External SSD Strangely Slow Read Speeds - Advice?

[Church13]

I have a Gen4 M.2 NVMe drive I've been using in an external enclosure for a while now, and for some reason it has insanely slow read speed. Writes are mostly consistent with what I'd expect, but performance is honestly all over the place.

I'm on a MacBook Pro 14", M3 Pro / 18GB / 512GB. The drive is formatted APFS. Drive is a Samsung PM9A1, but I had the exact same issues with an SK Hynix external SSD. My primary use case for this drive is as an external game drive, and it's basically useless for this with the real-world performance I'm getting.

I have tried two separate 40Gbps SSD enclosures (a Ugreen and now an Orico), and the behavior is the same. The cable is a nice thick USB4 40Gbps cable, so there's no issue there.

My Blackmagic speed test results are below. It'll report these speeds consistently.

Paradoxically, AmorphousDiskMark gives me the speeds I would expect for a properly functioning external drive, but as I said, in real world performance it's miserable, games and files are taking forever to load etc.

Any thoughts / advice would be appreciated! I basically never use the thing at this point because it's so awful.

 

[StoneJack]

Change samsung

 

[monokakata]

Can you get your hands on another Mac? That way, you could exclude the possibility that the issue lies with your MacBook, not the drive. Yes...a long shot. But taking that possibility off the table would be helpful.

 

[bzgnyc2]

I have a Gen4 M.2 NVMe drive I've been using in an external enclosure for a while now, and for some reason it has insanely slow read speed. Writes are mostly consistent with what I'd expect, but performance is honestly all over the place.

I'm on a MacBook Pro 14", M3 Pro / 18GB / 512GB. The drive is formatted APFS. Drive is a Samsung PM9A1, but I had the exact same issues with an SK Hynix external SSD. My primary use case for this drive is as an external game drive, and it's basically useless for this with the real-world performance I'm getting.

I have tried two separate 40Gbps SSD enclosures (a Ugreen and now an Orico), and the behavior is the same. The cable is a nice thick USB4 40Gbps cable, so there's no issue there.

My Blackmagic speed test results are below. It'll report these speeds consistently.

View attachment 2485082

Paradoxically, AmorphousDiskMark gives me the speeds I would expect for a properly functioning external drive, but as I said, in real world performance it's miserable, games and files are taking forever to load etc.

View attachment 2485083

Any thoughts / advice would be appreciated! I basically never use the thing at this point because it's so awful.

You will probably need to change out the Samsung as someone else mentioned but if you want to be sure, trying the drive on another system as another person mentioned would be a good test.

I would also check the drive's SMART stats using either DriveDx or "smartmontools". That model of SSD has a relatively low TBW of 600 TB. Between that and it hitting the 5 year mark (https://www.techpowerup.com/ssd-specs/samsung-pm9a1-1-tb.d786), it could be just that its time is up. I would have thought writes would suffer more than reads, though if it hit its TBW limit.

Also, by chance are you running Sequoia and if so do you have encryption enabled [APFS (Encrypted)]?

P.S.I'd be curious if your AmorphousDiskMark results change if you increase the file size to the max (there are so many caches in the chain these day, it's hard to know what you are exercising with just 1GiB).

 

[HobeSoundDarryl]

• Run Disk Utility app "First Aid" on the drive.
• Try another cable. "Nice & thick" is fine but it can still be a bad cable or cable that has gone bad.
• Try another MB port.
• Fully power down the MB, wait a minute, power up and try again.
• Power #3 is definitely worth a try... or even a PC just to retest read/write and rule out/in that MB.
• If none of that does anything, how is the drive formatted?
• Do you have it set up to properly dissipate heat?

 

[UnifiedMelody]

I tried to do my 970 Evo Plus 2TB NVMe on external drive and holy smokes it gets hot way too quick.
Not to mention the transfer speed is slow [limited due to protocol]. USB 3.1. [i cbf using TB4 lol for this one, its mostly storing work photos which wouldn't even require >500MB/s.

