-
The only thing coming down the pike of interest is a feature; virtualization. All AMD plans to do performance wise is switch to DDR2, which will possibly hurt performance(bad latency). The first m2 chips will be just what's out now but with the m2 interface + features(there's also DRM crap but that's just a screw to the consumer, not a feature). AMD's move to 65nm supposedly third or fourth quarter. That's purportedly their roadmap. As bad as that is, they're also usually late(so.. 2007?). I can only hope Intel gets up to speed to force AMD to do more, provided they're able to respond. Oh, I've read one account of 65nm coming out second quarter(that'd be nice, but AMD being AMD, optimistic).
-
The tweaked 90nm process might be a boon for am2 as well if the rumours about it prove to be accurate.
-
-
DDR2 FB-DImms
No arguement on Socket F being DDR2
-
uh yeha, so where's the DDR3 gonna go?you realize that FB dimm means that the memory controller is on the DIMM itself, with less interconnects between the chipset and the memory, which, although it makes motherboard design simpler(69 pins for DDR2, in comparison to the 268 currently used), it also add latency to the current K8 tech as the memory control is not really on-die anymore, as well as allowing manufacturer's to use 2-layers of pcb interconnects over 3 for a single channel, and could thereby introduce greater crosstalk as OEM's try to make the boards as cheap as possible. of course, they have implemented the same sort of routing as netwrok cables to deal wit hthis problem, with the data being inverted and passed along another path to ensure data integrity,and we all know the problems that can arise from such situations.
Now, ok, FBdimms can handle DDR3 as well as DDR2, but the only real advantages are greater capacity, brought about by the fact that you can easily route more channels in the same space, which is also what brings the greater bandwidth. Because the interconnection is serial, like pci-e, you can see why they chose fbdimms to go with intergated pci-e control to the cpu, but you don't seem to realize that this tech is only really benficial in real-world work in a highly threaded environment, and not the computing environments that almost every user is used to now...in other words, it's only practical for servers that need large ram capacity, spread accros multiple channels.
the only thing new about FBdimms is that they break up the communication between the controller and module from travelling the same paths(which really isn't new at all), as current DDR tech uses, and routes each signal on a different trace. this means it might be possible to squeeze more info on that trace, but with the greater capacity adding simply by adding additional channels, and not higher-bandwidth connections, there is no real advantage to FBdimms over standard dimms, at least to us.
Maybe when new OS'es pring out that are multithreaded, but i definately isn't gonna help a windows machine.
-
Bingo.. You hit it on the head..
EVERY SINGLE major analyst says Thread Level parrelization is the Future of Processing..
ANd look FB-Dimms are made Specifically for that... Hmm isn't that surprising...
FB-Dimms bring a Brand new level to the bandwidth equation..
Right now there isn't a Memory Setup that would provide enough Bandwidth in a cost effective manner for an Eight Core system...FB-Dimm make that possible..
First it will go to the servers and then it will change and reach the rest of us
