Tong Family

Well, I ended up with a bunch of different drive technologies here are the results so here are the results using three different SSDs against three different SATA to USB 3.0 adapters. And also my new ZikeDrive Z666.

What we want to do is figure out what combination of old drives in the dustbin and SATA adapters work the best and there are some surprising results. Some adapters are 2-5x slower than others

TL;dr

Wow, although expensive at $120 for the enclosure and $250 for the 4TB drive, you definitely get a very fast SSD running at 3GBps. It’s still a long way off from the MacBook Pro internal drive running at 7GBps. But for more normal random reads and writes, it is super fast.

If you want more storage and want to more economical then a $25 Startech USB3S2SAT3CB with a cheap SATA SSD that costs say $100 will give you 520GB of additional storage. I wouldn’t use this for doing video streaming, but it is plenty fast for archiving after you are done editing on your main Mac. Even drives that are 10 years old like the Samsung 850 EVO run at 300MBps over a USB 3.0 connection. That’s not bad.

It turns out the choice of adapter makes a big difference, so I would recommend in performance order from best to worst:

1. RIITOP USB 3.1 TO SATA 5G to USB C with ASMT 1153 controller paired with the SanDisk Extreme Pro 480GB. this had a very good performance close the the Startech above but it doesn’t need an adapter for modern USB C devices.
2. RIITOP USB 3.1 to SATA 5G to USB C with ASMT 1153 and pair it with the Samsung Evo Pro 512GB
3. Startech USB3S2SAT3CB with the ASM225CM controller paired with the Samsung Evo Pro 1TB. This is a little hard to find but has excellent performance. It is USB B connected
4. Startech USB3S2SAT3CB with ASMedia ASM225CM with Toshiba OCZ Tri150 480GB. Slightly lower performance but you won’t really know the difference.

Not recommended. Sabrent EC-SC31. It has great specs, but for some reason really bad random I/O performance, so retire these parts.

Benchmarks for Disk

I’ve used a bunch of these benchmarks:

1. The go-to has been the Blackmagic Disk Speed Test, this has been the easiest to use and typically I just run at 5GB file size. The main problem with it is that it is really only testing sequential read and write, so while it is great for video edits and storage, there are many more aspects to it
2. Amorphous Disk. Inspired by CrystalDiskMark for Windows, this gives a much better view of performance, you can install it with Homebrew. For all this work we set the number of runs to 5 and the size read at 1GiB for no other reason than that this is the default on the home page of the app 🙂 This means that the disk is really saturated with 32 requests constantly there

SSD Adapters (SATA to USB 3 and NVMe to USB 4)

1. Startech USB3S2SAT3CB. $15 at Amazon. Underneath this generic SATA to USB Adapter is an As media ASM225CM controller which first came out in 2010. So it’s a big old, but it speaks USB 3.1 5Gbps and its performance is pretty interesting. The main limit is that it is running SATA at 3Gbps instead of 6GBps so the maximum is more like half that since the SATA interface is limited the throughput more than the USB 3.0 interface (3Gbps is 375MBps or thereabouts). Although USB 3.0 runs from 5Gbps to 625MBps, it’s a little lower because of error encoding, which is 8b/10b and gives you theoretical real data throughput of 500MBps. And with flow control, 400MBps is more likely. However, you will get lower performance depending on the controller in your computer. For instance, the Etron controller on some older Intel boards tops out at 250MBps.
2. RIITOP USB 3.1 to SATA 5G ASM1153U with USB C Connector. This is a no-name adapter I bought for and reports as an ASM1153U which a USB 3.2 Gen 1 to SATA Adapter. It supports only up to 3Gbps SATA though so USB C isn’t going to make much difference. I have two of them.
3. ZikeDrive 666. This is a completely different adapter and only worth it if you need really high speed copying for say video editing. it is an NVM 3.0 and costs $120 compared to $10-20 for the adapters above, but it is NVMe to USB 4.0. A combination of NVMe and the 40 Gbps USB 4.0 is much faster than USB 3.0. The promise 3.8GBps read and 3.1GBps write with a M.2 NVMe SSD 2280 drive. Benchmarks show that in practical sequential reads, it is doing 3.4GBps read and 3 GBps pretty amazing write. There are not many of these, but this one seemed pretty fast.
4. Sabrent SATA to USB 3.0 Adapter or maybe SATA to USB 3.1. I’m not sure what chipset this is using, but the results are below seem to show it is faster.

