Thunderbolt 3 M.2 NVMe SSD Enclosures

[EzisAA]

Understood (and I assume that by "Pro M1" you mean your MBP has the M1 Pro CPU?). I only use BlackMagic since it does help provide a good apples-to-apples performance comparison even if it doesn't represent real life use. Perhaps your 2.6GB/s is due to the 2TB capacity or perhaps BlackMagic might give you something closer to what I got?

BlackMagic give you ZERO help!!! Because BlackMagic test DRAM cache memory. You can test SSD with BlackMagic diferent PCI 3/4 gen SSD with diferenc memory chips - QLC, TLC, MLC SLC and BlackMagic will show you +/- the same speed for all. But real life QLC can be the same slower as 2.5 inch HDD 5400 rpm.

Modern SSD consists of different types of memory chips. And the SSD controller initially lets you write to the fastest memory chips, also known as DRAM cache. That's why BlackMagic's speedtest is useless because it tests fast memory when it's available.

In essence, all these numbers written on the SSD package are the speed of the fast memory, but it can be from 1% to 100% of the full volume of the SSD (SATA SSD from Samsung EVO and Pro models 100% of volume, but NVMe SSD about ~10% of volume), depending on the model, and usually these volumes are not written anywhere, they can only be found out by actual testing - filling the fullness of the SSD.

Just as well BlackMagic can you show 3GB/s, but after 30GB on write, speed drop to 350MB/s, in the same time another SSD can you show 2.1GB/s and after write 200GB it can be the same 2.1GB/s .

And who you will chose, if need transfer hundreds gigabyts!? Beautiful numbers from BlackMagic speed test or real life test numbers.

Also answer of your misunderstanding - My MacBook Pro M1 (M1 not M1 Pro, but M1) - in BlackMagic 980 Pro show 2.8GB/s write and 2.75GB read - but in real life it is 2.6GB/s (1,6GB/s after DRAM cache)

 

[Mac-ra]

BlackMagic give you ZERO help!!! Because BlackMagic test DRAM cache memory. You can test SSD with BlackMagic diferent PCI 3/4 gen SSD with diferenc memory chips - QLC, TLC, MLC SLC and BlackMagic will show you +/- the same speed for all. But real life QLC can be the same slower as 2.5 inch HDD 5400 rpm.

Modern SSD consists of different types of memory chips. And the SSD controller initially lets you write to the fastest memory chips, also known as DRAM cache. That's why BlackMagic's speedtest is useless because it tests fast memory when it's available.

In essence, all these numbers written on the SSD package are the speed of the fast memory, but it can be from 1% to 100% of the full volume of the SSD (SATA SSD from Samsung EVO and Pro models 100% of volume, but NVMe SSD about ~10% of volume), depending on the model, and usually these volumes are not written anywhere, they can only be found out by actual testing - filling the fullness of the SSD.

Just as well BlackMagic can you show 3GB/s, but after 30GB on write, speed drop to 350MB/s, in the same time another SSD can you show 2.1GB/s and after write 200GB it can be the same 2.1GB/s .

And who you will chose, if need transfer hundreds gigabyts!? Beautiful numbers from BlackMagic speed test or real life test numbers.

Also answer of your misunderstanding - My MacBook Pro M1 (M1 not M1 Pro, but M1) - in BlackMagic 980 Pro show 2.8GB/s write and 2.75GB read - but in real life it is 2.6GB/s (1,6GB/s after DRAM cache)

Interesting! I haven’t seen any BM or other tests for these TB enclosures showing anything close to 2.8GB/s for the M1 MBP 13 (based on the results I’ve seen for the M1 Mac Air since that’s typically what’s reported vs. the M1 MBP 13). Not sure why/how your M1’s TB ports are substantially out-performing those on my M1 Pro Mac…perhaps because I formatted my SN770 to exFAT?…anyway….

I agree about the importance of testing sustained speed for much-larger files than BM’s 5GB max. However I am not a “pro” or gaming SSD user (just a geek that wants to get the most out of his new MBP). So I don’t do much of my own SSD testing (actually some of my 1st experiences with doing any SSD testing was for my post above). Instead I‘ve been mostly using Tomshardware.com reviews to get a sense of how SSD’s perform beyond BM. Their review of the SN770 showed that the 500GB SN770 (which I’m extrapolating from their test of the 1TB) should write ~140GB @ 4.9GB/s (not via TB of course) and then degrade to 550 MB/s, while the 980 Pro 500GB writes 100GB @ 4.5GB/s and then degrades to 1GB/s. Since I rarely deal w/ files over (or even close to) 100GB the SN770 should work fine for me and hence why I chose not to spend the extra $$ for the 980 Pro.

