Would it be nice if one could transfer large amounts of data at a fast transfer rate when working with servers and NAS-storage. Well that is what I thought before I failed miserably.
We have used external NVMe to USB-C 3.2 Gen2 enclosures before for random SSD-drives and have had reasonable experience with them. Usually the use has been read heavy when doing data recovery from a dead laptop and transferring files over to a new machine.
This time we thought that surely a speedy 4TB NVMe drive would be great for transferring data over from a server or from a NAS.
Seems that modern SSD-drives have firmware that is basically not compatible with the enclosures when doing write heavy tasks.
What we found out was that depending on the model of the drive it would operate for a short time at the speeds we would expect (1000MB/s) and then the speed would drop to anywhere from <1MB/s to 80MB/s dependind on the drive. Usually the drive would not recover from this low performance mode even if the transfer was aborted and the drive left alone for a while.
The enclosures we tested:
Two different m.2 m-key to USB-C 3.2 gen2 enclosures capable of 1GB/s sustained transfers - Bought from Ali Express - Tested with various cables including 10cm USB-C to USB-C and 30cm Thunderbolt3 certified oem-docking station cables
The enclosures were based on Micron and Realtek chipsets
The drives tested:
4TB KingstonKingston FURY Renegade SSD PCIe 4.0 NVMe with 1GB/s sustained write capability according to Toms Hardware - Worked well for the first 15-30 seconds and then dropped to around 1-2MB/s
4TB Crucial P3 Plus PCIe 4.0 (NVMe) - This drive actually worked fine until the write cache filled up at around 20-40GB written and the speed dropped to 80MB/s
When testing we ruled out overheating as an issue by applying extra heatsinks to the drives as well as the enclosure controllers. My guess is that this is simply a firmware issue where the new high performance and high capacity SSDdrives do not perform well in these kind of enclosures.
I’ve had the same idea, and found the problem to be heat. The enclosure basically insulated the drive, causing it to overheat quite quickly. It would work great for a few minutes and then never recover. And because it was being insulated, the casing didn’t get hot either, so there was little indication that would be the problem. If I opened the enclosure and ran it with the SSD just sticking out in the air, performance greatly improved.
If your enclosure is metal, try sticking a thick thermal pad between the SSD and the case. It may be enough heat transfer to let the drive work.
When testing with a Win PC I used Crystal disk info to read smart values from the drive and never got temps over 72C. Also we did try to run the case open and stuck a heatsink on the drive and the USB converter controller chip.
Most drives will report up to three temperatures, with one being roughly 10C higher. I would have to assume that one is the controller. If CrystalDisk is reporting the average or NAND temps, then your controller could be hitting 80C+ and that is definitely thermal throttle territory.
I would recommend using HWInfo64 as it will find every sensor it can.
How do you find the luck in what you are getting being good from AliExpress? I just find it hard to trust those websites when I look at the pricing compared to what I know the equivalent legit item would cost.
I saw a telescope on there once for $50, but being interested in astronomy I know an equivalent good one is hundreds to thousands more depending.
I just couldn’t not assume that the AliExpress one must be garbage.
You get what you pay for. The 50$ telescope is likely to be a children’s toy. But one for a few hundred dollars will probably be perfectly fine for taking photos of the moon or whatever it is one does with a telescope. I buy a fair deal of electronics on aliexpress, and for the most part they’re exactly as advertised. I wouldn’t use them in anything critical like a server, but that’s more for liability reasons than because I don’t think it would work. It’ll probably work just fine, but on the off chance it doesn’t there’s no good way to get the supplier to take responsibility, and even if you know a bit of Chinese it can be difficult to get them to reply.
Controller is the only one that matters, and I find a lot of disks only report flash temperature. If you’ve got a heat sink strapped to the drive the two are probably close enough, but without that there’ll be very little heat transfer between the flash and the controller and the reported heat value is basically useless. PCIe4 and up really do need good cooling, it’s not an optional thing any more. 99% that’s the problem even if the drive itself reports decent temperatures.
This silverstone enclosure has an asmedia chipset and seems good. I have been eyeing it but have not purchased yet (waiting to upgrade ssd in my nas first):
Apart from the chipset itself or heat, perhaps some of these have buggy firmwares that can’t deal with 4TB drives?
Also, paradoxically, it is actually a good sign if the case gets warm or even hot. If it stays cool, it probably does not have contact to the drive, and the drive is cooking inside with nowhere to dump the heat. The air inside will be a quite good insulator!
And even if the drive is cool, on a badly designed enclosure the usb controller may be overheating!
It is true that there is a large variance in the quality of products sold on Ali and similar sites. End of the day, most tech comes from china and most new tech is able to be bought from there a significant time before it is available anywhere else. These usb-c NVME enclosures are a good example. They have been available for years and are now beginning to be sold widely in western outlets.
While continuing the testing I found out that Additional cooling and larger heatsinks actually do make a difference. The ones I tried earlier did not do the trick so now I tried with some larger ones and the drive is managing to keep cool and function for the whole 200GB test transfer at expected speeds reaching 53C at the end.
As a confusing side note it seems that the machine I used for testing has files that simply cannot be read at a fast datarate.
Tested the transfers with large 12GB image files that I had in my downloads folder that were aroung 6 months old. The files could be read at a very slow 20-70MB/s speed from an internal 4TB Gen 4 SSD with trim enabled. Other files would read at 2-4GB/s data rates no problem but specific downloaded images were extremely slow for some reason.
After doing research it seems that stale data on SSD-drives has a tendency of becoming slow and that is actually a thing on SSD-drives. IT seems the rabbit hole is deep and consumer SSD’s are in fact not that great. Who would had thought…
Its not firmware (except rare cases), its due TLC/QLC drive overall design combined with poor thermals, which compound the issue further.
If you were to redo your testing slotted in onboards slots, I would bet they would behave similarly. The final steady state would likely be better performing, but there would be massive perf drop anyway.
Good rule of thumb is - advertised speeds of consumer drives are attainable only:
for burst workload with idle preiods following
empty drives
There simply isnt enough overprovisioning, slc cache and thermal headroom for any other result.
I’m curious as to how long it took to transfer the complete 200 GB after your cooling fix was done. Not surprised to learn that Micron (Crucial) gave you better performance than the Kingston NVMe. Cache size is everything when it comes to NVMe and methinks the boys @ Samsung know this too. I’m also thinking perhaps this is one more reason why Intel dropped Optane? Just a thought.
Yup. I knew it was a bizznizz decision and there’s no sense in operating a company if it can’t make a profit. That aside, Intel seems to be dropping the ball on a lot of things in these past couple of years. If I’m not mistaken they dropped support for IRST as well as Optane. Turbo Boost Technology is no longer a thing either. I’m beginning to wonder if a good deal of all this EOL has to do with the TPM Consortium. I’m keeping my tin foil hat handy but I don’t really need it as anyone who has done their homework knows the Consortium is real. Anyway, thanks for the information. I appreciate it.
the 200GB transfer went at around 870MB/s in an external enclosure with extra cooling.
What comes to the Crucial performance it was extremely poor compared to the kingston drive. Crucial falls of to around 70-100MB/s after roughly 20-30 sec at 1GB/s and the kingston drive seems to be able to sustain 1GB/s speeds for at least the first 900GB.
You are lowering potential best case performance even lower.
1-2 MB/s is suspiciously low, it might be combined effect of insufficient cooling on both drive and pcie/usb chipset device.
I.e if enclosure chipset thermally throttles, then thats it. Either use forced cooling or accept it, not much else to with noname junk.
But if you could, try attaching thermal probe inside to check how hot does the bridge chipset get before performance drop.
BTW: steady state performance and cache recovery was usually ignored drive spec during testing, but nowadays even lower quality reviewers like tomshardware publish at least something usable:
Thanks for sharing. Icydock rocks. I’ve purchased a number of their enclosures and never have any trouble with the bridges in those. I also appreciate their generous use of metal instead of plastic which gets brittle over time. I’m very pleased with my WD BLACK SN770 NVMe and this picture you just shared shows why. I’m going to snag that picture and send it to my buddy. Were at loggerheads on the SN770 vs the Sabrent Rocket. That’s too bad about the Crucial P5 Plus. I bought a stick just to check out. Now I know why I got it so cheap. Yeah, I own a few PCs. I can’t help myself.