Ended up purchasing a 4TB Samsung T9 instead. Which has better thermal management and guaranteed 1k MB/s speeds which is fine.

I felt like NVMe's were only good if you can actually a fan onto it otherwise it gets hot too quick lol.

 

[bzgnyc2]

I tried to do my 970 Evo Plus 2TB NVMe on external drive and holy smokes it gets hot way too quick.
Not to mention the transfer speed is slow [limited due to protocol]. USB 3.1. [i cbf using TB4 lol for this one, its mostly storing work photos which wouldn't even require >500MB/s.

Ended up purchasing a 4TB Samsung T9 instead. Which has better thermal management and guaranteed 1k MB/s speeds which is fine.

I felt like NVMe's were only good if you can actually a fan onto it otherwise it gets hot too quick lol.

I think it varies based on the NVMe drive and enclosure. Note that many if not most USB 3.x SSD drives that do ~ 1GB/sec are using NVMe drives internally and the main thermal advantage you would be getting in that case is de facto throttling. Otherwise I have not heard that the NVMe protocol over USB4/TB3+ has higher overhead than USB 3.x.

It is also possible that the first time an already written NVMe SSD is connected directly to a system (i.e. via USB4/TB3+), the system will initiate a large set of TRIM operations that could cause the drive to run hot until complete.

In any case I agree one drawback with buying a separate enclosure and NVMe drive is that one essentially becomes a system engineer responsible for among other things the heat management. Often people select their SSD based on a peak speed and/or good review but these often do not take into consideration heat and other criteria that are critical to the total solution. Tom's Hardware reviews of SSD usually include not just peak performance, but sustained write performance and power efficiency, particularly important for the external enclosure situation:

WD Black SN850X SSD Review: Back in Black (Updated)

The WD Black SN850X is one of the fastest drives around for gaming.

www.tomshardware.com

Also this database that can be helpful getting all the specs for a particular drive or finding drives that meet key criteria:

TechPowerUp

Solid-state-drive database with specifications for products launched in recent years. Includes specs, photos, and technical details.

www.techpowerup.com

Still one runs the risk that an otherwise seemingly good drive doesn't work in a particular case like the issues with TRIM on Samsung 980+ drives under macOS Monterey+. In other cases, there are reports of drives not properly handling NVMe power states until the firmware is upgraded.

 

[Church13]

Hey all, thanks for the responses!

Change samsung

As I mentioned in the original post, the Samsung is the second drive I've tried and it's the same issue. Previous one was an SK Hynix Gold 1TB.

Can you get your hands on another Mac?

Yes, my wife has an M2 Pro MacBook Pro and it exhibits the same behavior there! I'm really thinking this is some oddity in how macOS handles external drives.

You will probably need to change out the Samsung as someone else mentioned but if you want to be sure, trying the drive on another system as another person mentioned would be a good test.

I would also check the drive's SMART stats using either DriveDx or "smartmontools". That model of SSD has a relatively low TBW of 600 TB. Between that and it hitting the 5 year mark (https://www.techpowerup.com/ssd-specs/samsung-pm9a1-1-tb.d786), it could be just that its time is up. I would have thought writes would suffer more than reads, though if it hit its TBW limit.

Also, by chance are you running Sequoia and if so do you have encryption enabled [APFS (Encrypted)]?

P.S.I'd be curious if your AmorphousDiskMark results change if you increase the file size to the max (there are so many caches in the chain these day, it's hard to know what you are exercising with just 1GiB).

It's actually a brand new drive. Yes the model is five years old but this has only 100 power on hours, and 3 terabytes written/read in total. Sure, there's a chance of it being a faulty drive, but again, this isn't the only drive I've had the issue with. I am running Sequoia, but no encryption enabled. I've run the same test with 8 GiB blocks on AmorphousDiskMark and still getting the same values -- BlackMagic never crosses over 100MB/s read under any circumstances.

