I assume I’m not the only one who has played the same games on different type of storage: commonly HDDs and SSDs, but I also set up two RAID 0 filesystems (one on two HDDs, one on two SSDs), and I even installed Deep Rock Galactic on RAM.

However, more often than not loading times have been too similar across storage media.

Personal experience (tl;dr):
  • Deep Rock Galactic is so small it easily fits in your system’s memory, so you probably won’t be surprised to read that in my tests its loading times have been the same between HDD, SSD and RAM; any big chunk of data on disk is cached by the OS after being loaded for the first time, and it’s not like ALL data needs to be read at once in the first place.
    Quantitatively: loading times range from 2s to 40s (~15s on average), presumably because world generation and netcode hijinks take most of the time.
    All of this makes sense to me so far.
  • On the opposite side, Project Zomboid greatly benefits from faster storage if you’re using lots of mods.
    I haven’t measured world loading times, because it takes much more time to load and unload mods (it’s a Java game) than reading a bunch of jpegs and some kilobyte-sized files deciding where to place them: the former process takes ~ twice the time to complete if the game’s installed on an HDD rather than a SSD, haven’t tested it on RAID0/SSD; it’s a somewhat CPU intensive process, but some mods are BIG - my game’s workshop directory weighs 24GB.
    All of this still makes sense to me so far.
  • Then there’s Baldur’s Gate 3: the game is so chonky and I play it so infrequently that I have to keep it on my RAID0/HDD filesystem for logistic reasons, but at some point I had it on an SSD; I haven’t timed loading screens, but they are very long and I barely noticed an improvement on the SSD.
  • Helldivers 2: same as above, but netcode hijinks make metrics less reliable; besides, considering all the spaghetti code in it, that game is more Italian than me.
  • BeamNG: same as BG3, but less chonky and currently on RAID0/SSD.

Some other games I’ve played on several media matched the usual “SSD faster” expectations, namely Satisfactory, X3, X4, Abiotic Factor (or as I like to call it, “Antibiotic Factory”), Halo:MCC.

I’m asking this mostly because I’m considering getting two SSDs dedicated to a RAID0 setup, as of now my RAID0/SSD filesystem is “only” 200GiB wide and it’s sharing its drives with the OS and other things, but since I’m not short on space it may or may not be worth the price to set up a reasonably large FS with fast I/O.
I also suspect that my game loading times may be limited by the fact that I’m running most of them on Linux via proton, if everyone’s experience contradicts mine then that’s probably why; in fact, I’m pretty sure VKDX shader compilation adds some CPU-bound time.

  • spit_evil_olive_tips@beehaw.org
    link
    fedilink
    arrow-up
    1
    arrow-down
    1
    ·
    8 hours ago

    short answer: buy NVMe. plug it directly into your motherboard, don’t use an enclosure. forget about wonky RAID0 crap.

    longer answer:

    SATA SSDs (which you say in the comments below are all you’ve got) are an evolutionary dead-end. they’re SSDs pretending to be very fast hard drives. they end up being bottlenecked by the assumptions that the SATA protocol makes about how fast a hard drive can be.

    look at this chart for example. SATA (AHCI) limits a device to having 32 commands queued up at once, which means the operating system needs to jump through hoops in terms of maintaining its own queue of pending reads & writes and issuing them to the device as queue space becomes available.

    NVMe raises that limit to 64k, which for any non-server workload is effectively unlimited. the NVMe drive can respond to IO requests pretty much as quickly as the OS can dispatch them.

    if you want to know more nitty-gritty details, Scaling ZFS for NVMe is an interesting talk, much of it isn’t specific to ZFS, but instead is about how NVMe devices are so fast that they’re forcing filesystem developers to rethink long-standing assumptions about drives being slow.