 

[masotime]

AmorphousDiskMark might make for a better benchmark - https://apps.apple.com/us/app/amorphousdiskmark/id1168254295?mt=12

I believe you can adjust the size of the test files.

 

[Mac-ra]

AmorphousDiskMark might make for a better benchmark - https://apps.apple.com/us/app/amorphousdiskmark/id1168254295?mt=12

I believe you can adjust the size of the test files.

Will check it out esp. since I need to learn about how queue depth affects actual performance. Thx!

 

[Chancha]

Ever since ATTO Disk Benchmark became available in the App Store, I think this replaces even AmorphousDiskMark since it gives results in graphs and has quite a few adjustable parameters.

Anyway BlackMagic has its place, it gives a simple number for comparison which everyone can understand. It is like when 99% of router or network tests use outbound SpeedTest instead of something local like an iPerf3 server installed within your LAN, the results are almost absolutely useless but that's what the average folk can understand.

 

[Mac-ra]

OK so now I have two additional tests to read about and then use on my new Acasis & SN770 external SSD…Thx!

 

[Chancha]

I just got an ACASIS AC-TBU405 enclosure and WD SN850X 4TB for Black Friday and they just arrived. The enclosure is seemingly better built than the other Chinese stuff I am used to like JEYI or WAVLINK. Heat dispersion feels better also as the surface only reaches 30-ish degrees instead of the usual 40-ish on others.

Haven't got to use it extensively but a short test with AmorphousDiskMark gives me this, which is about the best I can expect from a Thunderbolt port (to a base Mac Studio):

 

[Mac-ra]

Wow - not sure I’ve seen TB enclosure test results for the 850x til now (or for the 990 Pro either). That’s definitely at the max (a marginal pun) that TB3/4 can support…very cool!

 

[Chancha]

The SN850X is massively overkill for Thunderbolt, but I got it anyways as it went on sale on Amazon Black Friday for $375 which isn't that far off from the other 4TB gen3 sticks (I totally need a single 4TB volume on the road now). And I can use it for many years down the road, probably when TB5 happens or if the Apple Silicon Mac Pro has PCIe slots that you can add gen4 NVMe cards.

And there is a bonus feature that I didn't consider when buying the ACASIS TBU405 but turned out to be really useful: it is a dual TB4 / USB4 board. Meaning you can plug it to older Macs without type-C, or other Windows computers with type-C but not Thunderbolt, they all work, as long as the file system on disk is supported on the computer. For instance, if you plug any previous TB3 NVMe enclosures directly to a 2015 MBP or any Thunderbolt 2 Macs it just wouldn't work, because the Apple TB3 to TB2 dongle cannot power the enclosure that is bus-power required (it can work with a powered Thunderbolt dock in the mix but that's too much hassle for a portable drive). But with this ACASIS it works as is, I just tested in on a Mac mini 2012 and got 400-500MB/s, so typical USB3.0 speed, and the same APFS volume works without tricks.

 

[EzisAA]

The SN850X is massively overkill for Thunderbolt, but I got it anyways as it went on sale on Amazon Black Friday for $375 which isn't that far off from the other 4TB gen3 sticks (I totally need a single 4TB volume on the road now). And I can use it for many years down the road, probably when TB5 happens or if the Apple Silicon Mac Pro has PCIe slots that you can add gen4 NVMe cards.

And there is a bonus feature that I didn't consider when buying the ACASIS TBU405 but turned out to be really useful: it is a dual TB4 / USB4 board. Meaning you can plug it to older Macs without type-C, or other Windows computers with type-C but not Thunderbolt, they all work, as long as the file system on disk is supported on the computer. For instance, if you plug any previous TB3 NVMe enclosures directly to a 2015 MBP or any Thunderbolt 2 Macs it just wouldn't work, because the Apple TB3 to TB2 dongle cannot power the enclosure that is bus-power required (it can work with a powered Thunderbolt dock in the mix but that's too much hassle for a portable drive). But with this ACASIS it works as is, I just tested in on a Mac mini 2012 and got 400-500MB/s, so typical USB3.0 speed, and the same APFS volume works without tricks.

Acasis TBU401E also work in USB mode up to 10Gb/s

 

[Mac-ra]

The SN850X is massively overkill for Thunderbolt, but I got it anyways as it went on sale on Amazon Black Friday for $375 which isn't that far off from the other 4TB gen3 sticks (I totally need a single 4TB volume on the road now). And I can use it for many years down the road, probably when TB5 happens or if the Apple Silicon Mac Pro has PCIe slots that you can add gen4 NVMe cards.

