About buses. INTEL/AMD

[patroclus02]

Hello!
I just want to make sure.

Intel, for core2duo platform, uses a single FSB of up to 1066MHz, to connect CPU to northbridge and memory.
So, if memory is 800Mhz, and, in hypothetical situation, all bandwidth is being used, there would be 266Mhz left for northbridge comunication.

AMD CPUs are connected to northbridge using hypertransport at 800 or 1000MHz (dual rate, so effective data transfers is 1600 or 2000MHz).
While, FSB is dedicated to memory accesses, working at the same freq as the memory. So, a 800Mhz memory would deliver 12.8Gb/s max (dual channel).
Adding a 2000Mhz HT bus, that can move 4 bytes per clock, we get 8Gb/s

Adding both together, we get 20,8Gb/s
While core2duo delivers a max bandwidth of 17Gb/s (dual channel, 800MHz memory). But northbridge comunication isn't dual channle, is it? If it isn't dual channel, only 8 bytes per clock can be moved (64 bit bus), so max bandwidth of 8,5Gb/s for northbrige exclusively.

AMD beats Intel here, because:
- More bandwith, and separate buses.
- Integrated mem controller.

BUT, even though, core2duo beats X2.
Imagine a mem controller and HT in core2duo
Terrific

 

[Frank]

 

[Jawed]

What applications are typically bandwidth limited?

Jawed

 

[Frank]

What applications are typically bandwidth limited?

Jawed

Most server apps, and some games.

Edit: and probably encoding video streams by now.

 

[Blazkowicz]

Quake 3 and winrar

 

[NocturnDragon]

Imagine a mem controller and HT in core2duo
Terrific

That would be Common System Interface for you.
Link

Intel is getting ready to introduce a chip communications technology that mirrors an approach central to recent successes of rival Advanced Micro Devices.

 

[patroclus02]

Even for winrar and games, core 2 duo whipes out AMD X2...
I've been an AMD fan since I was 13 years old, but I'm building a PC now, and I'll choose a core 2 duo. Great arquitecture. I'm sure AMD will come back

Common System Interface , I did not know about it... cool idea.

 

[hoom]

Intel has the shared & large L2 combined with apparently very good predictive pre-fetch populating the L2 with potentially useful stuff, both of which reduce the amount of time the core sits around waiting for data to process.