Because the SSDs themselves can have performance issues, we are going to test this against the in order of performance from fastest to slowest. These are the drives I have and the issue is that we want to maximize performance for this hardware:

1. MacBook Pro 2021 M1 Pro Max 2TB. This should be very fast and sort of scales what you can get. The Amorphous Bench shows it running 6.9BBps read and 7.7GBps write (the disk is 80% full) for sequential reads and 667MBps for random reads and 138MBps writes or 4KiB block. This is pretty common, doing random reads and writes is much harder and generates more I/O overhead.
2. Addlink S95 4TB. This is a different beast, it is it’s a 4TB NVMe Gen 4 so theoretically 2x the speed of NVMe Gen 3 at 7.2BGps . Note that the ZikeDrive is really NVMe Gen3 so performance will be lower. But what a deal at $285 for 4TB on Newegg. It uses the Phison E18 Gen4 controller set and a 96 Layer Micron 1200 MT Flash Array. The benchmarks show that it can reach 7GBps reada and 6.6 GBps write which is pretty crazy
3. Samsung 850 Evo 1TB. This is a 2014-era drive, so over 10 years now. The 1TB is different from the 500GB below, it uses a Samsung MEX controller and is capable of up to 540 MBps read/520 MBps write. The benchmarks show that the 850EVI 1TB should be very fast like 373MBps in actual performance in Anandtech Storage Bench 2013 so quite a big faster than the SandDisk Extreme Pro 480.
4. SanDisk Extreme Pro 480. This is supposed to be a good-performance SSD also from the 2014 era. It uses the Marvell 88SS9187 controller and the 480GB version has 1GB of DRAM and sequential read/write at 550MBps and 515MBps maximum. The memory is 8-way interleaved at 8x8GB dies per package. This is quite a fast drive running 332MBps in the Anandtech Storage Bench 2013.
5. Toshiba OCZ RT150 480GB . This is a 2016-era drive and has the same nominal 550 MBps read and 530 MBps write and it uses the Toshiba TC58. In a different benchmark, the OCZ Trion ran at 507/477 MBps read and write while the Samsung 850 EVO 1TB ran slightly slower at 490/448 MBps read/write. In Storage Bench 2013 Destroyer, it is running 240MBps
6. Samsung 850 Evo 500GB. This is the same family but has lower performance at more like 205 MBps on Storage Bench 2013.

Startech USB3S2SAT3CB with ASMT 2115 (two tests)

In this case with a MacBook Pro M1 Pro Max through a CalDigit Thunderbolt 4 Dock and its USB 3.2 Gen 2 Hub that supports up to 10GBps for the Startech. We would expect the main limit to be the Sata 3 and also I wonder if it matters how full the disk is, so in order of what I would have expected to be the fastest are these benchmarks. One thing to now is that Random reads seem to be stuck at 65MBps, this could be because of the overhead in the USB protocol as we are only running at 5GBps. Note that we are also testing the effect of disk capacity as we move things from these drives to the 4TB ZikeDrive and it’s addlink S95.

What is interesting is that large performance advantage of the EVO 1TB with The Destroyer doesn’t seem to be apparent with this USB 3.0 adapter. There are a couple of theories here, first is that it doesn’t seem to be related to how full the drive is as the performance is the same at 11% full and 80% full and also at 50% full. I also tried it on the different ports of the CalDigit and it looks like the rear ports are faster even though manual says they should all be USB 3.0 Gen 2 ports, that is 10Gbps.

The other theory is that the USB 3.0 conversion may cause this. But the main thing to learn here is that the SanDisk Extreme Pro has the best performance, but they are all quite similar except for the Toshiba OCZ which strangely does well in sequential write at 430MBps but quite slow in random read and write.

Runnig a retest shows that somehow the Samsung 850 EVO slowed down quite a bit on writes for the first few times down to 200MBps and then when I ran again it went back to normal. One aside here is the sequential reads and writes are over 300MBps which

Startech USB3S2SATCB MBps R/W	Full	The Destroyer	Amorphous Seq	Amorphous Rand
Samsung 850 EVO 1TB	80%	373	338/356	65/32
Samsung 850 EVO 1TB	11%	373	339/354	65/33
Samsung 850 EVO 1TB	50%	373	340/195, 338/363	66/31
SanDisk Extreme Pro 480GB	85%	332	354/412	65/34
Toshiba OCZ TR150 480GB	56%	240	356/430	65/26
Samsung 850 EVO 512GB	43%	205	341/363	56/33
MacBook Pro 2021 2TB	80%	N/A	7700/6900	667/138
ZikeDrive Z666 addlink S95 4TB	1%	N/A	3330/2727	526/129

What we can see from the chart above is that for writes, it looks like how full the drive is does have an impact, particularly for random reads and writes. This makes some sense as there is going to be more writes to update the file system. Notice that the ZikeDrive really does deliver some incredible performance that is close to the NVMe 3.0 limit that is driven by the maximum performance of PCIe 3.0 with 4 lanes which is 1GBps/lane or 4GBps. NVMe 4.0 allows up to 2GBps/lane, so 8GBps maximum. That is pretty impressive. The next limit we will hit is the USB 4.0 performance limit of 40Gbps raw where NVMe 3.0 can nearly saturage that with 32Gbps throughputs.

This is a 6Gbps SATA controller that attaches to a USB3 10Gbps system and as such it has great performance.I suspect the limit is probably.

The thing is the Random write is 35MBps which nearly double the Startech, so definitely a keeper if I can figure out where I bought it 🙂 One place where performance on writes isn’t great is the Toshiba which has the same performance as with the Startech. It is interesting to see how the read speeds are all capped at an identical 393MBps which implies some sort of limit in the SATA controller which is quite a bit faster than 300MBps SATA II and confirms that this is a SATA III 6Gbps 600MBps

Startech USB3S2SAT3CB MBps R/w	Full	Destroyer MBps	Startech Seq	Sabrent Seq	Seq	Random
Samsung 850 EVO 1TB	50%	373	339/363	244/236,	393/411	69/35
SanDisk Extreme Pro 480GB	85%	332	354/412	342/451	393/444	74/29
Toshiba OCZ TR150 480GB	56%	240	356/430	344/472	393/349	52/26
Samsung 850 EVO 512GB	43%	205	341/363	195/249	393/411	67/32
MacBook Pro 2021 2TB	80%	N/A	7700/6900	=	=	667/138
ZikeDrive Z666 addlink S95 4TB	1%	N/A	3407/3008	=	=	526/129

Sabrent SATA to USB 3.1 EC-SS31 (great specs, not recommended)

Then I tried the Sabrent SATA to USB 3.1 Adapter. The Sabrent looking at the Apple System Report is a VIA Labs Product ID 0x0715. I think that it is this $13 at Amazon. This supports up to USB 3.1 or 10Gbps compared with the Startech above. What we don’t know is if the Sabrent has a 3Gbps or a 6Gbps SATA interface. I would hope it is the later to better match the 10Gbps USB 3.1 link performance. It also supports the higher performance UASP protocol so it has less link overhead and stands for USB Attached SCSI Peripheral and can be up to 70% faster on reads and 40% faster on writes using up to 80% less computer processing overhead. So lots of hope here, let’s see how it goes.

Looking at Apple > About > System Report, it looks like the USB Controller is a VIA Labs Product ID 0x0715 and a google search shows this is the VL817 USB 3.1 SATA Controller presumably the 2-port version.

Net, net that Sabrent I’m guessing is a real 6Gbps SATA controller, so it’s preferred, so a lot will depend on whether the Sabrent has a 3Gbps SATA or a 6Gbps SATA interface. If it is 3Gbps, the 10Gbps speed of USB 3.1 won’t make any different. This is a good case where the theory and actual don’t match, the Sabreent is actually much slower, note that the Samsung 850 EVO is retested at 50% utilization and is slightly different 332 vs 338MBps for reading so this probably indicates that there are differences in the ports on the system, so for instance the front port of the CalDigit gives the Samsung 850 EVO 1TB 333/127 MBps sequential, but the real port is 339/183 MBps. But in either case compared with the 80% and 11% cases I did above the write speed has slipped quite a bit and I’m going to run five tests now to get a better average. The performance on random was worse going from 65/34 to 20/20 which I really don’t understand.

