~HeadShot~
Skilled
AMD and their upcoming sockets
Penstar Systems has an [RANK="www.penstarsys.com/#amd_socket"]article up on their site[/RANK] which should help to squash some of the confusion of AMD's upcoming socket changes (S1, M2, and Socket F) for the desktop and server processors:
Socket S1 is intended for the mobile sector, and it looks to be a very small version of Socket 754. This will be low profile, and aimed primarily at the low power/mobile sector. Socket M2 is the desktop replacement for Socket 939, and it has a total of 940 pins (though it is not pin compatible with the Opteron Socket 940). AMD is planning on using this socket with DDR-2, though there has been a lot of discussion that AMD may skip DDR-2 altogether (though I doubt DDR-3 will be ready in volume by Spring of 2006). This supposedly will give AMD the legs to continue to offer dual core Athlon 64's with plenty of power and bandwidth. In many ways, Socket 939 was a bit overdesigned and could handle some pretty hefty power loads, but it is just not smart to retain the same socket yet transition to DDR-2 (I can foresee many fried chips with a DDR memory controller trying to work with DDR-2).
[RANK="www.anandtech.com/news/shownews.aspx?i=24233"]Source[/RANK]
Original Article:
AMD, Sockets, and 64
AMD is of course making some big splashes these days, and the Athlon 64 X2 release was the biggest event this spring it seems. AMD will really push the X2 to the productivity market, but it will continue to present the single core chips to the gamers and enthusiasts out there. The FX-55 will continue its dominance as the highest speed processor out there until the release of the FX-57 in the Fall (or so I hear). The X2s are supposed to be available in June, but only through a few select partners. By August these parts should be widely available to users.
Now, currently these X2 parts are all aimed at the socket 939 platform, and none will show up on the Socket 754. By early next year though, AMD is planning another socket migration. This is an area where some people are really unhappy. In the past AMD has been great at keeping the socket infrastructure consistent, we can look at Socket 7, Slot A, and the very long lived Socket 462. Since the Athlon 64 has been released, on the desktop side we have already gone through the Socket 754 and now are into the Socket 939. The Opteron has held steady at 940 pins, but that is going to change as well.
Socket S1 is intended for the mobile sector, and it looks to be a very small version of Socket 754. This will be low profile, and aimed primarily at the low power/mobile sector. Socket M2 is the desktop replacement for Socket 939, and it has a total of 940 pins (though it is not pin compatible with the Opteron Socket 940). AMD is planning on using this socket with DDR-2, though there has been a lot of discussion that AMD may skip DDR-2 altogether (though I doubt DDR-3 will be ready in volume by Spring of 2006). This supposedly will give AMD the legs to continue to offer dual core Athlon 64's with plenty of power and bandwidth. In many ways, Socket 939 was a bit overdesigned and could handle some pretty hefty power loads, but it is just not smart to retain the same socket yet transition to DDR-2 (I can foresee many fried chips with a DDR memory controller trying to work with DDR-2).
The other socket that AMD is going to release is the Socket F, which is comprised of 1207 pins for the Opteron. This will also feature DDR-2, but one has to question why so many pins when the current dual core Opterons work fine on Socket 940? I think there are two distinct possibilities here. The first is that faster dual core chips will consume more power, so more power and ground pins need to be included for future products. However, with the current 90 nm dual core products, we won't see them scale much higher in speed for the next year, and that is nothing that the current socket can't handle. When AMD makes the jump to 65 nm, those power requirements will go down. So, why so many extra pins?
The first explanation is that AMD will be going with a quad core product for the server market when 65 nm is reached. The quad core at 65 nm would be about the same die size as a dual core at 90 nm (this is pure speculation, but we can approximate looking at how the design has changed in the past). However, quad cores running at 2.6 GHz+ will eat up a lot of juice, no matter what process is being used. But does this truly account for the greater amount of pins? This leads to the second spot of speculation.
Will AMD release the first CPU to introduce a 256 bit memory bus for DDR-2? If AMD is in fact going for a quad core, then the current 128 bit memory controller will be very bandwidth limited, even with high speed DDR-2 at its disposal. By making a 256 bit memory bus with DDR-2 800, then a dual core or quad core Athlon 64 core would be happily fed with bandwidth, all of which features the low latency of an integrated memory controller. This bit of speculation leads to other big questions, such as will DDR-2 receive a new pinout (can you imagine a 400+ pin DDR-2 module)? Or will AMD simply require these high end machines to utilize 4 DIMMS to enable the full 256 bit path?
While this is a bit far out, I think this is very possible. Would DDR-2 and extra power and grounds really take up all 267 extra pins? I think that another 128 data pins for the memory, plus the extra 139 pins for power and ground would fit quite nicely into that 267 pin budget. If this is true, then the server people must be drooling gallons by now. Imagine a dual socket motherboard with 2 x 256 bit memory paths which can host quad core processors. 8 logical processors in two sockets, with 12.8 GB/sec of bandwidth per socket (assuming this is DDR-2 800). That would be pretty amazing, not to mention very cost and space efficient. The best part of this speculation is that it can physically be done, and there are few barriers standing in AMD's way to implement such a design when they migrate to the 65 nm node. Looking at AMD's prices for their top of the line dual core Opteron products, I can only imagine how expensive a quad core would be...
