PCIe compatibility

[SemperFitz]

Morning all,

I'm going to buy a M2 for my MSI B350 gaming pro MB, and started looking at various options/manufactures/pricing, but then i got confused with PCIe gens.

The spec on my MB says -

1 x M.2 slots (Key M)
- Supports PCIe 3.0 x4

and i was thinking of buying WD_BLACK SN850X, but that says -
PCIe Gen4

Are they compatible?

Yours confused

 

[alec]

Yes, fully backwards compatible
And not much drawback except lower peak bandwidth

 

[SemperFitz]

thank you

 

[mickyflinn]

You can save a £20+ if you go with the WD Blue SN5000, the 1tb goes for around £52 ATM and doubt you notice any difference in performance.

 

[Tetras]

and i was thinking of buying WD_BLACK SN850X, but that says -
PCIe Gen4

Are they compatible?

If you look at the datasheet WD provide for the drive, it mentions at the bottom that it is backwards compatible, so I assume WD would keep this compatibility for the drives produced after the datasheet.

 

[Avalon]

All PCIe generations are backwardly compatible in the same way a USB 3 device will work in an original USB port, generation and bandwidth per lane are the only things that change.

 

[jigger]

All PCIe generations are backwardly compatible in the same way a USB 3 device will work in an original USB port, generation and bandwidth per lane are the only things that change.

That’s not entirely accurate, there some potential gotchas especially with .x versions. The two mains problems to watch for are power delivery and signal quality. There are also some operational and vendor specific parts to the PCI-E standard* and some Intel implementations of gen 4 and 5 are problematic.

 

[Avalon]

That’s not entirely accurate, there some potential gotchas especially with .x versions. The two mains problems to watch for are power delivery and signal quality. There are also some operational and vendor specific parts to the PCI-E standard* and some Intel implementations of gen 4 and 5 are problematic.

The use of a OEM modified PCIe connectors are no different to the mining days where our risers used a non standard USB connector to provide the required power and data lines, use those cables in a standard USB slot and you will potentially have a bad day. Some OEM’s have chosen to use non standard PCIe slots to expose PCIe lanes for various reasons, HP, Dell and ASUS server class kit spring to mind (eg that’s what I tend to encounter), I am sure other OEM’s will have done similar, but the clue with a non standard interface is the non standard part, it kind of gives the game away.

Other general exceptions to the PCIe standard do exist, the SMBUS pins for example are normally not used on consumer grade boards, so they aren’t connected by many OEM’s, but over the years a number of Dell (Broadcom) fiber cards, the Sun F20 NAND accelerators, early HBA’s (Broadcom) and even Nvidia 50 series GPU’s have had a hissy fit when it finds they are connected, but those are widely known and documented, as is the fix: a tiny bit of Kapton tape over pins 5 & 6.

Then let’s look at BIOS implementation, most consumer OEM’s didn’t bother to enable rebar despite it being part of the PCIe standard for many years. Then AMD couldn’t stop saying it so often that OEM’s enabled it for the most part. We’ve also seen various BIOS updates to improve timing accuracy with NVMe, again less to do with ‘the standard’ and again more to do with the OEM’s producing hardware for it needing the goal posts moved slightly.

So yes, any PCIe compliant card is backwards compatible with earlier standards of PCIe, that’s one of the basic principals of PCIe standard. No you can’t fit a bespoke OEM custom PCIe card into a PCIe slot without a suitable adapter (Ali-Exlress sells some weird custom to standard adapters), and per the examples provided, some OEM’s can and do interpret the standard with a certain level of latitude that sometimes requires intervention, though often not by the OEM who caused the issue (motherboard manufacturers are often required to push BIOS fixes for other OEM’s stupidity).

 

[LuckyBenski]

All PCIe generations are backwardly compatible in the same way a USB 3 device will work in an original USB port, generation and bandwidth per lane are the only things that change.

Well, if you're talking about storage maybe. Plenty of USB devices require a certain version to work e.g. you won't be able to attach a USB 4 dock to a USB 1.1 host device, and expect anything to work.

 

[jigger]

Well, if you're talking about storage maybe. Plenty of USB devices require a certain version to work e.g. you won't be able to attach a USB 4 dock to a USB 1.1 host device, and expect anything to work.

Not even pci-e storage is fully compatible. Some drives pull a little too much power.

 

[jigger]

The use of a OEM modified PCIe connectors are no different to the mining days where our risers used a non standard USB connector to provide the required power and data lines, use those cables in a standard USB slot and you will potentially have a bad day. Some OEM’s have chosen to use non standard PCIe slots to expose PCIe lanes for various reasons, HP, Dell and ASUS server class kit spring to mind (eg that’s what I tend to encounter), I am sure other OEM’s will have done similar, but the clue with a non standard interface is the non standard part, it kind of gives the game away.

Other general exceptions to the PCIe standard do exist, the SMBUS pins for example are normally not used on consumer grade boards, so they aren’t connected by many OEM’s, but over the years a number of Dell (Broadcom) fiber cards, the Sun F20 NAND accelerators, early HBA’s (Broadcom) and even Nvidia 50 series GPU’s have had a hissy fit when it finds they are connected, but those are widely known and documented, as is the fix: a tiny bit of Kapton tape over pins 5 & 6.

Then let’s look at BIOS implementation, most consumer OEM’s didn’t bother to enable rebar despite it being part of the PCIe standard for many years. Then AMD couldn’t stop saying it so often that OEM’s enabled it for the most part. We’ve also seen various BIOS updates to improve timing accuracy with NVMe, again less to do with ‘the standard’ and again more to do with the OEM’s producing hardware for it needing the goal posts moved slightly.

So yes, any PCIe compliant card is backwards compatible with earlier standards of PCIe, that’s one of the basic principals of PCIe standard. No you can’t fit a bespoke OEM custom PCIe card into a PCIe slot without a suitable adapter (Ali-Exlress sells some weird custom to standard adapters), and per the examples provided, some OEM’s can and do interpret the standard with a certain level of latitude that sometimes requires intervention, though often not by the OEM who caused the issue (motherboard manufacturers are often required to push BIOS fixes for other OEM’s stupidity).

The PCI-E standards aren’t really back compatible with each other. You can run into some issues on occasion.

 

[Avalon]

The PCI-E standards aren’t really back compatible with each other. You can run into some issues on occasion.

You've made the same vague statement twice now, i'm happy to admit i'm wrong in the face of facts, but you're going to have to at least provide some examples that I haven't taken the time to qualify, rather than just repeating your opinion as if saying it twice makes it true The power point is usually fixed via firmware update by the storage OEM limiting peak power, rather than the board OEM upping the limit beyond spec, that said it's not unheard of for a board OEM to get it wrong, and it still works, till it's in heavy IO workloads.

 

[aaronyuri]

Generally speaking: a PCI-E 4.0 NVMe will work just fine in a PCI-E 3.0 slot. There is a small chance that it won't for a myriad of potential factors, but it'll probably be fine.

 

[Avalon]

Well, if you're talking about storage maybe. Plenty of USB devices require a certain version to work e.g. you won't be able to attach a USB 4 dock to a USB 1.1 host device, and expect anything to work.

You'll notice I specifically said USB 3, that's because while USB 3.x has to be backwards compatible with USB 1 & 2, USB 4 only requires compatibility back as far as USB 2.0, because 4 is based on Thunderbolt 3, though just for fun, it doesn't have to be Thunderbolt compatible, though it can be.

 

[LuckyBenski]

You'll notice I specifically said USB 3, that's because while USB 3.x has to be backwards compatible with USB 1 & 2, USB 4 only requires compatibility back as far as USB 2.0, because 4 is based on Thunderbolt 3, though just for fun, it doesn't have to be Thunderbolt compatible, though it can be.

There are USB 3 devices that won't work on previous USB generations, but it's usually just down to data rate.

 

[Tetras]

but you're going to have to at least provide some examples that I haven't taken the time to qualify

Not sure if due to PCIE, BIOS issues or what not, but it comes up semi-regularly on the forums/other tech forums where cards for whatever reason don't work or have major issues.

The most recent example is 5000 series graphics in Z390/Z490 boards.

 

[Avalon]

Not sure if due to PCIE, BIOS issues or what not, but it comes up semi-regularly on the forums/other tech forums where cards for whatever reason don't work or have major issues.

The most recent example is 5000 series graphics in Z390/Z490 boards.

Some EVGA (and they aren't the only one) motherboards had the SMBUS pins connected, and 5000 series GPU's do indeed crap themselves when faced with this as I specifically mentioned. That appears to be a choice on Nvidia's part, almost as if the boards that usually have SMBUS active are usually server class, and they may not like the idea of consumer cards being used in that sort of environment, as opposed to the considerably more expensive Quadro's.

 

[Rroff]

Yes, fully backwards compatible
And not much drawback except lower peak bandwidth

While debatable how noticeable the difference is for a lot of uses the bandwidth difference between PCIe gen 3 and 4 NVMEs is quite substantial max sequential read/write on gen 3 is 3500MB/s compared to in excess of 7000MB/s for gen 4 devices.

 

[alec]

While debatable how noticeable the difference is for a lot of uses the bandwidth difference between PCIe gen 3 and 4 NVMEs is quite substantial max sequential read/write on gen 3 is 3500MB/s compared to in excess of 7000MB/s for gen 4 devices.

yeah, but the cases where max read/write is achieved are so rare to be basically non-existent outside of specific benchmarks

 

[aaronyuri]

yeah, but the cases where max read/write is achieved are so rare to be basically non-existent outside of specific benchmarks

Exactly... If you transfer a 7GB file the difference is 1 second. 1 second! I'm sure no one would notice that second in day-to-day use.