Originally Posted by willyolio it seems like intel's recent strategies have been: copy AMD's plans, use a whole lot of money to get it out before they do.
When you're dealing with this technology, eventually you end up with the same conclusions I'm guessing, and honestly, Intel may somewhat rip off on AMD, but they're furthering the technology and/or building other technologies on top to make it more performing
Nehalem will use Intel's CSI which is a serial link system like HT. The Register has had various stores about it for the past few years.
With PCIe, the limits seem to be mostly on the clock and number of links. the tech bares a remarkable resemblence to infiniband, but with the software interface of PCI.
I wonder if they will use something like the AMB on FBDIMMs so the platform is not necessarially stuck with a memory tech the same way AMD effectivly is.
New tech is always great.
I don't always understand everything in these articles, but they are hugely informative. Thanks Rich.
Once a mini-ITX comes out with either AMD or Intel integrated gpu and low power which can also do a decent job at DX9 gaming, I'm sold.
That'll be all I need.
I can't say that I am all that familiar with the protocol used by FBDIMMs, whether it is just serialized DDR2 traffic or an independent protocol. I do that it is a serial memory communication system with a part called an AMB on the DIMM that handle the more 'mundane' tasks with handling RAM. I believe it was implied that this is really the only part dependant on the memory technology used though.
If this is the case, then using faster memory would be a case of more bandwidth and a new AMB to make use of it.
The idea is almost anti-rambus as the idea is to use off the shelf memory tech rather than reinvent it to deal with the problems of DDR. The interface is another serial interface (it may even be based in CSI) between the memory controller and the AMB. The link allows for a lot less complexity in the wiring to the memory, and for it to be placed up to 12 inched from the memory controller. If the interface can scale like other serial link systems such as Hypertransport and PCIe, then it could be useful.
I will be curious to see what happens when the Xeon version of Nehalem hits to see if it is still using FBDIMMs (and if they have gotten any cheaper by then).
It sounds interesting and certainly beneficial, but FBDIMMS in their current state are even more expensive than DDR3 and require some massive heatsinks to keep cool. This is non-ideal for a consumer situation and it'll lead us back to the days of Prescott.
Even if it's not royalty free, market mechanisms are hugely behind DDR3 at the moment and whether that investment can be ported to FBDIMM manufacture easily is yet to be seen. :)
I will try to remember to ask Intel about CSI when I'm at IDF ;)
Comments 26 to 33 of 33
ReplyWhen you're dealing with this technology, eventually you end up with the same conclusions I'm guessing, and honestly, Intel may somewhat rip off on AMD, but they're furthering the technology and/or building other technologies on top to make it more performing
With PCIe, the limits seem to be mostly on the clock and number of links. the tech bares a remarkable resemblence to infiniband, but with the software interface of PCI.
I wonder if they will use something like the AMB on FBDIMMs so the platform is not necessarially stuck with a memory tech the same way AMD effectivly is.
That's good to know about CSI, cheers ;)
I don't always understand everything in these articles, but they are hugely informative. Thanks Rich.
Once a mini-ITX comes out with either AMD or Intel integrated gpu and low power which can also do a decent job at DX9 gaming, I'm sold.
That'll be all I need.
If this is the case, then using faster memory would be a case of more bandwidth and a new AMB to make use of it.
The idea is almost anti-rambus as the idea is to use off the shelf memory tech rather than reinvent it to deal with the problems of DDR. The interface is another serial interface (it may even be based in CSI) between the memory controller and the AMB. The link allows for a lot less complexity in the wiring to the memory, and for it to be placed up to 12 inched from the memory controller. If the interface can scale like other serial link systems such as Hypertransport and PCIe, then it could be useful.
I will be curious to see what happens when the Xeon version of Nehalem hits to see if it is still using FBDIMMs (and if they have gotten any cheaper by then).
Even if it's not royalty free, market mechanisms are hugely behind DDR3 at the moment and whether that investment can be ported to FBDIMM manufacture easily is yet to be seen. :)
I will try to remember to ask Intel about CSI when I'm at IDF ;)
-
« Previous
-
1
-
2
-
Next »
Discuss in the forums