Raspberry Pi SSD #679
Comments
|
On the site here: https://pipci.jeffgeerling.com/cards_m2/raspberry-pi-ssd.html |
|
lspci output: It looks like it's a Samsung NVMe controller, and here are more details: |
|
Benchmarking... here's a listing on harddrivebenchmark.com for the Samsung. And on the Pi 5, running my disk-benchmark.sh:
Also running PiBenchmarks.com: |
|
Hey Jeff, I was just curious if this supported 4K Native Advanced Format. You can check using either For example my WD Black (which does) has this: |
|
hi @Steve-Tech, Have some RPI SSD's myself and it does not like it support 4k blocks as shown below: |
|
Hi @leander091, alright, thanks for checking! |
|
Here is a newly purchased Raspberry Pi 512Gb SSD. It no longer claims to be a Samsung, but is rather is a Biwin SSD. It was purchased from a big brand well known very respectable supplier, and if it is a fake, it's an extremely good one. The packaging and label all look perfect. I wouldn't personally associate Biwin SSDs nor its "KingSpec" controller with quality like I would with Samsung, but if this is genuine I assume the RaspPi Foundation know what they're doing. I'll tell you this: power consumption is remarkably low with this SSD. Like, crazy low. PCIe 2.0: Write load is 0.5 watts. Wow. Read load is 0.7 watts. PCIe 2.0 throughput performance is very competitive. PCIe 3.0: Write load is 0.9 watts. Read load is 0.8 watts. PCIe 3.0 throughput performance is nothing to be sniffed at. DRAMful SSDs have a bit faster reads. Other details: The claimed power use states do seem to very much correlate with empirically measured power consumption. More info about the KingSpec NX SSD controller can be found at https://www.techpowerup.com/ssd-specs/kingspec-nx-series-1-tb.d1082 |
|
That is a bit less than the Samsung based SSD. In case it was a fake and I've been diddlied, I tried filling the drive to full: No i/o errors. Looks true sized. Obviously it ran out of SLC cache, that made it slow down to about one quarter peak write speed. I noticed that power consumption went up to 2.7 watts too, so running out of SLC cache does increase write power consumption considerably. That also agrees that the entire drive was filled. I see from searching the web that the Biwin-made drives appear to have become the default OEM for official RaspPi SSDs from Nov-Dec last year onwards. Multiple people report the same. I guess it's legit so. |
|
I've discovered something odd about the official Raspberry Pi NVMe SSD made by Biwin: it appears that TRIM has no effect despite that the SSD says it supports TRIM: Linux thinks it is supported. Apparently all space is allocated (it is not, 1.52Gb out of 457Gb is allocated). Manually force a trim: It takes about a minute, so it's doing something. Absolutely no change. Lets make sure it's not ZFS: Nope, My Pi with the Samsung SM961 MZ-VPW2560 trims just fine on ZFS, so I think it really is the official Raspberry Pi SSD here. Obviously ignoring TRIM is a big problem for longevity, never mind burst write performance. Jeff, I don't suppose you could do a blog post? I suspect given your reach it would achieve swift action. I have reported the issue to raspberrypi/linux#6627. |
|
@ned14 - Before raising the visibility further, I'd like to hear back in the pi/linux repo, see if it might be something overlooked in the kernel, or something that could be fixed elsewhere. Not supporting TRIM is certainly not a wonderful 'feature' of the new flash controller... The official Pi page for the SSD (and the Datasheet) don't mention TRIM support at all, but it'd be a bit of a cop-out to say that means they don't have to support it—because the initial batch of Samsung drives did... and online databases like This Techpowerup page for the Raspberry Pi SSD also include that in the spec (because that's what everyone had access to when the disks were first introduced). So I would definitely like to 'raise a little hell', so to speak, if people who buy newer Pi SSDs are getting an objectively worse piece of hardware than the original that I tested. If the performance were a couple percentage points different, or thermals were slightly different, that's one thing. But TRIM is a pretty essential feature for flash! I even wrote a blog post on it for Pi back in 2020! |
|
That's fair. My own personal suspicion is that the SSD's firmware needs an update. The fact no errors are reported either to the kernel or in the NVMe logs makes me think it's a firmware bug whereby it isn't implementing the request. I can tell you when first powered up, So I think the SSD understands just fine how much data has been written to it. This is supposed to be working. I didn't try playing with hdparm or sending TRIM requests by hand (though I did try I tried twiddling the This particular model (CE430T5D100 aka "Biwin AP425") is not a common one. Only a few laptops have shipped with it, and they seem to mainly promote it as an industrial SSD not a consumer one. Its product page is at https://www.biwin.com.cn/product/detail/95. No firmware update package is available. |
Raspberry Pi is now selling the Raspberry Pi SSD, a 2230-size NVMe SSD rated at PCIe Gen 3 speed.
I have a 256 GB model to test, and they are also releasing a 512 GB model. The product data sheet doesn't list certain specs like MTBF, TBW, shock ratings, or how many lanes of PCIe Gen 3 are supported (assuming you put it on a device other than a Pi 5).
I would like very much to find those things out, so once I get it hooked up, I'll post the full
lspcioutput here.The text was updated successfully, but these errors were encountered: