MP All Models External TB3 NVMe SSD - only 800 MB/s write speed

[bsbeamer]

so it appears to be specifically linked to 970 EVO Plus

Again, there are numerous macOS issues with the 970 EVO Plus blades. When they were originally released, they wouldn't even work in macOS or would fail miserably if trying to workaround the issue. The initial firmware update "fixed" the main issue, but many have complained about additional issues since then when using older firmware.

Unless you're going to update the firmware, all of this is just complaining to complain. Samsung has attempted to address most of the issues.

 

[MrAverigeUser]

With "new controller" I was referring to the 970 EVO - the EVO Plus has a different controller than the original EVO.

I have another newish crucial P1 that works fine in this enclosure. I also saw tests that 980 Pro seems to work fine (2800/2300), so it appears to be specifically linked to 970 EVO Plus.

Running Monterey 12.3

It just feels dirty to have to buy a 50% more expensive drive just so i can get the speed I already have with an older drive, especially because it's nerfed in the Thunderbolt enclosure anyway.

I've seen some tests that the PCIe4.0 controller of the updated 970 EVO Plus plays better with the M1. I'm not sure what the shop/service will do, still waiting for a response from them (not holding my breath). I'll probably just return it and splurge for the 980 Pro and hope for Thunderbolt 5 in M2 so I can actually utilise it

Ok,
well... I have no experience with the M1 Macs, only with the MBP mid 2015.

I found an article in an online-IT Publication (German language, last updated september 2021) that gives some facts about modifications of the 970 EVO plus. which I will nevertheless link at the end of this posting because there are very interesting screenshots and diagrams (their source for the description is an asian YT-channel, linked in the article).

Evidently, due to the pandemic situation Samsung had problems to produce the newer version of the Phoenix controller (with its newest 2.x firmware) and decided to produce the 970 EVO plus using the ELPIS controller, at that time with its 3.x firmware, now with the newer 4.x firmware. The Elpis controller (developed for the younger 980 PRO) uses more Cache than the Phoenix controller and therefore shows - dependent on the file-size being written - much faster write speeds than the "old" 970 EVO plus version , but (at least if paired with the 3.x firmware) only if the file has less than about 115 GB.
If you write more than this at one time, the cache is full and cannot be as fast, yet writes from that point on slower than the 970 EVO plus with the Phoenix controller (tested last year with firmware 3.x).

The "old" 970 EVO plus with Phoenix controller is far behind, because its cache is much smaller and saturates very fast.

BUT:
since the new 970 EVO plus gets much slower than the constantly writing old one, the old 970 EVO plus gets on par at about 150 GB ( that means at 150 GB both have needed the same time for about 150 GB) and is even faster if you write even more than 150 GB at one time.

Bottom line:
if you write/copy mostly very big files of more than 150 Gb at one time you are better off with the Phoenix-type and firmware 2.x
If you write/copy mostly files of less than 150 GB at one time the Elpis-type with firmware 3.x will be better for you.

reminder: This test of the Elpis-version was done only with firmware 3.x and NOT with the newer 4.x version!

I did not find any comparison with Elis and firmware 4.x which might lead to different results.

Perhaps my findings will help you a little bit.


-----------------------------------------------------------

I have no time to translate the whole article, but give you some translations to understand better the screenshots and their descriptions:

links = left
rechts = right

neu = new
alt = old

and the asian word for "old" looks like "18"

970 Evo Plus: Auch Samsung tauscht SSD-Komponenten aus

Zu den Herstellern, die bei bestehenden SSD-Serien entscheidende Komponenten wechseln, zählt auch Samsung.

www.computerbase.de

 

[Ploki]

Ok,
well... I have no experience with the M1 Macs, only with the MBP mid 2015.

I found an article in an online-IT Publication (German language, last updated september 2021) that gives some facts about modifications of the 970 EVO plus. which I will nevertheless link at the end of this posting because there are very interesting screenshots and diagrams (their source for the description is an asian YT-channel, linked in the article).

