This is not what we were discussing if you'd actually read the posts. The discussion was about adding more lanes directly from the CPU, not via the chipset, so there's nothing untrue about it.
Going via a multiplexed solution via the chipset is an option, but the bandwidth to the CPU is then shared by all devices which can be a bottleneck. For most people it won't be and as I mentioned in another post in this thread, at least as far as NVMe drives are concerned, it doesn't seem to benefit AMD to have a direct link to the CPU as long as you only have one drive.
No, that's not what we are discussing. Re-read the original post, what lanes btarunr is talking about, and what you responded to on the first page, the post I responded to. We are talking about the lanes off the chipset.
particularly to address 300-series chipset's main shortcoming, just 6-8 older PCI-Express gen 2.0 general purpose lanes (while Intel chipsets put out up to 24 gen 3.0 lanes).
We are specifically talking about the chipset lanes. The actual fact is that Ryzen has
more PCI-E lanes coming off the CPU than Intel does. The Ryzen CPU provides 24 PCI-E 3.0 lanes, Intel's Coffee/Kaby/Sky Lake only provide 20 PCI-E 3.0 lanes.
Ryzen = 24 lanes = 16 for Graphics, 4 for NVMe, 4 for link to Chipset
Intel CFL = 20 lanes = 16 for Graphics, 4 for link to Chipset
With Intel, none of the NVMe drives are running from the CPU itself because of the limited PCI-E lanes from the CPU.
So, there are two options:
1.) We were talking about the PCI-E lanes coming off the CPU, in which case you were wrong in your statement about AMD being too stingy, because they obviously were not since they have more lanes than Intel.
OR
2.) We were talking about the lanes coming from the chipset(we were) in which AMD has the same bandwidth between the chipset and the CPU, which means they have plenty of room to expand the lanes provided by the chipset and you were wrong.
You can pick which situations you think we were actually talking about, but either way, you were wrong.
So you're suggesting adding M.2, 10Gbps Ethernet and anything that's going through the chipset today and don't think there will be a bottleneck between the chipset and CPU if multiple things are used at once? Good luck with that, as it's already been shown to be a bottleneck on Intel's platforms.
Except it hasn't really been shown to be a bottleneck in Intel's system, not that I've seen. I've run NVMe and 10Gb/s network through the chipset just fine. The link between the CPU and chipset is a 31.5Gb/s link. If you use 10Gb/s ethernet, that still leaves 21.5Gb/s for everything else. That is still an insane amount of bandwidth. Yeah, that might slightly bottleneck the fastest NVMe drives, but not to a noticeable degree. It's still 2.6GB/s of bandwidth! That's enough for an NVMe drive, a SATA storage drive, and some USB 3.0 ports to be active without any noticeable slowdown.
I really wish people could read before posting arguments for or against something without know where the discussion started...
Yes, indeed it would be nice if people wouldn't post without knowing where the discussion started...
Because, the fact is we were talking about the chipset lanes. Nothing in your original post suggests you, for whatever reason, were talking about the CPU other than you stating they need a new socket for more PCI-E lanes, which doesn't make any sense when you actually know what you're talking about. No one before you was talking about the CPU lanes, and you never mention in your post you were talking about the CPU. Everyone else in this thread is talking about the chipset.
And the fact is, AMD could actually make the link to the Chipset twice as fast an Intel very easily. They have 8 extra PCI-E lanes on their CPU while Intel only has 4. Right now, AMD divides the 8 lanes as 4+4, 4 for the chipset and 4 for an NVMe drive. However, if they develop a more capable chipset with more PCI-E 3.0 lanes, they could take the 4 lanes currently used for NVMe connection and instead use those for the connection to the chipset, giving twice the bandwidth between the CPU and chipset that Intel has. That would be 63Gb/s, or 7.8GB/s. More than enough for an NVMe RAID array, 10Gb/s ethernet, all the USB ports you would want, and plenty of SATA ports.
This likely could be done with the current processors too. The motherboard would decided where the PCI-E lanes are directed. If a CPU is put in an older board, then the current configuration is used. If the CPU is put in a motherboard with a newer chipset, then all 8 lanes are used for the chipset connection, and the chipset then provides more lanes for NVMe and other expansion cards.