I think that NV actually has a real limitation with SLI HB. That link states NV is doubling the SLI bandwidth but yet it's only limited to two 1080 cards. We know that a 1080 has 2x SLI fingers. It's highly likely that NV's new SLI HB bridge is getting 2GB/sec
because it combines both of the SLI fingers on 2 cards.
Since the cards require 2 separate SLI fingers
working together to double the bandwidth, if you introduce a 3rd card, this type of setup wouldn't work. It means there is no flexibility at all in this type of an arrangement.
It also means old SLI bridges won't work without a performance penalty; and of course NV will start selling these
upgraded SLI bridges for a fee. In AMD's case, the upgraded XDMA was a free feature. :sneaky:
What makes it worse, is you have to buy a new SLI bridge depending on the mobo you have because it's not PCIe.
$70/100 premiums for reference coolers + $20-30 for SLI bridge, it all starts to add up. More $$$ to NV, less flexibility to PC gamers.
It's also not as easy anymore to just buy the new SLI HB bridge. Because they are rigid and larger in size, you have to do research as certain after-market cards may not have sufficient space between the SLI fingers and the extended AIB heatsink.
AIBs will need to think carefully now about how they design after-market heatsinks to not interfere with the rigid SLI HB bridge.
SLI BRIDGE (1st gen)
What is the function of the SLI connector?
The SLI connector is a proprietary link between GPUs that transmits synchronization, display, and pixel data.
The SLI connector enables inter-GPU communication of up to 1GB/s, consuming no bandwidth over the PCI Express bus.
http://www.geforce.com/hardware/technology/sli/faq
By deducation then => SLI HB =
2GB/sec
vs.
AMD's Crossfire Bridges were limited to 0.9GB/sec:
"In AMDs current CFBI implementation, which itself dates back to the X1900 generation, a CFBI link directly connects two GPUs and has 900MB/sec of bandwidth." ~
AT
Year 2013 AMD XDMA =
16GB/sec
Shockingly after introducing NV Link, NV's Pascal GP104 has 8X less available bandwidth between 2 cards than AMD 2013 XMDA does. This means AMD has more available bandwidth even in Quad-CF on an X99 platform or Z170 platform with PLX chip than NV has on a dedicated SLI HB bridge with 2 cards.
If PCIe 4.0 comes out, AMD's bandwidth would suddenly jump to 32GB/sec!
Thanks a lot! I am nowhere near as knowledgeable as some of the guys on this forum who deeply understand GPUs. I actually have to research and read up on the technology before I post because often I don't even know how the technology works. I am not an electrical engineer either and don't work in the GPU industry. What I do have is pretty good memory which allows me to find key information quickly since I recall reading it on website ABCD for product XYZ.
If you have the interest in tech overall, there are a lot of great articles that help learn. For example, just
stumbled on this so I'll read it to see if I can learn something new.