My anecdata: tl;dr: All 3 of my Samsung M.2 NVMe SSDs have failed in less than 3 years. 100% failure rate. My first SSD was a 1TB Samsung 970 EVO. It failed after 2 years and 8 months. It was replaced under warranty with a 1TB 970 EVO Plus. That replacement has now also failed after 1 year and 9 months. I bought a 2nd 1TB 970 EVO Plus in May 2019. It has now also failed (2 years and 7 months). Both are expected to be replaced under warranty. The 2 970 EVO Plus SSDs clearly had hardware errors (that were not accurately reflected in SMART data) that caused everything from system hangs, game crashes to file corruption on OTHER drives. I couldn't believe it at first but after 5 days of testing and trial and error, I had it confirmed. As soon as I removed those SSDs, my PC was completely stable again. In the meantime, I have bought a Kingston KC3000 1TB drive as I no longer trust Samsung M.2 NVMe SSDs. On the other hand, I have a Samsung EVO 850 SATA drive which has been rock-solid.
Neutral
I've been running the unsloth 200GB dynamic quantisation with 8k context on my 64GB Ryzen 7 5800G. CPU and iGPU utilization were super low, because it basically has to read the entire model from disk. (Looks like it needs ~40GB of actual memory that it cannot easily mmap from disk) With a Samsung 970 Evo Plus that gave me 2.5GB/s read speed. That came out at 0.15 tps Not bad for completely underspecced hardware. Given the model has only so few active parameters per token (~40B), it is likely that just being able to hold it in memory absolve the largest bottleneck. I guess with a single consumer PCIe4.0x16 graphics card you could get at most 1tps just because of the PCIe transfer speed? Maybe CPU processing can be faster simply because DDR transfer is faster than transfer to the graphics card.
Neutral
That is highly interesting and contrary to a number of reports I've read about the Samsung 970 EVO Plus Series (and experienced for myself) specifically! Can you share more details about your particular setup and methodology? (Specific model name/capacity, Firmware release, Kernel version, filesystem, mkfs and mount options, any relevant block layer funny business you are conciously setting would be of greatest interest.) Do you have more than one drive where this can happen?
Doesn’t like it
Yeah, it happens on two of the 970 EVO Plus models. One on the older revision, and one on the newer. (I think there are only two?) It happens on both Linux and Windows. Uhh, I'm not sure about the kernel versions. I don't remember what I had booted at the time. On Windows I've seen it happen as far back as 1607 and as recently as 21H2. I've also seen it happen on someone else's computer (laptop.) It's really easy to reproduce (at least for me?) and I'm pretty sure anyone can do it if they try to on purpose.
Doesn’t like it
The thing in the picture won't work at all if you plug it into a USB 3.0 port so that problem solves itself. I believe Thunderbolt ports are required to provide a minimum of 15W. But overall you might be right, there is not an end-to-end negotiation that would avoid a worst-case brownout of you had a very power-hungry SSD. The SSD in the article has a high-power active mode of 7.5W max (I am not sure why it says 3.3v*3.3A on the sticker[1]) and the TB controller in the enclosure claims 1.2W, so in this case it seems fine. It's possible to configure NVMe devices with power caps, but the 970 Evo in the article loses 95% of its performance if you cap it to 3.6W, which is its lowest operational power state. 1: Edited to say this is because mine is a 970 Evo Plus apparently the original really did draw that much current.
Neutral
(I was answering to the previous comment wondering if ramdisks still have any interesting usage nowadays.) About 150 TBW endurance on a 250 GB Samsung M.2 NVMe Evo 970+. On paper that is, but since it is the OS SSD with Windows Server 2022 STD (sole DC AD/DHCP/DNS/Hyper-V), in production, I won't take any risk. RAID 1 in this scenario would have changed nothing. On the side, I have 40 TB RAID10 for storage. So I cancelled the first C2 exec (with C2 encrypt on) when I reached 195 TBW on the SSD. Monitoring the ramdisk use still shows about 3:4 ratio on complete snapshot. I have about 1 million files on the 2,29 TB data to backup. I had indeed the RAM sticks available for free, simply had to take them (2x16 GB ) from a decomissioned ML350 Gen9 (which uses DDR4-2133P ECC Reg). It now serves me as a bench, litterally .
Likes it
I've bought 6-8 m.2 Samsung 970 EVO Plus and 980s since 2018, and none have failed to date. Anecdata is the worst, I'm sorry to hear about this happening to you. It's surely frustrating and upsetting.
Likes it
what I mean is, the typical consumer will probably never notice the performance difference between a 970 evo plus (or even 970 pro) versus any other middle-of-the-pack nvme drive. I personally went with the evo plus because I expect the drive to spend its life mostly full and I'm using it to compile c++ projects from. the extra $50 was worth the most consistent performance to me. as for reliability, this is hard to judge from anecdotes on the internet. you really need access to something like the backblaze data to start drawing conclusions. I had an oem samsung drive die in my xps 13 a week after the one year warranty expired. I have a WD hard drive that's still going strong after eight years. unless you have a lot of data, I think it's best to just back everything up, accept that you're rolling the dice, and hit a $/perf point that's acceptable to you.
Likes itDoesn’t say they own it
The models that never lost data: Samsung 970 EVO Pro 2TB and WD Red SN700 1TB Correction: “Plus” not “Pro”. Exact model and date codes: Samsung 970 Evo Plus: MZ-V7S2T0, 2021.10
WD Red: WDS100T1R0C-68BDK0, 04Sept2021 Update 2: models that lost writes:
SK Hynix Gold P31 2TB SHGP31-2000GM-2, FW 31060C20
Sabrent Rocket 512 (Phison PH-SBT-RKT-303 controller, no version or date codes listed)