Evidently, due to the pandemic situation Samsung had problems to produce the newer version of the Phoenix controller (with its newest 2.x firmware) and decided to produce the 970 EVO plus using the ELPIS controller, at that time with its 3.x firmware, now with the newer 4.x firmware. The Elpis controller (developed for the younger 980 PRO) uses more Cache than the Phoenix controller and therefore shows - dependent on the file-size being written - much faster write speeds than the "old" 970 EVO plus version , but (at least if paired with the 3.x firmware) only if the file has less than about 115 GB.
If you write more than this at one time, the cache is full and cannot be as fast, yet writes from that point on slower than the 970 EVO plus with the Phoenix controller (tested last year with firmware 3.x).

The "old" 970 EVO plus with Phoenix controller is far behind, because its cache is much smaller and saturates very fast.

BUT:
since the new 970 EVO plus gets much slower than the constantly writing old one, the old 970 EVO plus gets on par at about 150 GB ( that means at 150 GB both have needed the same time for about 150 GB) and is even faster if you write even more than 150 GB at one time.

Bottom line:
if you write/copy mostly very big files of more than 150 Gb at one time you are better off with the Phoenix-type and firmware 2.x
If you write/copy mostly files of less than 150 GB at one time the Elpis-type with firmware 3.x will be better for you.

reminder: This test of the Elpis-version was done only with firmware 3.x and NOT with the newer 4.x version!

I did not find any comparison with Elis and firmware 4.x which might lead to different results.

Perhaps my findings will help you a little bit.


-----------------------------------------------------------

I have no time to translate the whole article, but give you some translations to understand better the screenshots and their descriptions:

links = left
rechts = right

neu = new
alt = old

and the asian word for "old" looks like "18"

970 Evo Plus: Auch Samsung tauscht SSD-Komponenten aus

Zu den Herstellern, die bei bestehenden SSD-Serien entscheidende Komponenten wechseln, zählt auch Samsung.

www.computerbase.de

I read that, found an english article - but the speeds there are 3500mb/s vs 3300mb/s (or something like that)
this is 900mb/s vs 1700mb/s for the old EVO non-plus, and it indeed copies smaller files (~10GB) faster on the old EVO non-plus.
I'd be happy with either EVO Plus if it deliver the performance of the EVO non-plus i have.

Again, there are numerous macOS issues with the 970 EVO Plus blades. When they were originally released, they wouldn't even work in macOS or would fail miserably if trying to workaround the issue. The initial firmware update "fixed" the main issue, but many have complained about additional issues since then when using older firmware.

Unless you're going to update the firmware, all of this is just complaining to complain. Samsung has attempted to address most of the issues.

Couldn't update the firmware, the boot drive updated didn't find the drive unfortunately and i don't have access to a windows machine. I told the service provider to attempt a firmware upgrade, else i'll probably just go for the 980 Pro instead, ironically, to get advertised 970 EVO Plus speeds. Or i'll push them to replace the one i bought with the ELPIS (980 Pro) controller. Something.
Wish i did research prior, but my extremely positive experience with the 970 EVO (non-plus) gave me confidence.

 

[ScratchyMoose]

With the Orico M2V01 and a Sabrent 4 TB Rocket (SB-ROCKET-4TB), I get 1.700 - 1.800 MB/sec write and 2.700 - 2.800 MB/sec read

hi there micheloupatrick, can you confirm that that’s the enclosure you’re using, because it’s meant to only go up to 2tb according to the Amazon page for them M2V02?
Would love to know if it can infect take 4tb!!!

 

[ScratchyMoose]

Hi All, just to feedback incase someone finds this info useful.

I'm using a 4TB Sabrent NVMe SB-RKTQ-4TB with a Yottamaster Thunderbolt 3 enclosure. I was having real problems when assessing the speed with Blackmagic disk speed, and also AJA System Test light. At first I was getting around 2100 write and 2400 read MB/s speeds, but then within 30 seconds or so, it'd drop down to about 1100 and 700. With both testing apps. Tried different cables, and also a different enclosure, Fledging Shell Thunder. Both enclosures (T3) were showing more or less the same slow down.

I called Sabrent in the US, and they advised me to try using the Sensei app for speed tests, and low and behold (with both enclosures) I'm now getting very stable speeds of 2100 and 2600 (w/r)! See the screen grab for two from consecutive days.

@ubercool can you add these to your fab spreadsheet?!

In terms of Sensei being off - the guy from Sabrent said that it would seem they're more accurate than the Blackmagic or Aja ones ... I'm going to do some transfers from the internal (faster) drive to the NVMe to see how long it takes and then work out the speeds that way, over a large transfer, which should then check if the Sensei results are accurate, but for now, I think I'm going to accept them. Be interested to hear people who have had slow speeds experiences when comparing Blackmagic, if they have a go with Sensei and see what happens?

Anyway, good luck everyone
ps when doing the benchmarks or transferring large files (like 1Tb) the NVMe stays at about 28C in the Yottamaster enclosure

 

[bsbeamer]

I'm using a 4TB Sabrent NVMe SB-RKTQ-4TB with a Yottamaster Thunderbolt 3 enclosure. I was having real problems when assessing the speed with Blackmagic disk speed, and also AJA System Test light. At first I was getting around 2100 write and 2400 read MB/s speeds, but then within 30 seconds or so, it'd drop down to about 1100 and 700. With both testing apps. Tried different cables, and also a different enclosure, Fledging Shell Thunder. Both enclosures (T3) were showing more or less the same slow down.

I called Sabrent in the US, and they advised me to try using the Sensei app for speed tests, and low and behold (with both enclosures) I'm now getting very stable speeds of 2100 and 2600 (w/r)!

If Sensei can run an extended test or with larger files, enable that. You likely will have same results as Blackmagic and AJA. Just using a different app to test does not change anything with the hardware. You're just masking the "issue" with their recommended software.

 

[ScratchyMoose]

If Sensei can run an extended test or with larger files, enable that. You likely will have same results as Blackmagic and AJA.

I don't get the feeling that that's the case because the Sensei test itself runs for as long or longer than the other two, and I can see the figures changing very rapidly, many times a second, which makes me think that the results are not being artificially pegged or just using the top score every 1/2 or 1/4 second. I know the weak point with this argument is that you'll say it's being creamed off every 1/2000s and as my monitor is 1/60 how could I tell? With the other two, the speed drops to a max of 700 for quite a few mins before it comes back. If the drive was outputting maximum spikes of speed at 700, then how could Sensei come up with 2400 without cheating? Would the developer do this? I don't know, but I'm guessing not. If the spike is 2400 but only happens 1 out of 2000 times, would the developer risk their reputation on saying that the speed of the drive is 2400?

Also the figures are more or less the same with the other two speed tests, it's just that they drop off (and it's not heat related as the drive stays a constant temperature), which means the drive can do it.

Of course that could all be wrong (and the best way to test it would be to transfer something like 1Tb while timing it) but for now, I'm feeling ok about the situation.

 

[bsbeamer]

Sensei is a subscription based "all-in-one" drive utility that advertises itself as having the same capabilities as CleanMyMac... says all I need to know about avoiding the software.

 

[funkahdafi]

Sensei is a subscription based "all-in-one" drive utility that advertises itself as having the same capabilities as CleanMyMac... says all I need to know about avoiding the software.

You're not doing them enough justice. The developers of the Sensei app are well established and have been providing Mac apps for over a decade. It's the same dev that allowed all of us to enable Trim on our SSDs when Apple didn't allow us to do that (yes, that actually happened, long time ago when SSDs started to become a thing in Macs).

The Sensei app is pretty good, and it combines hardware monitoring, drive health monitoring (SMART), benchmarks and a couple other things as well.

 

[ScratchyMoose]

that advertises itself as having the same capabilities as CleanMyMac

Haha, that made me not take lunch and do some tests!

So, what I did was create a folder with 2,960 camera raw files, which came in at 100 Gb exactly. First I copied that to and from my 8HDD Raid (which clocks the following speeds on average on Blackmagic: 1721 / 1954 w/r, while on Sensei the Raid was getting 1587 / 1063 w/r). I did the same with the NVMe. Each time the pairing was either Raid / Mac Studio or NVMe / Mac Studio (the Mac Studio gets 5000 + in both directions, so shouldn’t be limiting).

Here are the results (in seconds, average of 3 runs):

Writing (copy onto):
NVMe - 44.1
Raid - 126.8

Reading (copy from):
NVMe - 39.1
Raid - 153.5

I couldn’t figure out why the Raid would have it’s quickest time writing as that’s not the way round in Blackmagic (that lead me to do a Sensei test on the Raid, giving me the numbers quoted before. Seems the Sensei figures are more realistic here). The NVMe is quickest Reading, which is along the lines I’d expect from the speed tests, but because of the Raid anomaly, I did the copying for a third time to check, and those figures were consistent with the previous two runs.

Then I thought that the figures from the NVMe might drop after a while - which was the whole problem that I started with - so I repeated the Read (copying from) x7 in quick succession, and I got a very consistent 39.3 average.

But at first (before the x7 run I mentioned above), the times were getting longer and longer, before then I realised that that was because my McS SSD was filling up to the brim - once I emptied it, and emptied it each time after copying - the times kept very very stable (ie less than 0.3 secs difference on my stopwatch for the x7 runs. You can see the tail end of this graph from iStat Menu showing that the blocks are solid (x5 of them, with the time taken to empty the McS SSD being the gaps between), with a max of 2400. The second attachment is of the Raid from iStat menu, showing that the Raid was operating at 655.

The numbers from the iStat screen grabs and the timed 100Gb blocks:
NVMe - 2400
Raid - 655
NVMe - 39.1s
Raid - 153.5s

Those as %s would be:
27% iStat
25% seconds (flipped 1/x)
Quite similar numbers (as are 2400 x 39 and 655 x 153)!

What I’m taking away from this is:
1. The NVMe is over 2.5 times faster than my Raid to write
2. The NVMe is nearly 4 times faster than my Raid to read
3. The NVMe doesn’t slow down under repeated use
4. Blackmagic and Aja showing that the NVMe and Raid are similar in speed is wrong.

So … I’m keeping with my original thoughts - that the NVMe’s speed is better represented by the Sensei app, than the Blackmagic and AJA apps. But talk about confirmation bias - because this makes me very happy, I’m sure I’m being blind to something, but hopefully that something isn’t great enough to invalidate the above!!

 

[ScratchyMoose]

You're not doing them enough justice. The developers of the Sensei app are well established and have been providing Mac apps for over a decade. It's the same dev that allowed all of us to enable Trim on our SSDs when Apple didn't allow us to do that (yes, that actually happened, long time ago when SSDs started to become a thing in Macs).

The Sensei app is pretty good, and it combines hardware monitoring, drive health monitoring (SMART), benchmarks and a couple other things as well.

Good to know, thanks for saying

 

[tommy chen]

You're not doing them enough justice. The developers of the Sensei app are well established and have been providing Mac apps for over a decade. It's the same dev that allowed all of us to enable Trim on our SSDs when Apple didn't allow us to do that (yes, that actually happened, long time ago when SSDs started to become a thing in Macs).

The Sensei app is pretty good, and it combines hardware monitoring, drive health monitoring (SMART), benchmarks and a couple other things as well.

but does not recognize raids :- (

 

[funkahdafi]

That's because it looks at disks, not volumes (like Blackmagic). It still works, though. Select the disk, start a benchmark, and it will ask you for the volume you want to use for the benchmark. Select your raid volume.

 

[micheloupatrick]

hi there micheloupatrick, can you confirm that that’s the enclosure you’re using, because it’s meant to only go up to 2tb according to the Amazon page for them M2V02?
Would love to know if it can infect take 4tb!!!

Yes, it's the M2V01-C4 and it works perfectly fine with the Sabrent Rocket 4TB (the blue one).

 

[godzfire]

Watching this thread because I bought a 1TB 970 Evo Plus back in Nov 2020 not knowing any of these issues specific to Samsung and this exact model, regarding to the incompatibilities and transfer speed issues with macOS and especially Monterey. Trying to update my firmware via Mac boot USB firmware updater doesn't work. In Parallels booting the same USB drive updater does work but says there's no drive/drive to update.

So am I just screwed? What TB3 enclosures do people have that also have a 970 Evo Plus that actually is pulling the proper transfer speeds?

Samsung SSD 970 EVO Plus 1TB:
Capacity: 1 TB (1,000,204,886,016 bytes)
TRIM Support: Yes
Model: Samsung SSD 970 EVO Plus 1TB
Revision: 2B2QEXM7
Serial Number: S59ANMFN909743P

 

[Ploki]

I got offered a refund or wait for Elpis batch. Thinking of simply going for the 980 Pro and be done with it

 

[haddy]

I got offered a refund or wait for Elpis batch. Thinking of simply going for the 980 Pro and be done with it

Yep they are what I use

 

[godzfire]

I got offered a refund or wait for Elpis batch. Thinking of simply going for the 980 Pro and be done with it

I'm going to contact Samsung tomorrow about the situation and demand something similar. None of this was ever highlighted on Amazon's page for the 970 Evo Plus back in Nov 2020 when I purchased it regarding issues with Mac, otherwise I would have never chosen that model and wasted $120.

 

[Ploki]

I'm going to contact Samsung tomorrow about the situation and demand something similar. None of this was ever highlighted on Amazon's page for the 970 Evo Plus back in Nov 2020 when I purchased it regarding issues with Mac, otherwise I would have never chosen that model and wasted $120.

There's only official samsung channels here for phones/tablets, not for drives and computers, so i can only deal with authorised resellers that are very unhelpful.
The guy i talked with didn't even know the new batch has different controllers and went on a rant how "thunderbolt is less and less popular and supported" or something.

Given how well the original 970 EVO i had performed, i'm really disappointed in this whole ordeal

 

[godzfire]

Swapped the OWC Envoy Express with an ACASIS FA-401, I'm now getting the proper 2500 MB/s read, but still only 1200 MB/s write, half of what the case and the 1TB EVO Plus is capable of.

I have the Trebleet case coming sometime between the 18th and 26th, but I have a feeling I am going to be getting the same results. I'm definitely calling Samsung's SSD warranty division and asking for a replacement that doesn't have these issues, either the newer 970 EVO or the 980.

Does the newer 970 Evo also have these issues?

I've heard people talk about an ORICO case. I can't find it. Does anyone have a link for it? Does it work any better?
EDIT: I tried going on the Oricho site, and unless you do some digging, you can't easily find any mention of modern 40 GBs TB3 NVME cases anymore, and just doing a Google search they aren't available really anywhere including Amazon.

 

[F-Train]

@godzfire,

Sonnet Tech's SSD compatibility list has a footnote on the 970 EVO Plus saying that it won't do better than 1200MB/s write in Sonnet's own Thunderbolt 3 enclosures. It's the only SSD to which the note applies.

Yesterday, before seeing that and reading this thread, I ordered a 970 EVO Plus 2TB direct from Samsung, but I'll be returning it when it arrives and buying something else.

 

[godzfire]

@godzfire,

Sonnet Tech's SSD compatibility list has a footnote on the 970 EVO Plus saying that it won't do better than 1200MB/s write in Sonnet's own Thunderbolt 3 enclosures. It's the only SSD to which the note applies.

Yesterday, before seeing that and reading this thread, I ordered a 970 EVO Plus 2TB direct from Samsung, but I'll be returning it when it arrives and buying something else.

Can you link me that please?

 

[F-Train]

Can you link me that please?

This is the Sonnet Tech .PDF on SSD compatibility as of February 2022: https://sonnettech.com/support/downloads/manuals/M2_compatibility.pdf

Go to Samsung > 970 EVO Plus > footnote #9.

Footnote #9 text: "When used via Thunderbolt, write speed limited to ~1200MB/s."

 

[Ploki]

I returned the 970 EVO plus 2tb and upgraded to 980 Pro, didnt want to deal with another return if the elpis 970 EVO Plus performed just as bad.

It’s frustrating since 970 Evo pre-plus works flawlessly. Oh well

 

[indiekiduk]

Has anyone tried the 980? Which is supposedly the successor to the 970 Evo Plus (They dropped the Evo name for the 980 generation). It has no DRAM so is cheaper and instead uses a little bit of system RAM, which is called HMB. I have a feeling that is not going to work well in a TB3 caddy.