And there is a bonus feature that I didn't consider when buying the ACASIS TBU405 but turned out to be really useful: it is a dual TB4 / USB4 board. Meaning you can plug it to older Macs without type-C, or other Windows computers with type-C but not Thunderbolt, they all work, as long as the file system on disk is supported on the computer. For instance, if you plug any previous TB3 NVMe enclosures directly to a 2015 MBP or any Thunderbolt 2 Macs it just wouldn't work, because the Apple TB3 to TB2 dongle cannot power the enclosure that is bus-power required (it can work with a powered Thunderbolt dock in the mix but that's too much hassle for a portable drive). But with this ACASIS it works as is, I just tested in on a Mac mini 2012 and got 400-500MB/s, so typical USB3.0 speed, and the same APFS volume works without tricks.

Yeah, I was thinking “overkill” too since the not-too-long-ago top-end drives like the SN850 & 980 Pro already max out what TB3/4 is capable of but your reasons make sense. And yeah, for most of us it currently makes no practical sense to have a TB3/4 enclosure that doesn’t also connect via all versions of USB2/3 (regardless of whether the ports involved are physically TB2, USB-A or USB-C)...hence why the Samsung X5 seemed destined for a few owners and then oblivion? These current enclosures are great products!

 

[Mac-ra]

Mistaken duplicate post that I can't figure out how to delete!?

 

[jazztastic]

For Christmas, I got an Orico 20 Gbps 3.2/3.1 Gen 2X2 Enclosure (M223) and a SAMSUNG 980 PRO SSD 1TB PCIe 4.0 NVMe Gen 4 drive.
On my Mac M1 Mini 2020 8/256 and I am using the enclosure supplied USB-C to C cable to connect to a Thunderbolt 3/USB 4 port. However, speeds are barely approaching 10 Gbps in tests.

What am I missing here?

 

[joevt]

For Christmas, I got an Orico 20 Gbps 3.2/3.1 Gen 2X2 Enclosure (M223) and a SAMSUNG 980 PRO SSD 1TB PCIe 4.0 NVMe Gen 4 drive.
On my Mac M1 Mini 2020 8/256 and I am using the enclosure supplied USB-C to C cable to connect to a Thunderbolt 3/USB 4 port. However, speeds are barely approaching 10 Gbps in tests.

What am I missing here?

Apple Silicon Macs have slower USB ports (940 MB/s) than Intel Macs (1060 MB/s). You can try connecting it to a Thunderbolt 3 dock but I don't think you'll notice the extra 100 MB/s which is only achievable with sequential reads/writes.
https://forums.macrumors.com/thread...ally-10gb-s-also-definitely-not-usb4.2269777/

Macs don't have USB-C ports that support USB 3.2 gen 2x2 (20 Gbps).

macOS doesn't support USB 3.2 gen 2x2 even if you got a USB 3.2 gen 2x2 PCIe card (though I haven't checked the latest Ventura release yet).

 

[EzisAA]

For Christmas, I got an Orico 20 Gbps 3.2/3.1 Gen 2X2 Enclosure (M223) and a SAMSUNG 980 PRO SSD 1TB PCIe 4.0 NVMe Gen 4 drive.
On my Mac M1 Mini 2020 8/256 and I am using the enclosure supplied USB-C to C cable to connect to a Thunderbolt 3/USB 4 port. However, speeds are barely approaching 10 Gbps in tests.

What am I missing here?


View attachment 2132774

You need Thunderbolt 3 or 4 or USB4 enclosure, them you will get up 2,6GB/s speed.
Or wait new Mac's with Thunderbolt 5 , maybe them will support USB 3.2 2x2 (up 20Gb/s)

 

[Mac-ra]

You need Thunderbolt 3 or 4 or USB4 enclosure, them you will get up 2,6GB/s speed.
Or wait new Mac's with Thunderbolt 5 , maybe them will support USB 3.2 2x2 (up 20Gb/s)

Yeah, there's plenty of tests on this thread showing the 2x-3x faster speeds you'll see from the 980 Pro if you use it in a TB3 enclosure like the Acasis TBU401/405 or Orico M2V01 (though there's plenty of other choices on Amazon which all seem to use the same TB3 & USB3 Gen. 2 "bridge" chips...search on "40gbps nvme enclosure for thunderbolt").

 

[Mac-ra]

I finally downloaded AmorphousDiskMark and ran it on my Acasis TBU401 w/ an SN770:

The sequential tests are close enough to the BM results (~2.3/2.7GBps R/W) that I can't get too excited about that. However the random R/W tests look pretty marginal though I don't have a frame of reference for judging this. Either way this continues to confirm the expected ~400MBps reduced peak write speeds when connected to an M1 Pro vs. Max CPU...I guess?

 

[Mac-ra]

...and then I ran ATTO's test on my SN770/Acacis combo:

Other than the expected slower speeds at smaller files sizes I'm not sure what this is telling me vs. BlackMagic. The max write speeds are ~400MB/s faster vs. BM but ATTO shows this similar "boost" in post #726. So I'm still concluding that the M1 Pro CPU is the weak link in the TB write speed vs. the M1 Max, though maybe there's other conclusions to draw from this that I don't have the ability to see. Either way these speeds are ~3x of what I can get from a Sandisk Extreme or Extreme Pro so I can't complain about the smoking performance of this $155 DIY external SSD!

 

[micheloupatrick]

Another combo : ACASIS TBU405 + FireCuda 530 4TB on a MBP 16" M1 Max :

The numbers are similar to those observed by Chancha with the same enclosure and the WD SN850X 4TB.

In a real world test, copying a 470 GB folder (containing photoshop files) to the MBP internal drive took 2 minutes 40 seconds without any slowdown. The enclosure got pretty hot though.

 

[micheloupatrick]

Same config as above, but this time with the Orico M2V01 enclosure (SSD and MBP are the same) :

So both enclosures perform very similarly (and very well).

 

[mackiemesser2]

@micheloupatrick
@Chancha

Does anybody of you 4TB guys have a chance to test if these drives can be formatted with a single 4TB (NTFS) GPT partition?

I read somehwere that most NVMe ensclosures ony support MBR (partition 2TB max.) not GPT.

What are your max.temps?

 

[MediaGary]

I had two 4TB Western Digital SN700 drives in a Sabrent EC-T3DN enclosure. The enclosure was uncomfortably hot to the touch, but my digital thermometer was lost at the time so, I don't have specific values. I added an external 40mm x 100mm x 11mm heat sink and a USB-powered fan to keep things in check.

I think that many enclosures only support single-sided NVMe M.2 drives, and therefore, that's the source of the 2TB limit; they don't have a way to wick away heat from the bottom of double-sided drives, which seems to be the typical packaging of 4TB M.2's.

Coincidentally, MBR's 32-bit counting limit for 512-byte sectors tops out at ~2.2TB, so that limit can get conflated with the hardware packaging limit.

I won't be doing any more measurements on the Sabrent dual NVMe M.2 enclosure because those WD SN700's are now happily on my Sonnet 4x4 PCIe x8 card.

 

[Low Shrimp]

For integrated Thunderbolt controllers with two ports (Ice Lake and Tiger Lake) it is known that the upstream of the controller is greater then PCIe gen 3 x4.

Found this on AnandTech. It appears that the upstream for integrated controller is still PCIe 3.0 x4:

Here’s a more detailed schematic, showing the complexities of adding in TB3 into a chip, with the four PCIe x4 complexes shown moving out to each of the individual PHYs at the bottom, and connected back into the main SoC interconnect fabric.

@joevt, are there any tests that show higher speeds that would be possible with PCIe 3.0 x4?

 

[joevt]

Found this on AnandTech. It appears that the upstream for integrated controller is still PCIe 3.0 x4:

It's definitely not PCIe 3.0 x4. I think the integrated Thunderbolt controllers actually report themselves as PCIe 1.0 x4 even they they transmit more than 1000 MB/s. Use pcitree.sh to test that. g1 = PCIe 1.0 = PCIe gen 1 = 2.5 GT/s. g2 = 5 GT/s. g3 = 8 GT/s.

@joevt, are there any tests that show higher speeds that would be possible with PCIe 3.0 x4?

Create a software raid in Disk Utility.app with two partitions of two different NVMe devices and use AmorphousDiskMark.app to test sequential read.
An easier test is to use two separate NVMe with the ATTO Disk Benchmark.app since that can test multiple disks at the same time. The ATTO Disk Benchmark in Windows doesn't seem to have that feature.

 

[Low Shrimp]

An easier test is to use two separate NVMe

Thank you! I'm going to assume we are talking about NVMe devices connected via TB3. It seems testing multiple NVMe devices connected to a single external TB3 controller would be limited by that controller before they hit any limits of the integrated TB3 controller. But how do I connect multiple external TB3 controllers to a single TB3 port of the integrated TB3 controller?