• Run Disk Utility app "First Aid" on the drive.
• Try another cable. "Nice & thick" is fine but it can still be a bad cable or cable that has gone bad.
• Try another MB port.
• Fully power down the MB, wait a minute, power up and try again.
• Power #3 is definitely worth a try... or even a PC just to retest read/write and rule out/in that MB.
• If none of that does anything, how is the drive formatted?
• Do you have it set up to properly dissipate heat?

Done all the basic due diligence before posting here (steps you listed + more). I'm still unwilling to accept a cable issue given that I get the speeds I expect during basic (large) file transfers and in AmorphousDiskMark. In System Information I can see that it's linking at the full expected 40Gbps. Heat is also not an issue, both the Ugreen and Orico enclosures are effectively slabs of aluminum which the drive transfers heat to through a thermal pad, and they are actively cooled with built-in fans.

I tried to do my 970 Evo Plus 2TB NVMe on external drive and holy smokes it gets hot way too quick.
Not to mention the transfer speed is slow [limited due to protocol]. USB 3.1. [i cbf using TB4 lol for this one, its mostly storing work photos which wouldn't even require >500MB/s.

Ended up purchasing a 4TB Samsung T9 instead. Which has better thermal management and guaranteed 1k MB/s speeds which is fine.

I felt like NVMe's were only good if you can actually a fan onto it otherwise it gets hot too quick lol.

Like I responded to HobeSoundDarryl, it's definitely not a cooling issue (both passive and active cooling in enclosures), though at this point I'm strongly considering just getting a bespoke external drive like a T9 which is purpose-built.

Still one runs the risk that an otherwise seemingly good drive doesn't work in a particular case like the issues with TRIM on Samsung 980+ drives under macOS Monterey+. In other cases, there are reports of drives not properly handling NVMe power states until the firmware is upgraded.

I think this is really what I'm honing in on. I suspect that the drive is being commanded to a lower power state and staying there for most read-intensive workloads, which are then bottlenecked as a result. AmorphousDiskMark, for some reason, is able to pull it out of that low power state and get the drive to perform as expected. I think macOS is poorly handling the power state management of the external drive, and real-world performance in varied I/O tasks is suffering.

I'll try to see if I can update the firmware on the drive from a Windows or *nix system. I know the Hynix drive I was using previously had the most current firmware at the time, but that's just an anecdotal observation which is no longer relevant at this point.

 

[Church13]

Update -- and this is really interesting. I have a Dell monitor with an integrated Thunderbolt 4 dock which I use as my primary external monitor. When I plug it into a standard 10Gbps USB-C port daisy-chained off the monitor/dock, BlackMagic shows more correct R/W numbers, in this case maxing out the 10Gbps link. Truly bizzare. This is with the same USB4 40Gbps cable and Orico enclosure. I'm really thinking macOS is trying to be too smart and manage the drive's power state etc when it's on the Thunderbolt/USB4 link. In fact, under System Information, it does come up under Thunderbolt/USB4 when plugged directly into the MBP, but it comes up under USB when plugged into the monitor/dock which has a completely different arbitration method/service in macOS.

 

[Church13]

Some other information I've turned up online might suggest that it's a firmware issue with the enclosure itself -- see below:

https://www.reddit.com/r/UsbCHardware/comments/1bt10ue/ugreen_40gbps_nvme_ssd_enclosure_mac_os

Going to go down this rabbit hole as well and report back.

 

[macnavarra]

Open terminal in boot mode : csrutil disable, boot... open terminal in macOS: sudo nvram boot-args="vm_compressor=2" ,... sudo trim force disable , open terminal in boot mode again... csrutil enable... , boot....
Open the Terminal from /Applications/Utilities.
Type sudo trimforce enable...

 

[Church13]

Open terminal in boot mode : csrutil disable, boot... open terminal in macOS: sudo nvram boot-args="vm_compressor=2" ,... sudo trim force disable , open terminal in boot mode again... csrutil enable... , boot....
Open the Terminal from /Applications/Utilities.
Type sudo trimforce enable...

