To illustrate the changed situation and altered possibilities with the Intel dual DDR motherboards, let us take two 2.4GHz processors.
The first is the kind of processor you can buy today, running at a default 133MHz FSB.
The second is a future 2.4GHz processor running at 200MHz FSB.
Let’s presume for argument’s sake that both are capable of about 3.25GHz.
A New Problem With The Old Way
Standard procedure for PIVs up to now has been to yank the FSB up, and set the memory to run even faster than the FSB.
We take our 133MHz FSB 2.4, and eventually get it cranked up to 180MHz using a 1:1 FSB/memory ratio, giving us a CPU speed of 3.24GHz. We see to our pleasure that Sandra memory bandwidth has jumped up to well over 4000, and we can’t wait until we crank the memory up to 225MHz.
To our horror, we get nowhere near the improvement we’ve seen before when we did this with earlier PIV boards, actually not much better than we got running 1:1.
Why? It’s due to dual-channel.
With single stick DDR systems, the total available amount of FSB bandwidth at a certain frequency was always double the memory bandwidth of a single DDR stick at the same speed. Therefore, there was plenty of room left for memory bandwidth when you ran memory 25% faster than FSB.
With dual stick DDR system, the total available amount of bandwidth at a certain frequency is the same as the memory bandwidth of two DDR sticks running dual channel. There is little room left for memory bandwidth when you run memory in a dual-channel DDR system faster than FSB.
An Intel dual channel DDR system faces essentially the same FSB bottleneck as Athlon systems when using 133MHz FSB processors (though obviously at a much higher level).
New Processor, New Opportunities, New Problems
Let’s now take this 2.4 200MHz processor. At first, it looks like we have a big problem overclocking this puppy because there’s no RAM in the world that’s going to run at an FSB fast enough to get us a decent overclock from the CPU, much less faster than the CPU.
Then we find out that we can run memory slower than FSB, at a ratio of 4:5. So we take that route, and find that we can reach 270MHz FSB and a CPU of 3.24GHz, just like in the first example. Our high-end memory is running at 216MHz, slower than the 225MHz we tried before.
With fear and trepidation, we run Sandra again and to our amazement and disbelief we get a memory score rather higher than we did before.
How can this be? How can you run memory slower, and then get a higher memory score?
The answer is simple: no more bottleneck.
By increasing the FSB to 270MHz, you’ve done the equivalent of expanding the FSB bandwidth highway from four to six lanes. Running dual-channel memory at around 225MHz is like having five rows of cars. They don’t all fit on a four-lane highway, but they do on a six-lane highway.
What’s the catch? The catch is “Can you get the motherboard to run at 270MHz FSB or more easily?” Apparently, a couple extreme overclocking attempt have worked, but we don’t know yet if this can consistently be done with Canterwoods, and Springdales are anybody’s guess.
Does It Really Make That Much Difference
Depends on what you mean by “difference.”
If you worship SiSoft Sandra like some ancient stone god, it is indeed the difference between life and living death. If you take the 133MHz FSB route and provide such an unworthy offering to the god, you will be mocked and scorned and virtually stoned by fellow worshippers until your inevitable suicide (or at least your purchase of a 200MHz processor).
If, on the other hand, you are sane and don’t belong to the memory cult, we’re talking about maybe a 2-3% difference in a few games or programs, and less for the rest.
The 200MHz route also gets you hyperthreading. For the moment, the best simple way to look at hyperthreading for at least a while is that it won’t improve your scores in any one intense activity, but it will make your computer run more smoothly when you have multiple mundane tasks.
Neither is an unreasonable choice. Just depends on your priorities.