X38A). And the Core 2 Extreme QX9770 processor with the 1600 MHz FSB will be tried on two different motherboards. These motherboards come from the same manufacturer, the same series, and they are practically identical, except for their memory support (DDR2 and DDR3).

Hardware and Software

Testbed configurations

Components used in all tests:

Testbed processors:

Software

1. Windows XP Professional x64 edition SP1
2. 3ds max 9 x64 edition
3. Maya 8.5 x64 edition
4. Lightwave 3D 9 x64 edition
5. MATLAB R2006a (7.2.0.32) x64 edition
6. Pro/ENGINEER Wildfire 2.0
7. SolidWorks 2005
8. Photoshop CS2 (9.0)
9. Visual Studio 2005 Professional
10. Apache HTTP Server 2.2.4
11. CPU RightMark 2005 Lite (1.3) x64 edition
12. WinRAR 3.62
13. 7-Zip 4.42 x64 edition
14. FineReader 8.0 Professional
15. LAME 3.97
16. Monkey Audio 4.01
17. OGG Encoder 2.83
18. Windows Media Encoder 9 x64 edition
19. Canopus ProCoder 2.01.30
20. DivX 6.4
21. Windows Media Video VCM 9
22. x264 v.604
23. XviD 1.1.2
24. F.E.A.R. 1.08
25. Half-Life 2 1.0
26. Quake 4 1.3
27. Call of Duty 2 1.2
28. Serious Sam 2 2.07
29. Supreme Commander 1.0.3220

Testing

Essential foreword to charts

Our test method has two peculiarities of data representation: (1) all data types are reduced to one - integer relative score (performance of a given processor relative to that of Intel Core 2 Duo E4300, given its performance is 100 points), and (2) detailed results are published in this Microsoft Excel table, while the article contains only summary charts by benchmark classes. We will nevertheless focus your attention on detailed results, when needed. Besides, this table is unusual a bit: it states how much performance is gained owing to the new faster memory in each test. Hopefully, this information will come in handy to those inquisitive minds, who are not satisfied with brief general conclusions.

3D Modeling and Rendering

DDR3-1333 gives 2% performance gain to the QX9650. However, the QX9770 gains just 1.5% in this case, despite its faster FSB.

The most likely explanation is that in case of 1333 MHz FSB (real clock rate—333 MHz) and DDR3-1333 (real clock rate—667 MHz), the memory indeed operates at 667 MHz (FSB:DRAM = 1:2). But in case of the 1600 MHz FSB in the QX9770 (real clock rate—400 MHz) and DDR3-1333, the FSB:DRAM ratio is 2:3, so the real memory clock rate is just 600 MHz (so we actually get DDR3-1200).

In its turn, the QX9650+DDR2-800 combo uses the "awkward" FSB:DRAM ratio of 5:6, while the QX9770+DDR2-800 combo is operating synchronously—1:1. Thus, DDR3-1333 gets a performance bonus on the 1333 MHz FSB, while DDR2-800 gets it on the 1600 MHz FSB.

CAD/CAE

Here is an opposite situation - memory just does not need higher bandwidth on a generally slower 1333 MHz bus.

Digital Photo Processing

Detailed results are illustrative: the faster memory demonstrates the most noticeable effect with Sharp and Resize operations. Why it happens in the latter case is crystal clear to all readers, who understand the image resize algorithm.

Compile

The reaction is chilly. That may happen because compilers prefer using faster memory with lower latencies: cache. :)

Web server

That's not the worst result (~3% gain on the QX9650 and ~2% gain on the QX9770). But it's also not impressive.

Synthetics

CPU RightMark is traditionally indifferent to memory performance.

Archiving

Not bad. Strange as it may seem, WinRAR is more responsive to faster memory (see detailed results).

OCR

There's nearly no effect.

Audio Encoding

Old test, old software... and there is almost no effect from new technologies. :)

Video Encoding

It seems that video encoding should have responded to faster memory, because lots of data are processed. But no. We can only assume that codec developers do not rely on memory performance and process data in cache-sized chunks.

Games

That's probably the main disappointment: there is practically no effect in games! However, it's gamers that are traditionally enthusiastic about new products.

Total score

Conclusions

There is no miracle again: faster memory yields 1% performance gain on one of the fastest processors with the 1333 MHz FSB, while the only processor for the 1600 MHz bus gains 2% from it. As always, faster memory appears much earlier than we really need it. But let's not act as ignorant users, who shout about being cheated. There is no cheating here. Look at results and draw your conclusions. Sooner or later, we'll have processors that will be able to use the full potential of DDR3-1333. It would have been naive to think otherwise. It will just happen later, that's all.