We are already seeing SSDs with over double the read speed and ~70% more write speed. You don't see how that could make a difference? Even with 128GB of RAM on a workstation there are a lot of data sets that do not fit.

For training neural networks faster? I doubt SSD->RAM speeds are a bottleneck there. Are you reading training data faster than 6GB/s during training?

You don't even need to get to that complex of a scenario. Just running simple database queries would benefit from this.

Sure, I can see it being useful for servers, just not for 99% of home users.

moving from spinning rust to SSD was a huge improvement in day to day life. i still see lag (on my very fast machines) opening programs or loading data on my laptop that would be highly improved by doubling the speed of the hard drive. and i’m not even counting compiling or things that we might do that is different than other users.

That's because SSDs had way lower latency. Upping the pcie gen will speed up sequential reads which were already really fast, but won't speed up random reads (or reduce latency) at all.

It could allow for more parallelism (deeper queue depth) with those reads though

Right but nvme are at the level of speed where you need to optimize apps for it else they get CPU-bound. Difference between mid and high-end NVMe for most user stuff ought to be minuscule

Full-access database queries? Indexed queries are going to randomly read blocks and PCIe 5 doesn't help. The only way I've perceived improvements from SSD linear access > 4GB/s is when doing gigantic ETL jobs that were not CPU-limited. Unfortunately a lot of ETL jobs are CPU limited and some of the formats I work with are too intensive on the compute side to benefit from faster storage (like GeoJSON, ugh). Formats that are more CPU-friendly also tend to be smaller, making the storage less relevant.

I think for general use most desktop and workstation users are going to get more benefits from faster random access and won't notice very high linear access speeds. I have two recent SSDs and one of them has a median access time of 12µs and the other has a median of 30µs. Even though the latter has gaudy benchmark numbers, can stream at many GB/s and can ultimately hit almost a million IOPS, higher-level application benchmarks lead me to prefer the former because random access waiting time is more important.

I was mainly talking about my usage patterns, not every desktop or workstation user. I work with a lot of data daily.

Something more general might be game asset load speed. Those are often sequential reads. They put a ton of engineering effort into the latest consoles simply to improve that one thing.

> They put a ton of engineering effort into the latest consoles simply to improve that one thing.

Most of that effort was to ensure the processors could actually ingest data at the speeds that off the shelf SSDs could deliver it. The Xbox Series X shipped with what was a low-end NVMe SSD at the time, and the PS5 used a custom SSD controller primarily so they could hit their performance targets using older, slower flash memory rather than being constrained by the supply of the faster flash that was just reaching the market at that time.

And there are still hardly any games that even make a serious attempt to use the available storage performance.

PCIe 5 isn't what helps here. it's that nvme/ssd use very few lanes of PCIe (usually 2/4). they could have achieved these speeds already, but now it's just easier with a smaller slot.

Not only not fitting in RAM but also initial loads. Loading times could be reduced even further.