Hey...while I appreciate the advice, I want to put out a general statement to anyone reading this that it's NEVER a good idea to just type random commands into terminal, especially if you don't know what they do. They could be dangerous and directly damage your system if you don't have an understanding of what it will affect.

I do know what the commands above do -- but I'm not going to execute them for the following reasons. What you're recommending is to disable OS swap. This is not good idea in the first place unless you're worried about the write endurance of your storage medium, which I understand has been an issue for some of the entry level MacBook Airs and Mac Minis. However, this is all besides the point anyway, as disabling OS swap only applies to the internal drive anyway.

Enabling TRIM is a great idea -- if it was disabled in the first place. For anyone reading, TRIM is basically garbage collection for your SSD, it helps to erase blocks that are no longer used and recalculate free space. However, TRIM is already enabled and supported for both the internal SSD and also my external SSD, and this status is reported in System Information.

 

[bzgnyc2]

Hey all, thanks for the responses!



As I mentioned in the original post, the Samsung is the second drive I've tried and it's the same issue. Previous one was an SK Hynix Gold 1TB.

I think you might be unlucky here in that the Samsung 980+ have reported problems with TRIM under recent versions of macOS and the SK Hynix Gold had other reported problems:

Buyer's Guide: Choosing a Compatible SSD for macOS Boot Drive

Buyer's Guide: Choosing a Compatible SSD for macOS Boot Drive If you're familiar with macOS, you might be aware that macOS supports a very limited number of hardware components which are natively supported on macOS, i.e. plug n play. You just cannot buy any random hardware and expect it to work...

elitemacx86.com

GUIDE - How to Enable Unsupported NVMe Drives on macOS [Samsung, SK Hynix, Micron, Intel, LITEON, PLEXTOR]

How to Enable Unsupported NVMe Drives on macOS [Samsung, SK Hynix, Micron, Intel, LITE-ON, PLEXTOR] This guide covers enabling unsupported NVMe Drives on macOS and is supported by both, i.e. Clover and OpenCore bootloaders. By following this guide, you'll be able to enable the unsupported NVMe...

elitemacx86.com

I realize the above is for Hacintosh systems and boot drives but I believe the underlying issues are still applicable. Samsung apparently did release a firmware update to address the TRIM issue for 970 EVO but has not addressed the issue in any later drive. Your PM9A1 drive is apparently an OEM version of the Samsung 980 Pro. There does appear to be a firmware update for the SK Hynix Gold (aka P31) drive to should make it work well under macOS (all else being equal...):

SK Hynix P31 NVMe macOS support possible via NVMeFix? · Issue #1662 · acidanthera/bugtracker

SK Hynix P31 does not work with MacOS because apparently Apple's NVMe drivers are causing the P31 firmware to timeout on a request, resulting in a kernel panic. Is this something that is fixable vi...

github.com

Yes, my wife has an M2 Pro MacBook Pro and it exhibits the same behavior there! I'm really thinking this is some oddity in how macOS handles external drives.

It's actually a brand new drive. Yes the model is five years old but this has only 100 power on hours, and 3 terabytes written/read in total. Sure, there's a chance of it being a faulty drive, but again, this isn't the only drive I've had the issue with. I am running Sequoia, but no encryption enabled. I've run the same test with 8 GiB blocks on AmorphousDiskMark and still getting the same values -- BlackMagic never crosses over 100MB/s read under any circumstances.

Agree additional details rule out some of the previous suggestions. At this point seems most likely a macOS, enclosure chip/firmware, SSD firmware issue. Or unfortunately some combination.

I think this is really what I'm honing in on. I suspect that the drive is being commanded to a lower power state and staying there for most read-intensive workloads, which are then bottlenecked as a result. AmorphousDiskMark, for some reason, is able to pull it out of that low power state and get the drive to perform as expected. I think macOS is poorly handling the power state management of the external drive, and real-world performance in varied I/O tasks is suffering.

I'll try to see if I can update the firmware on the drive from a Windows or *nix system. I know the Hynix drive I was using previously had the most current firmware at the time, but that's just an anecdotal observation which is no longer relevant at this point.

If your Hynix SSD has firmware after ~ 2021 then less likely its firmware at issue.

Then perhaps an issue with the enclosure firmware as you mentioned. Can't tell with Orico enclosure you have but guessing it uses the same ASM2464PD chipset as the Ugreen and my impression is they are resolving important issues with each firmware update. I have seen other reports of issues with the ASM2464PD -- it's very fast when it works but hasn't (or perhaps not until latest firmware) reached the same level of compatibility and reliability as Intel TB3 chipsets.

 

[bzgnyc2]

Update -- and this is really interesting. I have a Dell monitor with an integrated Thunderbolt 4 dock which I use as my primary external monitor. When I plug it into a standard 10Gbps USB-C port daisy-chained off the monitor/dock, BlackMagic shows more correct R/W numbers, in this case maxing out the 10Gbps link. Truly bizzare. This is with the same USB4 40Gbps cable and Orico enclosure. I'm really thinking macOS is trying to be too smart and manage the drive's power state etc when it's on the Thunderbolt/USB4 link.

In fact, under System Information, it does come up under Thunderbolt/USB4 when plugged directly into the MBP, but it comes up under USB when plugged into the monitor/dock which has a completely different arbitration method/service in macOS.

Yes agree this configuration drives a completely different set of drivers/pathways. Former tunnels PCIe through USB4 to make the SSD appear directly accessible to the system via NVMExpress while the latter connects host's USB controller to ASM2464PD through UASP and then the ASM2464PD talks to the SSD via NVMExpress.

Then seems likely some sort of macOS versus ASM2464PD issue possibly resolvable with updated firmware for the latter. That is ASM2464PD works fine when it serves those SSD up via UASP but not when it enables the more direct connection to those SSD over NVMExpress.

Hopefully not some sort of SNAFU across the entire macOS/Apple Silicon/ASM2464PD/Samsung chain...which would be a lot to work through...

 

[Parowdy]

• Run Disk Utility app "First Aid" on the drive.
• Try another cable. "Nice & thick" is fine but it can still be a bad cable or cable that has gone bad.
• Try another MB port.
• Fully power down the MB, wait a minute, power up and try again.
• Power #3 is definitely worth a try... or even a PC just to retest read/write and rule out/in that MB.
• If none of that does anything, how is the drive formatted?
• Do you have it set up to properly dissipate heat?

Isn’t restarting the new “power it off and wait”?

 

[HobeSoundDarryl]

I don't know. Power off and wait conceptually flushes all kludge. Restarting MIGHT be able to retain some of it. Or I'm applying an old fashioned concept of off-wait-on instead of simply restarting.

 

[UnifiedMelody]

I think it varies based on the NVMe drive and enclosure. Note that many if not most USB 3.x SSD drives that do ~ 1GB/sec are using NVMe drives internally and the main thermal advantage you would be getting in that case is de facto throttling. Otherwise I have not heard that the NVMe protocol over USB4/TB3+ has higher overhead than USB 3.x.

It is also possible that the first time an already written NVMe SSD is connected directly to a system (i.e. via USB4/TB3+), the system will initiate a large set of TRIM operations that could cause the drive to run hot until complete.

In any case I agree one drawback with buying a separate enclosure and NVMe drive is that one essentially becomes a system engineer responsible for among other things the heat management. Often people select their SSD based on a peak speed and/or good review but these often do not take into consideration heat and other criteria that are critical to the total solution. Tom's Hardware reviews of SSD usually include not just peak performance, but sustained write performance and power efficiency, particularly important for the external enclosure situation:

WD Black SN850X SSD Review: Back in Black (Updated)

The WD Black SN850X is one of the fastest drives around for gaming.

www.tomshardware.com

Also this database that can be helpful getting all the specs for a particular drive or finding drives that meet key criteria:

TechPowerUp

Solid-state-drive database with specifications for products launched in recent years. Includes specs, photos, and technical details.

www.techpowerup.com

Still one runs the risk that an otherwise seemingly good drive doesn't work in a particular case like the issues with TRIM on Samsung 980+ drives under macOS Monterey+. In other cases, there are reports of drives not properly handling NVMe power states until the firmware is upgraded.

Ah yes, I had a spare 2TB 970 Evo Plus which i was going to put in my PC but never got around to doing it. [This is only PCIE3.0 lol at best]. So now I'm like I'm migrating to mac, F***, might as well toss it into an enclosure since i got nothing to lose... [the pc has a 500GB and 1TB NVMe in it but i cbb buying 'new components' as it would set me back nearly 3k++ just to change the mobo and CPU and RAM [diff compatibility esp ram now in DDR5/6+ god knows i lost track]]

For now I think I just use T9 4TB lol when it arrives as a standalone external SSD, its no speed machine, I'm aware its not gonna get 2000MB/s but 1000MB/s is like plenty for me. NVMe I felt you gotta throw it into a unit that will have dedicated cooling... Even my pc nvme are standalone with no heatsink on it... two of them

 

[bzgnyc2]

I don't know. Power off and wait conceptually flushes all kludge. Restarting MIGHT be able to retain some of it. Or I'm applying an old fashioned concept of off-wait-on instead of simply restarting.

Not sure, this person says a shutdown, wait, reboot is like an old-fashioned SMC reset but the details are sketchy:

https://support.apple.com/en-us/102605?displayMode=headless#silicon

how to reset pram on MacBook Pro with App… - Apple Community

discussions.apple.com

Otherwise, if you want to reset an Apple Silicon Mac, about the best you can do is start in Safe Mode, which:
"Clears some system caches, including font caches and the kernel cache. These are automatically created again as needed."

Start up your Mac in safe mode - Apple Support

To help determine whether an issue is caused by software that loads as your Mac starts up, you can start up your Mac in safe mode.

support.apple.com

We lost a lot of those "secret" startup key combinations:

Mac startup key combinations - Apple Support

Learn about the Mac features and tools that you can access by holding down one or more keys during startup.

support.apple.com

How to reset NVRAM on an M1 or Intel Mac

Resetting NVRAM can help when something is messing with a Mac, but on Macs with M1 Chip the process is different

www.macworld.com

 

[StoneJack]

Hey all, thanks for the responses!



As I mentioned in the original post, the Samsung is the second drive I've tried and it's the same issue. Previous one was an SK Hynix Gold 1TB.



Yes, my wife has an M2 Pro MacBook Pro and it exhibits the same behavior there! I'm really thinking this is some oddity in how macOS handles external drives.



It's actually a brand new drive. Yes the model is five years old but this has only 100 power on hours, and 3 terabytes written/read in total. Sure, there's a chance of it being a faulty drive, but again, this isn't the only drive I've had the issue with. I am running Sequoia, but no encryption enabled. I've run the same test with 8 GiB blocks on AmorphousDiskMark and still getting the same values -- BlackMagic never crosses over 100MB/s read under any circumstances.



Done all the basic due diligence before posting here (steps you listed + more). I'm still unwilling to accept a cable issue given that I get the speeds I expect during basic (large) file transfers and in AmorphousDiskMark. In System Information I can see that it's linking at the full expected 40Gbps. Heat is also not an issue, both the Ugreen and Orico enclosures are effectively slabs of aluminum which the drive transfers heat to through a thermal pad, and they are actively cooled with built-in fans.



Like I responded to HobeSoundDarryl, it's definitely not a cooling issue (both passive and active cooling in enclosures), though at this point I'm strongly considering just getting a bespoke external drive like a T9 which is purpose-built.



I think this is really what I'm honing in on. I suspect that the drive is being commanded to a lower power state and staying there for most read-intensive workloads, which are then bottlenecked as a result. AmorphousDiskMark, for some reason, is able to pull it out of that low power state and get the drive to perform as expected. I think macOS is poorly handling the power state management of the external drive, and real-world performance in varied I/O tasks is suffering.

I'll try to see if I can update the firmware on the drive from a Windows or *nix system. I know the Hynix drive I was using previously had the most current firmware at the time, but that's just an anecdotal observation which is no longer relevant at this point.

change samsung and hynix

 

[AlaskaMoose]

More than likely what I am to say won't make a difference to the OP, but I wonder if it's worth a try?

If you are using the right cable to connect the external SSD, how about moving the SSD as far as possible from your Mac, and then give it another go?

 

[Church13]

I've updated the firmware on both of the enclosures, and still no joy. I now genuinely believe that it's an incompatibility with macOS and its ability to manage the drives (specifically the Samsung and Hynix controllers/firmware). Still trying to find some more recent firmware for the PM9A1, but not optimistic. Going to pick up a WD SN770 tomorrow and see how it handles a tried and true drive.

 

[bzgnyc2]

I've updated the firmware on both of the enclosures, and still no joy. I now genuinely believe that it's an incompatibility with macOS and its ability to manage the drives (specifically the Samsung and Hynix controllers/firmware). Still trying to find some more recent firmware for the PM9A1, but not optimistic. Going to pick up a WD SN770 tomorrow and see how it handles a tried and true drive.

Those have generally worked though I would get the WD SN850/SN850X if available for an extra $30 if possible. The SN770 is a DRAM-less / Host Memory Bus (HMB) design that aren't ideal for use in external enclosures.

Not sure if Samsung has released updated firmware for the PM9A1 / 980 Pro and even if it does if it will address your issue (more than one person has reported issues with Samsung 980+ drives due to compataiblity issues around TRIM starting in Monterey and it appears Samsung is not going to fix it so those people had to switch to WD).

You should be able to find firmware for your Hynix P31 SSD though. You mentioned it was already updated and not sure if an even later firmware would address your issue but that is the one last thing I would try.

 

[Church13]

Those have generally worked though I would get the WD SN850/SN850X if available for an extra $30 if possible. The SN770 is a DRAM-less / Host Memory Bus (HMB) design that aren't ideal for use in external enclosures.

Not sure if Samsung has released updated firmware for the PM9A1 / 980 Pro and even if it does if it will address your issue (more than one person has reported issues with Samsung 980+ drives due to compataiblity issues around TRIM starting in Monterey and it appears Samsung is not going to fix it so those people had to switch to WD).

You should be able to find firmware for your Hynix P31 SSD though. You mentioned it was already updated and not sure if an even later firmware would address your issue but that is the one last thing I would try.

See, that's interesting you say that -- I don't think I'm overly concerned about lack of DRAM since I won't be heavily loading it with random writes. But that raises a more interesting question of whether the external SSD can actually utilize HMB. I know that while connected via UASP external drives cannot use HMB for write caching, but is it possible that the TB4/USB4 link allows it since it's handling the drive through an NVMe interface? Worth exploring, but I don't know that I care enough to go buy two SSDs.

I haven't been able to find Samsung firmware, and don't know that I will be able to since it's an OEM-only drive. The Hynix SSD is now deployed in a server and I don't plan to pull it, but the last time I used it (about 6 months ago) it had the latest firmware as of then.

 

[bzgnyc2]

See, that's interesting you say that -- I don't think I'm overly concerned about lack of DRAM since I won't be heavily loading it with random writes. But that raises a more interesting question of whether the external SSD can actually utilize HMB. I know that while connected via UASP external drives cannot use HMB for write caching, but is it possible that the TB4/USB4 link allows it since it's handling the drive through an NVMe interface? Worth exploring, but I don't know that I care enough to go buy two SSDs.

It's my understanding that HMB doesn't work when connected to the host via Thunderbolt 3. I assume the same holds for USB4/TB4. The SSD still works as an NVMExpress drive -- but it goes into a fallback mode. My impression from others is that performance is still good (i.e. better than USB 3.x) and not sure how much penalty say an SN770 in an TB5 enclosure would take relative to being directly in a host with OS support for HMB (e.g. Windows and Linux) for which type of operations. Or if there are other drawbacks (e.g. more wear on the NAND). But I've yet to see any complaints from SN770 owners.

In any case I know what you mean -- I am also curious to test a bunch of different NVMe representative of different designs with a the different chipsets connected to different hosts to see what makes a difference, but I am not so motivated to so that I am put out the actual time and money that would require...

I haven't been able to find Samsung firmware, and don't know that I will be able to since it's an OEM-only drive. The Hynix SSD is now deployed in a server and I don't plan to pull it, but the last time I used it (about 6 months ago) it had the latest firmware as of then.

Yes I don't believe you'll be able to get firmware updates for the Samsung PM9A1 as a non-OEM purchaser. OEMs probably get not just the firmware but a toolkit that let's that tweak different firmware parameters for their particular needs. Anyway, you could try pushing a Samsung 980 Pro firmware onto that SSD as they appear identical:

Samsung PM9A1 1 TB Specs

1 TB, TLC, PCIe 4.0 x4, M.2 2280

www.techpowerup.com

Samsung 980 PRO 1 TB Specs

1 TB, TLC, PCIe 4.0 x4, M.2 2280

www.techpowerup.com

though Samsung may block that from working for market segmentation reasons. Then even if it works it may not fix your issues as I've heard other people having trouble with recent Samsung (after the 970 EVO) even with their latest firmware.

 

[Church13]

change samsung and hynix

Sometimes the simple answer is the best answer. I think in general you'd get more traction with your input if you did more than just post it as a lowercase one-liner though, lol.

It's my understanding that HMB doesn't work when connected to the host via Thunderbolt 3. I assume the same holds for USB4/TB4. The SSD still works as an NVMExpress drive -- but it goes into a fallback mode. My impression from others is that performance is still good (i.e. better than USB 3.x) and not sure how much penalty say an SN770 in an TB5 enclosure would take relative to being directly in a host with OS support for HMB (e.g. Windows and Linux) for which type of operations. Or if there are other drawbacks (e.g. more wear on the NAND). But I've yet to see any complaints from SN770 owners.

In any case I know what you mean -- I am also curious to test a bunch of different NVMe representative of different designs with a the different chipsets connected to different hosts to see what makes a difference, but I am not so motivated to so that I am put out the actual time and money that would require...

I ended up picking up an SN850X, found a good sale and bought it for $15 more than a 1TB SN770, so definitely worth it in that regard. And of course, as soon as I formatted it and gave 'er a test:

Darned impressive numbers, maxing out the USB4 link for sure. Apple and macOS really just like the WD nvme controllers. Which brings me to what I've discovered is incompatible about the Samsung (and potentially the Hynix drive as well, though as we've discussed it's not confirmed whether I truly tested it with the latest firmware or not).

Yes I don't believe you'll be able to get firmware updates for the Samsung PM9A1 as a non-OEM purchaser. OEMs probably get not just the firmware but a toolkit that let's that tweak different firmware parameters for their particular needs. Anyway, you could try pushing a Samsung 980 Pro firmware onto that SSD as they appear identical:

Samsung PM9A1 1 TB Specs

1 TB, TLC, PCIe 4.0 x4, M.2 2280

www.techpowerup.com

Samsung 980 PRO 1 TB Specs

1 TB, TLC, PCIe 4.0 x4, M.2 2280

www.techpowerup.com

though Samsung may block that from working for market segmentation reasons. Then even if it works it may not fix your issues as I've heard other people having trouble with recent Samsung (after the 970 EVO) even with their latest firmware.

Yeah, no luck with the firmware, but what I have figured out is why it hates the Samsung drive especially. It all has to do with power state management...which macOS is not doing a great job with. It appears to be going into a low power state when not in use. Naturally, the writes "wake" the drive up, but then it returns to the low power state and the reads absolutely suffer.

AmorphousDiskMark loads the drive differently, and keeps it in the high power operating state. Blackmagic and regular use just don't put the drive in a high enough power / activity state for reads to perform where they should.