Diminishing returns. I think my mobo supports Gen 5 and I don't even care.

I don't even care that mine supports gen 4. My gen 3 Samsung 970 is plenty fast.

All modern mainboards share the M.2 PCI lanes with the GPU. How much performance loss on the dual graphics cards ? https://files.catbox.moe/9xn441.jpg

Robson:

All modern mainboards share the M.2 PCI lanes with the GPU. How much performance loss on the dual graphics cards ? https://files.catbox.moe/9xn441.jpg

AM5 boards have 24 PCIe lanes. Enough for an x16 GPU and dual M.2.

Its probably going to be a while before flash nvme is fast enough to max out PCIe 4.0. That being said, you watch Microsoft make PCIe 5 a hard requirement for DirectStorage for the PC.

Ojref:

Its probably going to be a while before flash nvme is fast enough to max out PCIe 4.0. That being said, you watch Microsoft make PCIe 5 a hard requirement for DirectStorage for the PC.

PCIe 4.0 has already been maxed out. DirectStorage is already enabled for any NVMe drive including PCIe 3.0 drives. It doesn't work on Storage Spaces volumes though. I would like to use striped PCIe 5.0 drives in Storage Spaces for games and other large media.

Monstieur:

PCIe 4.0 has already been maxed out.

There is no concrete evidence of that yet. Synthetic benchmarks are always going to over-stress the functionality.

schmidtbag:

There is no concrete evidence of that yet. Synthetic benchmarks are always going to over-stress the functionality.

What do you mean? The primary benefit of all new PCIe generations is higher throughput. If sequential transfer rate has been maxed out, the generation has reached its limit. Random read / write performance has nothing do with any PCIe generation.

Monstieur:

What do you mean? The primary benefit of all new PCIe generations is higher throughput. If sequential transfer rate has been maxed out, the generation has reached its limit. Random read / write performance has nothing do with any PCIe generation.

Higher throughput only matters when the lanes are saturated. That very rarely happens when gaming on a GPU with ample VRAM. This has been proven over and over again for the better part of a decade (look up PCIe scaling for the 4090 on Techpowerup). When there isn't ample VRAM, your system memory is bound to be a bottleneck before the PCIe bus, assuming your GPU didn't also have it's PCIe lanes crippled *cough*RX 6400*cough*. Where DS will really come in handy is by reducing load times since the GPU can feed data from more than just x4 lanes while also reducing dependency on the CPU. So long as developers don't treat the on-board storage as a replacement to VRAM (which it really ought not to be), I would be shocked if we're going to see a PCIe 4.0 @ x4 lanes become saturated for real world games. Again: synthetic tests (which from what I understand, is all we have data on) are going to push to unrealistic limits, because kinda the point of them is to know what the upper limit really is.

Glottiz:

Diminishing returns. I think my mobo supports Gen 5 and I don't even care.

seconded my boot drive is 4.0 (x570s) and although i bought a AM5 mobo (for 3DCache release) i'm not even close to being interested in a 5.0 drive given my use case scenario. all of my secondary (M.2) drives are 3.0, which is more than adequate for games and media (and actually my productivity related tasks too)

schmidtbag:

Higher throughput only matters when the lanes are saturated. That very rarely happens when gaming on a GPU with ample VRAM. This has been proven over and over again for the better part of a decade (look up PCIe scaling for the 4090 on Techpowerup). When there isn't ample VRAM, your system memory is bound to be a bottleneck before the PCIe bus, assuming your GPU didn't also have it's PCIe lanes crippled *cough*RX 6400*cough*. Where DS will really come in handy is by reducing load times since the GPU can feed data from more than just x4 lanes while also reducing dependency on the CPU. So long as developers don't treat the on-board storage as a replacement to VRAM (which it really ought not to be), I would be shocked if we're going to see a PCIe 4.0 @ x4 lanes become saturated for real world games. Again: synthetic tests (which from what I understand, is all we have data on) are going to push to unrealistic limits, because kinda the point of them is to know what the upper limit really is.

PCIe lanes aren't just for gaming. I max out my 4x NVMe drives' PCIe throughput copying video files. I need 4x Gen 5 RAID0.

Monstieur:

PCIe lanes aren't just for gaming. I max out my 4x NVMe drives' PCIe throughput copying video files. I need 4x Gen 5 RAID0.

The context implied DS, and gaming in general. Of course, once you exclude games then I could see how 4x NVMe drives could be bottlenecked, at least at gen 3. We're talking a pretty high-end niche where gen 4 drives in RAID 0 could be maxed out. More than likely, the CPU is the bottleneck in such a case.