I've had this with gen5 PCIe SSDs recently. My T710 is so fast it's hard to beli...

adgjlsfhk1 · 2025-12-04T15:56:38 1764863798

To me that reads 3x, not "almost 10x". The main differrence here is probably power. A desktop/server is happy to send 15W to the SSD and hundreds of watts to the CPU, while a laptop wants the SSD running in the ~1 watt range and the CPU in the 10s of watts range.

simlevesque · 2025-12-04T16:12:57 1764864777

There's over twice as much content in the first test. It's around 3.8gb/s vs 30gb/s if you divide both folder size and both du durations. That makes it 7.9 times faster and I'm comfortable calling this "almost 10 times".

ls65536 · 2025-12-04T16:32:42 1764865962

The total size isn't what matters in this case but rather the total number of files/directories that need to be traversed (and their file sizes summed).

simlevesque · 2025-12-04T17:46:17 1764870377

I responded here, it's essentially the same content: https://news.ycombinator.com/item?id=46150030

adgjlsfhk1 · 2025-12-04T16:28:41 1764865721

oops. I missed the size diff. that's a solid 8x. that's cool!

taneliv · 2025-12-04T16:39:54 1764866394

I believe you, but your benchmark is not very useful. I get this on two 5400rpm 3T HDDs in a mirror:

    $ time du -sh .
    935G    .
                                                                                                                          
    real    0m1.154s

Simply because there's less than 20 directories and the files are large.

simlevesque · 2025-12-04T17:14:13 1764868453

I should have been more clear: It's my http cache for my crawling jobs. Lots of files in many shapes.

My new setup: gen5 ssd in desktop:

    > time find . -type f | wc -l
    5645741
    ________________________
    Executed in    4.77 secs

My old setup, gen3 ssd in laptop:

    > time find . -type f | wc -l
    2944648
    ________________________
    Executed in   27.53 secs

Both are running pretty much non-stop, very slowly.