There seems to be some confusion about FB-DIMMs and why they are being used in the Mac Pro. There is no performance advantage in using FB-DIMMs, and in fact you get slightly less performance compared to regular DDR. FB-DIMMs are used by intel on their high end workstation/server boards for one reason only, they make it far easier and cheaper to allow mass quantities of RAM. In computational situations where you need 16, 32, 64GB or more of memory absolute speed may be secondary to having enough RAM available.
Performance is slightly worse than regular DDR memory because of the added latency each module will add when in place. Each FB-DIMM has what is called an Advanced Memory Buffer (AMB) which is essentially a small memory controller. Most other computers have a memory controller on the motherboard or the main CPU which essentially controls the memory directly. This has an added advantage of less latency because of the lack of go betweens. The disadvantage is as you add more sockets for more memory you need more pins on the memory controller and more wires on the motherboard. Adding more and more RAM capacity makes the problem get worse.
FB-DIMMS don't talk to the memory controller directly they have the AMB on each stick which gets the request first and also acts as a repeater for the requests that go downstream. The added benefit is that there are some fault corrections built in that impose no performance hit compared to a regular DDR based system. Theoretically a system based on the FB-DIMM architecture would have higher bandwidth and be cheaper and easier to manufacture with a much less complicated motherboard. The downside is all the AMB's inline will increase latency and they get hot so they will also introduce more heat into a computer that uses them. The price difference is mostly due the the low volume they are made compared to regular DDR, in fact the memory chips on a FB-DIMM theoretically can be anything as the AMB would be the controller that interfaces with the rest of the system and are really just regular memory chips. The good news on price is that intel's new Skull Trail if popular should make the FB-DIMM more popular and help bring down the prices.
As it stands today the reasons intel chose to use them in the server/workstation arena are to allow very large memory capacity, cheaper and easier to manufacture motherboards, and some better fault tolerance and correction which would appeal to professional, scientific and other high end usages.
In everyday use a system based on FB-DIMMs will likely be a little slower than a standard DDR one but it should be pretty close in any event. Mac Pros use Xeons which are essentially souped up Core Duo 2's with a few nice features which should actually more than make up for the latency associated with the FB-DIMMs, especially the brand new ones. The new Xeons have 12 MB L2 cache per processor (6 MB shared per pair of cores) and a 24 MB snoop filter on the FSB which essentially keeps track of all the cached data on both CPU's which should dramatically reduce traffic on the FSB and push bandwidth usage to the limits. There are quite a nice little bunch of tweaks in there.
Performance is slightly worse than regular DDR memory because of the added latency each module will add when in place. Each FB-DIMM has what is called an Advanced Memory Buffer (AMB) which is essentially a small memory controller. Most other computers have a memory controller on the motherboard or the main CPU which essentially controls the memory directly. This has an added advantage of less latency because of the lack of go betweens. The disadvantage is as you add more sockets for more memory you need more pins on the memory controller and more wires on the motherboard. Adding more and more RAM capacity makes the problem get worse.
FB-DIMMS don't talk to the memory controller directly they have the AMB on each stick which gets the request first and also acts as a repeater for the requests that go downstream. The added benefit is that there are some fault corrections built in that impose no performance hit compared to a regular DDR based system. Theoretically a system based on the FB-DIMM architecture would have higher bandwidth and be cheaper and easier to manufacture with a much less complicated motherboard. The downside is all the AMB's inline will increase latency and they get hot so they will also introduce more heat into a computer that uses them. The price difference is mostly due the the low volume they are made compared to regular DDR, in fact the memory chips on a FB-DIMM theoretically can be anything as the AMB would be the controller that interfaces with the rest of the system and are really just regular memory chips. The good news on price is that intel's new Skull Trail if popular should make the FB-DIMM more popular and help bring down the prices.
As it stands today the reasons intel chose to use them in the server/workstation arena are to allow very large memory capacity, cheaper and easier to manufacture motherboards, and some better fault tolerance and correction which would appeal to professional, scientific and other high end usages.
In everyday use a system based on FB-DIMMs will likely be a little slower than a standard DDR one but it should be pretty close in any event. Mac Pros use Xeons which are essentially souped up Core Duo 2's with a few nice features which should actually more than make up for the latency associated with the FB-DIMMs, especially the brand new ones. The new Xeons have 12 MB L2 cache per processor (6 MB shared per pair of cores) and a 24 MB snoop filter on the FSB which essentially keeps track of all the cached data on both CPU's which should dramatically reduce traffic on the FSB and push bandwidth usage to the limits. There are quite a nice little bunch of tweaks in there.
