Because they can.Originally Posted by oldblue
If you've got an Opteron part with dual-core you can simply patch two chips toghether with a HT link in between.
It's easy to do, they've got the tech ready and it brings a lot of cheap power to the server market (and that's where the cash is).
A big plus about non-native quad core is that the yields get higher. You just patch 2 working cores together, so the yield is the same as the original dual-core part. With a quad core just because of statistics the number of failures doubles (twice as much cores on a chip so twice the chance of it failing).
This makes native quad cores much more expensive, also because they are more complex the yields will be even lower.
With non-native quad core it's so simple and cheap, and still holds a big gain.
For AMD the difference between non-native and native isn't as big as with Intel, this is due to the fast hypertransport interface and the integrated memory controller.
AMD would be crazy not to.
After the die-shrink 65nm parts hit the market (raising the clockspeeds) we will see non-native quad core about 1 maybe 2 quateres later. K8L will be a bit later I'm afraid.
The rumour is still out there Rev G (Deerhound) will have some extra ALUs raising the raw speed (IPC) to the Conroe level.
Combine that with integrated memory controllers and hypertransport you've got a winner, put it in quad core and you can outrun Core 2 for the next year.
Bookmarks