I’m not quite sure why the Sabrent didn’t perform as well given the specs, but I think its a good example of where testing does matter. It could be there is something with the CalDigit Thunderbolt 4 dock or because these are disks that had content on them. I did try a direct attachment and bypassed the Thunderbolt with a USB A to USB C adapter and I got slightly higher numbers, but not enough to be significant.

The biggest problem with the Sabrent is that Random speeds are really low at 20MBps/11MBps compared with 65MBps/34MBps, so if you have one of these you might think of different USB Adapter.

What was interesting is to see how much better the SanDisk and Toshiba OCZ worked with this controller for sequential yet still very poor for random access. With just 8MBps write at random, it’s really, really slow so both of these are basically unusable with random writes.

Sabrent EC-SSH MBps R/w	Full	Destroyer	Startech Seq	Seq	Random
Samsung 850 EVO 1TB	50%	373	339/363	244/236,	20/12
SanDisk Extreme Pro 480GB	85%	332	354/412	342/451	16/8
Toshiba OCZ TR150 480GB	56%	240	356/430	344/472	13/8
Samsung 850 EVO 512GB	43%	205	341/363	195/249	19/12
MacBook Pro 2021 2TB	80%	N/A	7700/6900	=	667/138
ZikeDrive Z666 addlink S95 4TB	1%	N/A	3407/3008	=	526/129

RIItech USB 3.1 to SATA 5G with ASM1153U Adapter (Recommended)

This is another no-name adapter and here is what we got is asmedia ASM1153 USB 3.2 Gen 1 to SATA 3Gbps. So it spec-wise is between the Startech and the Sabrent. Hopefully it doesn’t disappoint with terrible random scores and it doesn’t. In fact, the scores are marginally higher than the Startech. This must be better design and I’m thinking its actually SATA III running at 6GBps because SATA II at 3Gbps should have a maximum throughput of 300MBps yet these numbers are higher, so I’m a little confused it sounds like the datasheets ;say SATA II, they are SATA III 6Gbps in reality.

The performance is nearly identical with the ASMedia ASTM2115 so the USB C connector is the real difference and a convenience. The small variations in performance are probably more statistical than anything else so the last two are identical. Note that we can see there is definitely a higher ceiling for these last two at 393MBps/447MBps writes.

ASMT 1153U MBps R/2	Destroyer MBps	Startech Seq	Sabrent Seq	ASMT2115 Seq	Seq	Rand
Samsung 850 EVO 1TB	373	339/363	244/236,	393/413	393/411	69/30
SanDisk Extreme Pro 480GB	332	354/412	342/451	393/444	393/446	76/31
Toshiba OCZ TR150 480GB	240	356/430	344/472	393/349	393/447	68/28
Samsung 850 EVO 512GB	205	341/363	195/249	393/411	393/447	67/32
MacBook Pro 2021 2TB	N/A	7700/6900	=	=		667/138
ZikeDrive Z666 addlink S95 4TB	N/A	3407/3008	=	=		526/129

What really matters is random I/O

If you are not just streaming gigantic files, then the specmanship of sequential is less important than random I/Os

Amorphous 4K Random 32Q MBps R/2	Full	The Destroyer	Startech USB3S2SATCB	Sabrent EC-SSH	Startech USB3S2SAT3CB ASMT 2155	ASMT 1153U
Samsung 850 EVO 1TB	50%	373	66/31	20/12	69/35	69/30
SanDisk Extreme Pro 480GB	85%	332	65/34	16/8	74/29	76/31
Toshiba OCZ TR150 480GB	56%	240	65/26	13/8	52/26	68/28
Samsung 850 EVO 512GB	43%	205	56/33	19/12	67/32	67/32
MacBook Pro 2021 2TB	80%	N/A	667/138	=
ZikeDrive Z666 addlink S95 4TB	1%	N/A	526/129	=

ZikeDrive Z666 Performance Notes

This is a completely different kind of drive as mentioned and with the 7GBps read/write drives that support NVM 4.0. It gets us to 2.9GBps and 2.8GBps so not as good as the benchmarks. This is directly attached to the MacBook Pro M1 Pro Max and it’s Thunderbolt 4 (which is also USB4 compatible, they are nearly the same).

As a comparison, the onboard SSD in the MacBook Pro M1 Pro Max is pretty incredible, I get 3.7GBps write and 5.3GBps read, but wow these speeds are pretty crazy good.