[RANK="www.penstarsys.com/#amd_socket"]Source[/RANK]
Penstar Systems has an [RANK="www.penstarsys.com/#amd_socket"]article up on their site[/RANK] which should help to squash some of the confusion of AMD's upcoming socket changes (S1, M2, and Socket F) for the desktop and server processors:
Socket S1 is intended for the mobile sector, and it looks to be a very small version of Socket 754. This will be low profile, and aimed primarily at the low power/mobile sector. Socket M2 is the desktop replacement for Socket 939, and it has a total of 940 pins (though it is not pin compatible with the Opteron Socket 940). AMD is planning on using this socket with DDR-2, though there has been a lot of discussion that AMD may skip DDR-2 altogether (though I doubt DDR-3 will be ready in volume by Spring of 2006). This supposedly will give AMD the legs to continue to offer dual core Athlon 64's with plenty of power and bandwidth. In many ways, Socket 939 was a bit overdesigned and could handle some pretty hefty power loads, but it is just not smart to retain the same socket yet transition to DDR-2 (I can foresee many fried chips with a DDR memory controller trying to work with DDR-2).
[RANK="www.anandtech.com/news/shownews.aspx?i=24233"]Source[/RANK]
Original Article:
AMD, Sockets, and 64
AMD is of course making some big splashes these days, and the Athlon 64 X2 release was the biggest event this spring it seems. AMD will really push the X2 to the productivity market, but it will continue to present the single core chips to the gamers and enthusiasts out there. The FX-55 will continue its dominance as the highest speed processor out there until the release of the FX-57 in the Fall (or so I hear). The X2s are supposed to be available in June, but only through a few select partners. By August these parts should be widely available to users.
Now, currently these X2 parts are all aimed at the socket 939 platform, and none will show up on the Socket 754. By early next year though, AMD is planning another socket migration. This is an area where some people are really unhappy. In the past AMD has been great at keeping the socket infrastructure consistent, we can look at Socket 7, Slot A, and the very long lived Socket 462. Since the Athlon 64 has been released, on the desktop side we have already gone through the Socket 754 and now are into the Socket 939. The Opteron has held steady at 940 pins, but that is going to change as well.
Socket S1 is intended for the mobile sector, and it looks to be a very small version of Socket 754. This will be low profile, and aimed primarily at the low power/mobile sector. Socket M2 is the desktop replacement for Socket 939, and it has a total of 940 pins (though it is not pin compatible with the Opteron Socket 940). AMD is planning on using this socket with DDR-2, though there has been a lot of discussion that AMD may skip DDR-2 altogether (though I doubt DDR-3 will be ready in volume by Spring of 2006). This supposedly will give AMD the legs to continue to offer dual core Athlon 64's with plenty of power and bandwidth. In many ways, Socket 939 was a bit overdesigned and could handle some pretty hefty power loads, but it is just not smart to retain the same socket yet transition to DDR-2 (I can foresee many fried chips with a DDR memory controller trying to work with DDR-2).
The other socket that AMD is going to release is the Socket F, which is comprised of 1207 pins for the Opteron. This will also feature DDR-2, but one has to question why so many pins when the current dual core Opterons work fine on Socket 940? I think there are two distinct possibilities here. The first is that faster dual core chips will consume more power, so more power and ground pins need to be included for future products. However, with the current 90 nm dual core products, we won't see them scale much higher in speed for the next year, and that is nothing that the current socket can't handle. When AMD makes the jump to 65 nm, those power requirements will go down. So, why so many extra pins?
The first explanation is that AMD will be going with a quad core product for the server market when 65 nm is reached. The quad core at 65 nm would be about the same die size as a dual core at 90 nm (this is pure speculation, but we can approximate looking at how the design has changed in the past). However, quad cores running at 2.6 GHz+ will eat up a lot of juice, no matter what process is being used. But does this truly account for the greater amount of pins? This leads to the second spot of speculation.
Will AMD release the first CPU to introduce a 256 bit memory bus for DDR-2? If AMD is in fact going for a quad core, then the current 128 bit memory controller will be very bandwidth limited, even with high speed DDR-2 at its disposal. By making a 256 bit memory bus with DDR-2 800, then a dual core or quad core Athlon 64 core would be happily fed with bandwidth, all of which features the low latency of an integrated memory controller. This bit of speculation leads to other big questions, such as will DDR-2 receive a new pinout (can you imagine a 400+ pin DDR-2 module)? Or will AMD simply require these high end machines to utilize 4 DIMMS to enable the full 256 bit path?
While this is a bit far out, I think this is very possible. Would DDR-2 and extra power and grounds really take up all 267 extra pins? I think that another 128 data pins for the memory, plus the extra 139 pins for power and ground would fit quite nicely into that 267 pin budget. If this is true, then the server people must be drooling gallons by now. Imagine a dual socket motherboard with 2 x 256 bit memory paths which can host quad core processors. 8 logical processors in two sockets, with 12.8 GB/sec of bandwidth per socket (assuming this is DDR-2 800). That would be pretty amazing, not to mention very cost and space efficient. The best part of this speculation is that it can physically be done, and there are few barriers standing in AMD's way to implement such a design when they migrate to the 65 nm node. Looking at AMD's prices for their top of the line dual core Opteron products, I can only imagine how expensive a quad core would be...
[RANK="www.penstarsys.com/#amd_socket"]Source[/RANK]
