SUMMARY: After many hours of blood, sweat and tears we've finished our look at nVidia's latest chipset for the Intel Pentium 4 platform. It's the nForce 4 SLI Intel Edition or as its become known around the labs, the NF4 IE. It follows the features found in the nForce series of chipsets but helps fill a gap in the gaming market offering SLI gaming performance on the Intel platform.
nVidia have pushed the fact that they have pioneered Dual Channel DDR since the release of the first nForce 2 many years ago, unlike AMD though Intel don’t have a memory controller built onto the processor. Instead it is built into a separate chip so nVidia took this as a chance to start reworking Dual Channel and we see a lot of new technology implemented with the launch of their new chipset.
We will soon see if it’s just a whole lot of technical words or something that actually makes a difference. The memory controller tech brief is far from brief, weighting in at 13 pages. We’ll compress it down and give you the key features of the NF4 IE memory controller.
- Saturating Memory Bandwidth
When the Athlon XP implemented the 64-bit Front Side Bus at 133MHz (or 266MHz) effectively the maximum amount on bandwidth needed was 2.1GB/s. With DDR266 running in Dual Channel mode the maximum available bandwidth available was 4.2GB/s leaving a large amount of bandwidth for the CPU and GPU to access. Moving away from the past and into the present a 1066 FSB Intel Pentium 4 processor can make use of up to 8.6GB/s and with DDR2-667 running Dual Channel we have a total bandwidth of 10.6GB/s so the CPU has to access both memory channels so maximum bandwidth can be held.
Interleaving Memory Types
There are two modes of interleaving and depending on your memory configuration the nForce 4 controller kicks in and gives it the correct mode. The two modes we are talking about are finer-grain interleaving and coarser-grain interleaving. While we have seen a lot of sites throw these terms around, the extent of their explanation involves a copy and paste from the tech briefs which doesn’t explain it the best - after a quick call to the nVidia, we got a better understanding.
- Finer-Grain Interleaving: This is the better of the two interleaving processes and can be achieved by utilizing two memory modules of the exact same size, speed and brand (e.g. 2 x 512MB DDR-2 PC5400 modules) otherwise known as a symmetrical setup. This wouldn’t sound like anything too different as this is the general understanding of how you get Dual Channel up and running but this is were we move over to the second mode.
Coarser-Grain Interleaving: - If for some reason you don’t have two modules of exactly the same size, you can still get the benefit of Dual Channel. For example, if you have one 512MB and one 256MB DDR2 5400 module, you can still make use of Dual Channel. This memory setup is known as asymmetrical, and does cause performance to take a hit as you would expect.
While the second of the technologies might not be utilized as often as the first, it’s nice to know that it does exist. No matter what interleaving mode it uses though with the new memory controller the bus always operates at 128-bit-wide as opposed to 64-bit-wide when running the memory asymmetrically offering people as much performance as possible when running Coarser-Grain Interleaving.
Shared Address Bus
Instead of simply saying nVidia share the address bus so you get the best performance and inserting a block diagram that no one will understand, we will give you a quick run down on the Shared Address Bus. Most people would assume a Dedicated Bus is better as opposed to a Shared Bus but as usual memory has its own rules and makes everything a lot more confusing then normal.
The shared address bus lets the memory send one signal to the chipset instead of your normal two in a dedicated address bus. By shortening the length of travel the memory modules have to perform, you are able to run a more aggressive timing known as 1T. Until the release nForce 4 for AMD, the most important thing about memory was its ability to run at CAS 2, now we want the memory to run 1T address timing. 1T has become more important to Athlon 64/FX users over the past couple of months.
The advantage of 1T is going to be seen in a couple of instances - when you are showing off your SiSoft Sandra memory bandwidth and using applications that relay on accessing the memory heavily like image manipulation programs and so on.
