ignorant memory/FSB speed question...

[Rift]

OK. I thought about posting this in the memory forum, but it seemed more newbie-friendly in here-heh. And I read the memory FAQ over there, but it seemed to only dance around my specific question.

I'm looking at upgrading my system to one of the new Core 2 DUO chips (probably E6600), which means a new motherboard and new memory as well. In my research, I noticed that they were using DDR2 800 memory with the test systems.

This confuses me. The FSB on the new CPUs is 1066. And I was under the impression that the memory clock speed needed to be in sync with the FSB of the CPU, which means it has to be either at that speed, or some derivative thereof (like 533 in the case of 1066 FSB speed).

Is this not the case? Is it just that it is ideal to have the memory clock and CPU FSB in sync? Or does it matter at all? Will the 800 memory "clock down" to 533 to sync up with the 1066? Would that make 1066 speed memory a better performer with the CPU? Someone explain this relationship to me please. Thanks.

 

[ArbY]

Great question! I'm sure you'll be receiving an answer shortly.

 

[weilin]

i dont have all hte answers but theoretically ddr2 533 in dual channel is all you will need to achieve intel 1066FSB. Anything faster would clock down and get better timings. People who buy ddr667 or faster really intend to overclock. As when u up fsb ur ram has to run faster to. Also your looking at test systems and becuase these people review comptuers for a living, htey have the best parts money can buy (or in their case gifts). So y not use them especially if they are backwards compatible.

 

[shadowduck]

OK. I thought about posting this in the memory forum, but it seemed more newbie-friendly in here-heh. And I read the memory FAQ over there, but it seemed to only dance around my specific question.

I'm looking at upgrading my system to one of the new Core 2 DUO chips (probably E6600), which means a new motherboard and new memory as well. In my research, I noticed that they were using DDR2 800 memory with the test systems.

This confuses me. The FSB on the new CPUs is 1066. And I was under the impression that the memory clock speed needed to be in sync with the FSB of the CPU, which means it has to be either at that speed, or some derivative thereof (like 533 in the case of 1066 FSB speed).

Is this not the case? Is it just that it is ideal to have the memory clock and CPU FSB in sync? Or does it matter at all? Will the 800 memory "clock down" to 533 to sync up with the 1066? Would that make 1066 speed memory a better performer with the CPU? Someone explain this relationship to me please. Thanks.

You are correct. FSB 1066 is basically 333 FSB quad pumped (4 operations per cycle) which is DDR2-533 RAM. You are correct there. However, any higher memory will clock down to sync with the FSB usually with tighter timings.

As weilin pointed out, DDR2-800 is mainly for people who wish to overclock. However, quad core Intel CPUs are rumored to run at FSB 1333 as not to stave the CPU for operations which would be DDR2-667. SO getting 800 is extending the timeperiod the RAM can be used in future upgrades as well.

 

[Scougs]

You are correct. FSB 1066 is basically 333 FSB quad pumped (4 operations per cycle) which is DDR2-533 RAM. You are correct there. However, any higher memory will clock down to sync with the FSB usually with tighter timings.

:roll:
Correction: FSB1066 is 4x266MHz. 4x333MHz would result in 1332.

 

[Rift]

Thanks guys--that definitely helps my understanding.

So then if DDR2 1066 is available, is there a distinct advantage of these over DDR2 533 when used with these CPUs w/ FSB 1066? I mean other than for the overclocking and/or future-proofing aspects.

 

[maury73]

Intel CPU are not connected directly to RAM, they lakcs an integrated memory controller.
The FSB is a bus that connects the CPU to the NorthBridge and SouthBridge and is the NorthBirdge that integrates the memory controller.
Speaking in very simple words, the CPU asks the NB to read a memory location, the NB does it and posts the read data back to the CPU.
It's the NB that has a memory BUS that runs at 400MHz (not 800, the real RAM clock is half the speed of DDR) no matter the FSB speed.

Of course the FSB must have enough bandwidth to sustain the RAM traffic and all other peripharals data transfers, but take on account that ion a typical system a DDR800 with tight timings transfers data at no more than 1/3 the theoretical maximum bandwidth.

 

[shadowduck]

You are correct. FSB 1066 is basically 333 FSB quad pumped (4 operations per cycle) which is DDR2-533 RAM. You are correct there. However, any higher memory will clock down to sync with the FSB usually with tighter timings.

:roll:
Correction: FSB1066 is 4x266MHz. 4x333MHz would result in 1332.
lol oops.. Well i was close enough.. FSB1333 is next anyway.

 

[theaxemaster]

Intel CPU are not connected directly to RAM, they lakcs an integrated memory controller.
The FSB is a bus that connects the CPU to the NorthBridge and SouthBridge and is the NorthBirdge that integrates the memory controller.
Speaking in very simple words, the CPU asks the NB to read a memory location, the NB does it and posts the read data back to the CPU.
It's the NB that has a memory BUS that runs at 400MHz (not 800, the real RAM clock is half the speed of DDR) no matter the FSB speed.

Of course the FSB must have enough bandwidth to sustain the RAM traffic and all other peripharals data transfers, but take on account that ion a typical system a DDR800 with tight timings transfers data at no more than 1/3 the theoretical maximum bandwidth.

It is actually more like half to two thirds, the new AMDs are getting that much. Perhaps the reason the intels aren't getting it is because of a NB running at 400 MHz? I'd never heard that, so I don't know the details. But the bandwidth figures I've seen for intels on DDR2 and AMDs on DDR400 would seem to suggest that.

I've never seen any evidence that a sync'd frequency or an even devision would hurt or gain you performance. But I tend to think that the closer the frequency they are running, the better, since the CPU should have to wait less.

 

[maury73]

AMD hasn't an FSB since many years: it has an integrated memory controller in the CPU, so it's obviously faster in RAM access because of lower latencies.

Without an integrated memory controller (like Intel) it doesn't matter if memory and FSB are in synch or not, it only matters the RAM speed and latencies as long as the FSB has enough bandwidth for RAM and all the other peripherals because it's a shared bus.