M.2 SSDs are now cheaper than 2.5" sata SSDs. I already have a 1TB and a 2TB m.2 SSD in my system. I’m planning on upgrading to the new AMD zen5 cpus. But even if I get a motherboard with 4 m.2 slots I might risk using up pcie lanes for the GPU.
Sure I could buy a 4tb 2.5" sata SSD instead of an m.2, but with sata the speed is capped at around 560MB/s. It just hurts a little paying more for less performance
I checked the speed on 4x pcie 4.0 and it can do 8 GB/s. I haven’t been able to find anyone making a pcie switch so that multiple m.2 SSDs share a 4x pcie connection regardless if it’s pcie 4.0 or 5.0 . It will still be way faster than going with SATA. Of course I will have one or two new m.2 mounted on the new motherboards m.2 slots, but for my old drives I don’t care if they are bottle necked slightly while they share pcie lanes.
I know it’s not a lot of money we’re talking about, just a couple of m.2 drives. I just want to keep using all my storage, while not paying a ‘premuim’ for slower sata drives. Any ideas?
I’m unsure there’s any good answers here. Three NVMes is usually the max which fit on an ATX motherboard before getting into M.2 slots located under a dGPU’s exhaust. Choose a mobo with an x4 slot instead of another M.2 and pretty much always it’s positioned where an x4 switch card’s pretty obstructive of 2+ slot dGPU airflow and mechanically incompatible with a >3 slot dGPU. Reviews are pretty mixed for the few x4 to x2/x2 bifurcating cards I know, plus they’re all 3.0. I’m not aware of a mobo supporting x2/x2 bifurcation of its x4 slot to enable a low cost breakout board, which is probably pretty iffy for 4.0 signal integrity anyways.
There’s a couple things I’d suggest.
Look at the PCIe block diagrams of ASRock’s X670E Taichi, Pro RS, and PG Lightning manuals. Other motherboards give somewhat different tradeoffs but those three pretty well span the range of AM5 lane allocation options.
Consider quite decent 2 TB 4.0 x4 NVMes are around US$ 125 to 130 new and 4 TBs are 300. Not your objective, I know, but there aren’t many 4.0 switch cards under those price points. Or the US$ ~850 for an 8 TB, actually.
Cabling off an M.2 to a U.2/U.3 2.5 enclosure an option in principle but I don’t know of anything bifurcating offhand. Might be something suitable in the neverending story, though the cost effective options are likely 3.0. Given its bandwidth similarity to 3.0 x1 and 3.0 x2, USB 3.2 also has some relevance. Dual enclosures are few but, ironically, the state of things makes putting individual NVMes on gen 2 or 2x2 ports is actually not a terrible expansion method. AMD’s mandating USB 4.0 on some of the 800 series mobos for Zen 5 will up that a bit.
It’s unclear to me quite what this is referring to. AM5’s 28 lanes provide a CPU connected x16, two x4s that are often assigned to M.2s, and the x4 chipset downlink. With X670 boards most commonly there’s one x4 M.2 off each Promontory 21. The Taichis are the boards I know offhand which add a PCIe switch for an Intel type setup where the x16’s brought out to two x8 slots, though probably there’s high end boards from the other three which also do this.
If you mean desktop’s hardware tendency to motivate x4/x4/x4/x4 bifurcation of the x16 to enable additional M.2s instead of having a dGPU when more than four or five NVMes are needed, yeah, that’s a thing. The next alternative’s 7960-70-80-90X, though for what Storm Peak costs you can do two AM5 desktop builds and use one with drives and the other with a dGPU.
Might be two more M.2s on the back, might be a mislabeled dual. If it’s actually a quad there’s a decent chance two of the M.2s would work from an x4 slot, otherwise I’d expect just one. But it’s probably 3.0 and an x2/x2 bifurcating, 3.0 x4 dual M.2 card should be more like US$ 70.
Yeah, the STC PE382-4I product page on Shinreal’s website shows two on the back. But, like their Ali listing, doesn’t say anything about x4 support. The PEX 8747 is odd here as its 48 lanes are double the 24 needed for quad x4 NVMe with x8 uplink.
As I understand PLX’s line of 3.0 switches, what’s supported depends pretty much entirely on the card’s firmware. In the time I’ve had to search I haven’t found any good indication of how common x8 implementations handle narrowing to x4.
