9

The 256GB and 512GB Steam Decks are advertised with faster storage (NVMe SSD) than the 64GB version's eMMC internal storage. But I read that the eMMC storage read times are more or less comparable with a decent SD card. Under what circumstances is the NVMe SSD noticeably faster than the eMMC or SD card options? If there have been benchmarks for this, they're buried under a lot of other press.

1
  • Typical eMMC's Read/Write speeds is 330MB/s and 200MB/s, and recent M.2 is 3,500MB-7,500MB/s and 3,000MB-7,000MB/s (depends on PCIe generation), so at least 10 times faster at all circumstances. I don't know about Steam deck's performance though.
    – Skye-AT
    Jan 16 at 2:47

1 Answer 1

3
+25

The specs describe the 256GB and 512GB models as using NVMe with either PCIe Gen 3 x4 or PCIe Gen 3 x2.

Maximum theoretical bandwidth for x2 is 1.969 GB/s and x4 is 3.938 GB/s (taken from this chart) and real world max performance will be less than that, varying by model, and then actual observed performance will be less than that, varying by workload.

According to Valve, the difference between x2 and x4's bandwidth only manifests in edge cases.

"SSD performance is currently gated by factors not related to PCIe bandwidth," says Yang. "In extremely uncommon cases, differences in read/write speed caps may minimally impact file transfer speeds, but OS performance, loading times, game performance, and game responsiveness are identical between the x2 and x4 drives."

Loading a lot of random files into memory and then crunching on those files will often present bottlenecks far lower than theoretical limits. Pretty much the circumstance where you'll see some real noticeable differences is when copying large files.

eMMC caps out at 330MB/s read and 200MB/s write. This seems a lot slower, but game workloads typically bottleneck much lower than NVMe speeds.

This order of magnitude of difference seems like it should matter, but for a lot of gaming workloads, it really doesn't, as the bottlenecks are elsewhere or the types of reading and writing are similarly slow on PCIe vs eMMC vs SD card.

At least one test has shown eMMC and SD card actually beating the order-of-magnitude faster SSD in game load times slightly, and others appear to show negligible differences or the SSD slightly winning.

Even the differences between speeds of compatible UHS-I SD cards don't appear to make that much difference to load times, often sub-second variations.

The write speed is more likely to make a difference, particularly in game downloads. UHS-I SD cards will have a U rating, with U1 being a guarantee of 10MB/s minimum sustained write performance, and U3 being a guarantee of 30MB/s. When writing a download to a card from say a 300+Mbps internet connection, that could make a big difference. (I say could, because many other factors could slow down the throughput and make the difference irrelevant.)

Note that some games stream assets from disk during gameplay, and this is a place where hypothetically disk read performance could matter, but this is a much less objectively visible operation than a game load, and so it's difficult to quantify how much it could help. It's more likely to matter with extremely high resolution assets intended for consumption on a high end system with a GPU that costs more than a Steam Deck.

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .