[SOLVED] SSD SATA vs. SSD USB - test results

The Electro Machine

Commendable
BANNED
Jan 28, 2021
163
2
1,595
I have performed tests in regards to my old question of "how much read and write speed will I lose if I decide to switch from using remaining SATA connections to USB - for the purpose of reducing the number of cables". That question was asked be me here: https://forums.tomshardware.com/thr...with-usb-getting-rid-of-power-cables.3685529/

The result are almost largely ambiguous and to some extent inconclusive - but also somewhat valid



The tests consistent of coping files with the usage of TeraCopy 3.6, on Windows 10 Enterprise 20H2 19042.746 installed from scratch on NVMe M.2 2280 Samsung 980 PRO 500GB PCle 4.0. The motherboard was Gigabyte X570 Aorus Ultra rev. 1.2

The files were always* copied to SSD 2"5 Samsung 870 QVO 4TB SATA III 560/530 MB/s drive. They were [in that order]:

1] single video files in the size of 75 GB, 20.5 GB, 10 GB, 5 GB and 1 GB - residing on SSD 2"5 Samsung 870 QVO 8TB SATA III 560/530 MB/s drive

2] various audio formats plus some JPGs and TXTs, 12378 of them, weighting altogether 64.2 GB, individually starting with weight from ~50 KB and ending with weight of ~100 MB but mostly [around 85% of them] in size between 1 and 10 MB - residing on NVMe M.2 2280 Samsung 980 PRO 1TB PCle 4.0 drive

3] various document formats but mostly PDFs, 2769 of them, weighting altogether 1.4 GB, individually starting with weight from under 1 KB and ending with weight of ~12 MB but mostly [around 86% of them] under the size of 340 KB - residing on SSD 2.5" Samsung 850 EVO 500GB SATA III 540/520 MB/s 512MB cache drive

Between each of sub-tests [i.e. between 5 GB video and 1 GB and then that whole set of audio files] I gave the drive dozens of second, to cool off itself and clear its cache and give the OS chance to release computing resources [all whatever is a possible benefit of being idle]. In summary: test started with huge files and ended with small and tiny ones, placed on three different drives



As I said, those files were copied to the same 870 QVO 4TB drive placed in:

A] newly bought NO NAME / AliExpress USB 3.0 enclosure [portable case] - and connected to USB-A 3.2 Gen 1 slot placed on the back I/O panel of mobo

B] old NO NAME SATA to USB 3.0 converter - and connected to USB-A 3.2 Gen 1 slot placed on the back I/O panel of mobo [the same one]

C] newly bought Unitek DiskGuard Limpid R USB-C to SATA Enclosure S1103D USB 3.2 Gen 2×1 - and connected to USB-C 3.2 Gen 2 slot placed on the back panel of mobo

D] free space on the desk because I do not have a PC case right now - and connected to SATA3 6Gb/s internal connector [one of those that I want to stop using along with the power cable coming out from PSU]

[*Plus there was also one extra copying done for comparison to SSD 2.5" ADATA Ultimate SU900 256GB SATA III 560/525 MB/s MLC drive utilizing the method "A". And that drive was the only one that has its cache-writing turned off with the option available at Windows System > Control Panel > Device Manager > Disk drives > (DRIVE) > Properties > Policies]



And finally the results: they were all over the place

Sometimes SATA was profoundly better even than USB 3.2 while with other files sometimes USB 3.0 was [illogically] faster than SATA. There were also numerous cases when transfer started steady but started to differentiate at alarming rate; or steady itself after 1/4 or 1/2 of file was done - with big video files and audio files I would get a range from 110 to 310 MB/s while that largest video and tiny documents would jump all over the place

The the only consistent and repeatable value, which I could spot early on, was the speed of verifying tests done by TeraCopy itself after the copying process had ended. For method "C" and "D" [i.e. conventional / ordinary SATA 6Gb/s method and high-end enclosure using fastest USB resulting with overall 10Gbit/s capability] it was 267-268 MB/s. While for method "A" and "B" [i.e. cheap hardware without specification provided by its unknown manufacturer] it was 235 MB/s. Those two values were establishing themselves quickly and did not change until TeraCopy was done with the chek-up. So based on that latter data, it seems that with older technology and when not paying attention to set-up I will achieve 86% of what I get with a traditional SATA or when switching to high-end workarounds with a proper connectivity

As for the actual copying process speeds [i.e. all that what TeraCopy was dong before check-ups], they would usually start with middle value, go down, then speed up to a high and come down to middle value- at least with big video files. Because with small audio and documents the changes were more profound and happening much quicker and showing often very oscillating values that were hard to decide what was the average during a period of [lets say] 10 seconds; which was logical as the small and tiny files varied a lot. But once I repeated almost right away copying of some video file and got staggering different results that a minute ago. Plus 3 times TeraCopy for whatever reason decided no to perform [or just not to show me as a separate task] the process of verification [even despite me trying to see it by over-copying the files], which in the end resulted with no data


I know I could list here even more parameters [like my other PC components] - and that also I should repeat those tests but in a much stricter environment [like turn of the constantly playing audio player and disconnect drives that were not in use in a given variant of test]. But I just do not have the time and energy and I am not running a YouTube tech channel. But I did do what I could without disrupting my light work, i.e. turned off all of the unnecessary software and background tasks that I could thing of and made sure I am using the fastest USB slots as my mobo has many different ones. I also made sure that there was ~20% free space on the destination drive before starting a test. So these tests were more close to how I use those drives and not to some hypothetical and perfect set and settings from factories benchmarks, done with real life various data. I would also like to take the opportunity here and point out that my real life speeds were half or 1/3rd of those stipulated by the manufacturer of those various drives [i.e. Samsung company]



All in all I can now backup my educated guest from aforementioned old topic, that I can ditch SATA connection and switch entirely to USB without loss - assuming I will buy the proper [more expensive] equipment

Back then I predicted a ~17% loss with bad [not adequate enough] hardware being used - and now [the consisted part of ] my empirical data shows measured quite precisely loss in the value of 12.5%





Does anyone has some other thoughts, ideas or empirical data?
 
Last edited:
  • Like
Reactions: Krotow
Solution
The strength of a ssd is in small random read operations.
The weakness is in sustained sequential writes.
Once the buffer in the ssd is filled, performance reverts to the capability of the underlying nand blocks.

The Electro Machine

Commendable
BANNED
Jan 28, 2021
163
2
1,595
Methodology I have already written in details - there is no way I am able to present it in other way / words. [I gave all the relevant hardware and described methods of connection]

As for the data itself - sorry, no can do. I cannot waste any more time on this topic, now that I pretty much know now for sure

[...]
that I can ditch SATA connection and switch entirely to USB without loss - assuming I will buy the proper [more expensive] equipment
[...]
for my drives [i.e. enclosures / connectors / cables]
 
I have performed tests in regards to my old question of "how much read and write speed will I lose if I decide to switch from using remaining SATA connections to USB - for the purpose of reducing the number of cables". That question was asked be me here: https://forums.tomshardware.com/thr...with-usb-getting-rid-of-power-cables.3685529/

The result are almost largely ambiguous and to some extent inconclusive - but also somewhat valid



The tests consistent of coping files with the usage of TeraCopy 3.6, on Windows 10 Enterprise 20H2 19042.746 installed from scratch on NVMe M.2 2280 Samsung 980 PRO 500GB PCle 4.0. The motherboard was Gigabyte X570 Aorus Ultra rev. 1.2

The files were always* copied to SSD 2"5 Samsung 870 QVO 4TB SATA III 560/530 MB/s drive. They were [in that order]:

1] single video files in the size of 75 GB, 20.5 GB, 10 GB, 5 GB and 1 GB - residing on SSD 2"5 Samsung 870 QVO 8TB SATA III 560/530 MB/s drive

2] various audio formats plus some JPGs and TXTs, 12378 of them, weighting altogether 64.2 GB, individually starting with weight from ~50 KB and ending with weight of ~100 MB but mostly [around 85% of them] in size between 1 and 10 MB - residing on NVMe M.2 2280 Samsung 980 PRO 1TB PCle 4.0 drive

3] various document formats but mostly PDFs, 2769 of them, weighting altogether 1.4 GB, individually starting with weight from under 1 KB and ending with weight of ~12 MB but mostly [around 86% of them] under the size of 340 KB - residing on SSD 2.5" Samsung 850 EVO 500GB SATA III 540/520 MB/s 512MB cache drive

Between each of sub-tests [i.e. between 5 GB video and 1 GB and then that whole set of audio files] I gave the drive dozens of second, to cool off itself and clear its cache and give the OS chance to release computing resources [all whatever is a possible benefit of being idle]. In summary: test started with huge files and ended with small and tiny ones, placed on three different drives



As I said, those files were copied to the same 870 QVO 4TB drive placed in:

A] newly bought NO NAME / AliExpress USB 3.0 enclosure [portable case] - and connected to USB-A 3.2 Gen 1 slot placed on the back I/O panel of mobo

B] old NO NAME SATA to USB 3.0 converter - and connected to USB-A 3.2 Gen 1 slot placed on the back I/O panel of mobo [the same one]

C] newly bought Unitek DiskGuard Limpid R USB-C to SATA Enclosure S1103D USB 3.2 Gen 2×1 - and connected to USB-C 3.2 Gen 2 slot placed on the back panel of mobo

D] free space on the desk because I do not have a PC case right now - and connected to SATA3 6Gb/s internal connector [one of those that I want to stop using along with the power cable coming out from PSU]

[*Plus there was also one extra copying done for comparison to SSD 2.5" ADATA Ultimate SU900 256GB SATA III 560/525 MB/s MLC drive utilizing the method "A". And that drive was the only one that has its cache-writing turned off with the option available at Windows System > Control Panel > Device Manager > Disk drives > (DRIVE) > Properties > Policies]



And finally the results: they were all over the place

Sometimes SATA was profoundly better even than USB 3.2 while with other files sometimes USB 3.0 was [illogically] faster than SATA. There were also numerous cases when transfer started steady but started to differentiate at alarming rate; or steady itself after 1/4 or 1/2 of file was done - with big video files and audio files I would get a range from 110 to 310 MB/s while that largest video and tiny documents would jump all over the place

The the only consistent and repeatable value, which I could spot early on, was the speed of verifying tests done by TeraCopy itself after the copying process had ended. For method "C" and "D" [i.e. conventional / ordinary SATA 6Gb/s method and high-end enclosure using fastest USB resulting with overall 10Gbit/s capability] it was 267-268 MB/s. While for method "A" and "B" [i.e. cheap hardware without specification provided by its unknown manufacturer] it was 235 MB/s. Those two values were establishing themselves quickly and did not change until TeraCopy was done with the chek-up. So based on that latter data, it seems that with older technology and when not paying attention to set-up I will achieve 86% of what I get with a traditional SATA or when switching to high-end workarounds with a proper connectivity

As for the actual copying process speeds [i.e. all that what TeraCopy was dong before check-ups], they would usually start with middle value, go down, then speed up to a high and come down to middle value- at least with big video files. Because with small audio and documents the changes were more profound and happening much quicker and showing often very oscillating values that were hard to decide what was the average during a period of [lets say] 10 seconds; which was logical as the small and tiny files varied a lot. But once I repeated almost right away copying of some video file and got staggering different results that a minute ago. Plus 3 times TeraCopy for whatever reason decided no to perform [or just not to show me as a separate task] the process of verification [even despite me trying to see it by over-copying the files], which in the end resulted with no data


I know I could list here even more parameters [like my other PC components] - and that also I should repeat those tests but in a much stricter environment [like turn of the constantly playing audio player and disconnect drives that were not in use in a given variant of test]. But I just do not have the time and energy and I am not running a YouTube tech channel. But I did do what I could without disrupting my light work, i.e. turned off all of the unnecessary software and background tasks that I could thing of and made sure I am using the fastest USB slots as my mobo has many different ones. I also made sure that there was ~20% free space on the destination drive before starting a test. So these tests were more close to how I use those drives and not to some hypothetical and perfect set and settings from factories benchmarks, done with real life various data. I would also like to take the opportunity here and point out that my real life speeds were half or 1/3rd of those stipulated by the manufacturer of those various drives [i.e. Samsung company]



All in all I can now backup my educated guest from aforementioned old topic, that I can ditch SATA connection and switch entirely to USB without loss - assuming I will buy the proper [more expensive] equipment

Back then I predicted a ~17% loss with bad [not adequate enough] hardware being used - and now [the consisted part of ] my empirical data shows measured quite precisely loss in the value of 12.5%





Does anyone has some other thoughts, ideas or empirical data?
Just going by bus speed sata wins but there is all kinds of stuff happening with moving data as your finding out.

