Does having half the pcie bandwidth halves the speed of the drive, or cap it to the interface bandwidth?

[Valent117]

hello
Thats a question i had for some time, but never had the time to answer it
OK so let's you have a SSD pcie gen 4x4, rated (and tested) for 6GB/s (max of pcie4x4 being 8GB/s)
If you plug it into a gen3x4 slot, do you get 4GB/s (maximum of gen3x4), or do you get 3GB/s (speed of the drive divided by 2)
Thanks for answering this strange question

 

[lilhasselhoffer]

The interface will always saturate if allowed to.

This...doesn't mean 4 GB/s....more like (4-overhead-losses)...because nothing is perfect. Think of this less as stepped values, and more of overflow allowing saturation of a tunnel.

 

[Athlonite]

hello
Thats a question i had for some time, but never had the time to answer it
OK so let's you have a SSD pcie gen 4x4, rated (and tested) for 6GB/s (max of pcie4x4 being 8GB/s)
If you plug it into a gen3x4 slot, do you get 4GB/s (maximum of gen3x4), or do you get 3GB/s (speed of the drive divided by 2)
Thanks for answering this strange question

Basically yes it'll be capped to the lower bandwidth your SSD that is PCIe Gen4x4 will be slower on a gen 3x4 bus but you'd be hard pressed to feel that difference in speed unless constantly reading/writing large files when it comes to 4K random reads/writes even an SATA 6Gbps SSD will keep up and because 90% of what your pc does when reading/writing to an SSD whether it's PCIe or SATA is 4K reads/writes I wouldn't worry about it

 

[chrcoluk]

The drive will still go as fast as it can possibly go, so if it was exceeding the limits of gen 3 whilst using gen 4, then it will be at those limits (minus overheads) on gen 3. Things like random i/o which wont exceed the limits anyway should be unaffected and will be more affected by the power of the CPU in the system.

I ran my 980 pro in gen 3 initially, and there is no visible feels like impact since moving it to gen 4.

 

[massaker]

If you plug it into a gen3x4 slot, do you get 4GB/s (maximum of gen3x4),

This. But not exactly 4GB/s -> minus overhead. The maximum of PCIe3.0x4 is ~ 3740 MB/s (mostly achievable in sequential reads), 4.0x4 ~7480 MB/s, 5.0x4 ~14960, only with 6.0/7.0 will scale slightly worse due to different bit-coding).
As for much more important random I/O (e.g. 4K random) - slightly goes down by about 3-5%, nothing to worry about.

 

[lexluthermiester]

hello
Thats a question i had for some time, but never had the time to answer it
OK so let's you have a SSD pcie gen 4x4, rated (and tested) for 6GB/s (max of pcie4x4 being 8GB/s)
If you plug it into a gen3x4 slot, do you get 4GB/s (maximum of gen3x4), or do you get 3GB/s (speed of the drive divided by 2)
Thanks for answering this strange question

That depends on how fast your drive is. If your drive can saturate the PCIe bus, then yes. If it can't, then no. However, depending on how fast it is or isn't, you may not even notice.

 

[Wirko]

This. But not exactly 4GB/s -> minus overhead. The maximum of PCIe3.0x4 is ~ 3740 MB/s (mostly achievable in sequential reads), 4.0x4 ~7480 MB/s, 5.0x4 ~14960, only with 6.0/7.0 will scale slightly worse due to different bit-coding).
As for much more important random I/O (e.g. 4K random) - slightly goes down by about 3-5%, nothing to worry about.

It's exactly 4000 * 128 / 130 MB/s, or 3938 MB/s for PCIe 3.0. That's about 5% more than what you listed.

With these exact calculations, the difference between MB and MiB becomes significant. For data rates, decimal k/M/G is always assumed. No matter if PCIe, DRAM, USB, Ethernet speeds or anything else.

On the other hand, when talking about capacities and amounts of data - well, it depends. The 4KB block size is really 4096 bytes, not 4000. Therefore a 4 GB/s interface can't possibly transfer a million blocks per second.

I think you used MiB/s here, which would be incorrect. If you didn't, but you accounted for the PCIe protocol overhead in addition to the 128/130 bit encoding, the difference would only be around 1%, according to this data on Wikipedia. That still amounts to 3911 MB/s for PCIe 3.0. Is there any other significant overhead or latency for sending each small (4096-byte) block over PCIe bus?

 

[massaker]

It's exactly 4000 * 128 / 130 MB/s, or 3938 MB/s for PCIe 3.0. That's about 5% more than what you listed.

I know, 3938 MB/s is a "brutto throughput" and i spoke about "netto data rate". Not that OP thinks he'll get 4GB/s in real-life transfers. Nothing to do with MB-to-MiB difference, no.

If you didn't, but you accounted for the PCIe protocol overhead in addition to the 128/130 bit encoding, the difference would only be around 1%, according to this data on Wikipedia. That still amounts to 3911 MB/s for PCIe 3.0

Just read your own source:

Like other high data rate serial interconnect systems, PCIe has a protocol and processing overhead due to the additional transfer robustness (CRC and acknowledgements). Long continuous unidirectional transfers (such as those typical in high-performance storage controllers) can approach >95% of PCIe's raw (lane) data rate.

So yes, in addition to the overhead for bit coding, there is further overhead, commands, addresses, checksums, etc. must be transmitted in order to be able to do anything with the data and therefore ~3740 MB/s is about maximum you can transfer over 4x PCIe3.0 lanes. Just test a good Gen4 (or even Gen5 one, though not necessary) SSD in a PCIe3.0-Slot and see by yourself.