A64 + motherboard mem controller..

[JeremiahTheGreat]

I'm just wondering if it was possible, with the Athlon 64 mem controller working as well. It could turn single-channel A64 into dual channel ! Is there any technical reason why this cannot be done?

 

[AndyHui]

You would have a horrible problem of non-symmetric latencies, which defeats the purpose of moving the memory controller to the processor.

 

[Peter]

Of course it can be done. That's in fact exactly what's happening when in a dual AMD64 setup, one processor accesses memory connected to the other processor.

You don't get dual channel memory from that though, because you don't parallelize two RAM controllers in a certain address location. What you get from that is separate RAM controllers, each owning a fraction of the system's total memory array.

 

[Peter]

Oh, and btw, it's up to the operating system to sort things out such that tasks running on a certain processor preferrably get allocated RAM that resides on the same processor node.

 

[Mingon]

Peter, out of curiosity do you know how the P4C syncs with the 865/875 chipset in dual channel mode? The p4's fsb iirc transfers on the begining and end of the signal both up/down, does this mean that one controller is delayed by half a cycle (if that makes sense) so as to sync the memory with the CPU.

 

[Peter]

You'd have to ask someone who designed that chipset - and I bet they won't tell

But in general, things aren't as synced as one might think they are. After all, memory isn't accessed by the CPU alone. AGP, other chipset parts, and PCI agents access memory too, without involving the CPU. And even the path to the CPU is deeply buffered and FIFOed.

 

[Mingon]

That sounds like a lot of buffering required, thanks for the response.