Please for the love of God no! Leave the 4 Dimms alone. At least that way when Quad channel memory comes into the mainstream we will already have the space set aside for it without losing MORE features.
The fact that PCI-e lanes are SO limited on the mainstream is pretty infuriating actually with the ability to do things like PCI-e bifurication for quad nvme drives on a single x16 slot also 10/40Gpbs being relatively easy to get into networking wise as well.
Im just looking back to the days of things like the X58 UD9 and imaging the mad setups people could do with that amount of PCI-E lanes even if they were all only Gen 4 at the moment.
What are you actually trying to achieve? The Asus X670e Pro Art has 4 M.2 slots, 10Gbe and USB4/TB4. You're either moving data around, or processing it in some way. 10Gbe is 1250MBps and 40Gb would be 5000MBps. Which is the max you can get data in or out of the computer, and both are lower than a single Gen4 NVME SSD. Never mind having the option of 2xGen5. That's also ignoring the fact that few operations are sequential to even get the max potential throughput, and NAND write holes. Most common disk access patterns are similar to or marginally better than SATA SSDs. Almost nothing nearly anyone would ever hope to do on the consumer platform needs or can even use all that bandwidth.
The 7950X can do decompression in something like 7-zip at a max of 2900MBps, and next to no other common CPU heavy task approaches even that. GPUs don't even really need the full x16 lanes for gaming or GPGPU tasks. Mining rigs run GPUs at 1x for example. Running the 2 x16 slots at x8 is going to be fine a majority of the time. Sure using the 3rd PCie slot on this board splits m.2#4 to x2, but that is still 4000MBps that can barely be put to full use anyway. You aren't really at risk of being IO limited for anything you're likely to ever do on the platform. Since you can put 4 on the board, a quad NVME card is effectively just a neat bucket of NAND, because there aren't a lot of ways to use the potential combined bandwidth. You'd end up needing a platform and CPU with more memory channels and cores before that. Gen 5 levels of bandwidth only might be meaningful by the time AM5 is EOL. Could stripe 2 Gen 5 NVME if you actually "needed" the bandwidth of a quad gen 4 card.
I can run a NAS in a VM with 4 threads and 8GB mem allocation and get 800MBps with 4 7200RPM HDDs, which is close enough to the peak of the 10Gbe interface without even using a NVME cache drive. On the onboard sata. I could double that with 8 HDDs and a HBA card in the 3rd slot. In theory a 4x sata SSD array would likely be north of 1500MBps. SSDs won't likely ever be cost effective for NAS use compared to HDDs though. If you just want several big buckets of NAND, might as well embrace the SATA ports.