Start moving out to usb only and see if something jumps out at you.
 

The Electro Machine

Commendable
BANNED
Jan 28, 2021
163
2
1,595
Couple of large enough M.2 NVMe drives would solve all problems. No cables at all + super speed.
How many motherboards have more than 3 M.2 NVMe slots?

How many motherboards have full speeds for all of the M.2 NVMe slots without some kind of limitations?

What is the average difference in price between 8GB SATA SSD and 8GB M.2 NVMe SSD?



I already have 0.5 and 1 TB M.2 NVMe drives - and I am waiting for arrival of 4 TB one. And at the same time I am using 18TB ad 16TB HDDs; plus 8TB, 4TB, 2TB and 1TB SATA SSD



The strength of a ssd is in small random read operations.

The weakness is in sustained sequential writes.

Once the buffer in the ssd is filled, performance reverts to the capability of the underlying nand blocks.
Thank you for that insightful info - but it is completely irrelevant to what I am stipulating: changing connection of SSDs [and possibly of also HDDs when using additional external power source] to USB



Just going by bus speed sata wins but there is all kinds of stuff happening with moving data as your finding out.

Start moving out to usb only and see if something jumps out at you.
Prior to these tests I had that 4TB SSD connected via some other NO NAME 2-year old USB 3.0 adapter for over 6 months. And I did not stumble upon issues with this solution. [It is only now that I know it is slightly slower than SATA connection and USB 3.2 gen 2 x 1]
 
How many motherboards have more than 3 M.2 NVMe slots?

How many motherboards have full speeds for all of the M.2 NVMe slots without some kind of limitations?

What is the average difference in price between 8GB SATA SSD and 8GB M.2 NVMe SSD?



I already have 0.5 and 1 TB M.2 NVMe drives - and I am waiting for arrival of 4 TB one. And at the same time I am using 18TB ad 16TB HDDs; plus 8TB, 4TB, 2TB and 1TB SATA SSD




Thank you for that insightful info - but it is completely irrelevant to what I am stipulating: changing connection of SSDs [and possibly of also HDDs when using additional external power source] to USB




Prior to these tests I had that 4TB SSD connected via some other NO NAME 2-year old USB 3.0 adapter for over 6 months. And I did not stumble upon issues with this solution. [It is only now that I know it is slightly slower than SATA connection and USB 3.2 gen 2 x 1]
Dont forget about PCIe to M.2 adapters.
 

The Electro Machine

Commendable
BANNED
Jan 28, 2021
163
2
1,595
Dont forget about PCIe to M.2 adapters.
But of course I remember [but to be precise I think it should be the other way around: M.2 > PCIe]

But I also remember that there are possible problems with adding e.g. a second GPU, because it will not have as many PCIe lines like the top most which is by the manufacturers usually recommended to host [a single] GPU on their motherboards. And thus the same goes for any other PCIe connected devices. From what I remember we have to wait for a combo of PCIe 5.0 and AMD Zen 4 processors to alleviate such issues
 
But of course I remember [but to be precise I think it should be the other way around: M.2 > PCIe]

But I also remember that there are possible problems with adding e.g. a second GPU, because it will not have as many PCIe lines like the top most which is by the manufacturers usually recommended to host [a single] GPU on their motherboards. And thus the same goes for any other PCIe connected devices. From what I remember we have to wait for a combo of PCIe 5.0 and AMD Zen 4 processors to alleviate such issues
Any way you turn around your a$$ is behind you. :)

PCIe x 16 slot > Adapter to 1 - 4 M.2 slots > 1-4 NVME (PCIe x2 or 4) depending on how many PCIe (free) lines your PC has. Even if you get only PCIe x2 on each drive, till helluva faster than any SATA or USB connection.
 

The Electro Machine

Commendable
BANNED
Jan 28, 2021
163
2
1,595
Last edited: