Sign in with
Sign up | Sign in
Your question

Why only 16 lanes??

Tags:
  • CPUs
  • Sandy Bridge
  • PCI Express
Last response: in CPUs
Share
September 12, 2012 6:28:47 PM

Hey guys,
I was just wondering why Intel decided to give basically its whole line-up (Excluding Sandy Bridge-E) only 16 lanes of PCI-e connectivity? Is it just to set it apart from Sandy Bridge-E, or is it something that is difficult to implement? While I do understand that your average user doesn't care about only having 16 lanes, but they could make a line-up for the enthusiast gamer. For instance the K line of processors could have say 32 lanes. Just one of those thoughts I had....

More about : lanes

a b à CPUs
September 12, 2012 7:32:11 PM

Because that would violate PCI-E specifications.
m
0
l
a c 85 à CPUs
September 12, 2012 7:42:53 PM

it'll be hardware dependent which would make the k series a different design and hence more expensive. But essentially it is a differentiator between SB-E and SB.

I thought that it more than 16 lanes available to it, but regardless it is indeed less than sb-e.
m
0
l
Related resources
September 12, 2012 8:26:03 PM

How would it violate the PCI-E specifications??? Unless you mean that the maximum slot width is only 16 lanes? If that is indeed what you mean then I completely understand it at this day and age more than 16 lanes for a single graphics card isn't really necessary (x8 is even considered good enough sometimes) What I meant is why did Intel only give the processor 16 lanes in total limiting us to basically 2x GPU's before running out of bandwidth? The reason I ask is not just GPU based. Other stuff could also run of off PCI-E lanes. Some of my SATA ports use PCI-E 2.0 lanes through an Asmedia controller. This is why I would think that it would be a good idea to also give your normal run of the mill processors more lanes. I was just wondering whether there is some limitation that makes it difficult to implement more the 16 lanes of connectivity?
m
0
l
a b à CPUs
September 12, 2012 8:55:20 PM

bigbasedrum said:
How would it violate the PCI-E specifications??? Unless you mean that the maximum slot width is only 16 lanes? If that is indeed what you mean then I completely understand it at this day and age more than 16 lanes for a single graphics card isn't really necessary (x8 is even considered good enough sometimes) What I meant is why did Intel only give the processor 16 lanes in total limiting us to basically 2x GPU's before running out of bandwidth? The reason I ask is not just GPU based. Other stuff could also run of off PCI-E lanes. Some of my SATA ports use PCI-E 2.0 lanes through an Asmedia controller. This is why I would think that it would be a good idea to also give your normal run of the mill processors more lanes. I was just wondering whether there is some limitation that makes it difficult to implement more the 16 lanes of connectivity?


More PCIe Lanes means more transistors, more heat, more power and more overall complexity. However, all of these pale in comparison to the cost of extra pins on the CPU package. Each PCIe lane requires 2 pins (a differential pair for noise cancellation) and two lanes routed across the board, so doubling the number of lanes to 32 would cost 32 extra pins minimum. Multiply that by several million CPUs and motherboards sold and the cost adds up.

For each generation of CPU that Intel releases, they carefully consider how many PCIe lanes are needed by a particular user group, and dedicate the required number of pins for that SKU (mobile, desktop, xeon). For enthusiast desktop that number is 16-20 pcie lanes. For extreme desktop and server it is 40 lanes (per socket).

If you find yourself needing more PCIe lanes at a reasonable price then pick up a quad core LGA2011, the i7-3820.
m
0
l
September 12, 2012 10:22:46 PM

Blandge said:
More PCIe Lanes means more transistors, more heat, more power and more overall complexity. However, all of these pale in comparison to the cost of extra pins on the CPU package. Each PCIe lane requires 2 pins (a differential pair for noise cancellation) and two lanes routed across the board, so doubling the number of lanes to 32 would cost 32 extra pins minimum. Multiply that by several million CPUs and motherboards sold and the cost adds up.

For each generation of CPU that Intel releases, they carefully consider how many PCIe lanes are needed by a particular user group, and dedicate the required number of pins for that SKU (mobile, desktop, xeon). For enthusiast desktop that number is 16-20 pcie lanes. For extreme desktop and server it is 40 lanes (per socket).

If you find yourself needing more PCIe lanes at a reasonable price then pick up a quad core LGA2011, the i7-3820.



While I agree to your 2º paragraph I don't agree with your first, AMD boards with 990FX aren't that much expensive than 990X boards, the problem is not the cost of the board is the CPU design and the chipset, the proof is that there are Z77 boards that support dual 16x but require and PLX PEX 8747 chip that is basically a "multiplexor"


But yeah normal desktop PC doesn't benefit from dual x16, some test even show that difference between SLI/CX dual 16x and dual 8x are negligible, also I doubt that there is some hardware out there targeted to desktop PC that actually uses all the bandwich of PCI 2.0
m
0
l
!