Upgrading 2013-2015 Macbook Pro SSD to M.2 NVMe

[kvic]

For people interested in a new WD SN550 1TB, it's currently available on Amazon UK for 69GBP

I think I haven't seen such low price in pounds..lol

 

[kvic]

Note that 2015 13-inch Macbook Pro came with dual core (four threads) processors. From my quick bench using fio, SN500's random 4k read is about 350k IOPS (close to WD's spec 410k). Given the machine under test is only dual core, I would think 350k is decent. The random 4k write is a miserable 50k IOPS (far below WD's spec 405k). Given the core counts, I give WD the benefit of doubt though I have expected way higher than 50k.

This has been spinning in my head for a long while. Finally I solved the riddle - why random 4k write is so miserable in macOS i.e. way below what the hardware is capable of. In fact, SN550 is innocent here.

I believe the issue applies to all 3rd-party NVMe drives in macOS and the problem is two folds - the filesystem and the NVMe driver in macOS.

The 50k IOPS (~200MB/s) was run on SN550 formatted as APFS. If formatted as HPF+, the same test should achieve double speed ~100k IOPS (~400MB/s). But that's still only about a quarter of what SN550 is capable of. Here comes the suspicion in the NVMe driver. I believe Apple does (perhaps deliberately) not optimise the NVMe driver for 3rd-party m.2 drives (as Apple's SSDs don't use it). I believe for majority users of this thread, it's "non-issue" as everyday laptop workloads will hardly hit Apple's cap.

Worth pointing out forum members saw the same issue and did some digging in the following thread: https://forums.macrumors.com/threads/cmp-slow-afps-4k-write-performance.2166560/. The key message is very clear: under the same test condition (CPU, SSD, benchmark s/w), HPF+ performs better than APFS on 3rd-party NVMe SSDs.

So about SN550, one con's less.

Peace

 

[phybersplice]

For people interested in a new WD SN550 1TB, it's currently available on Amazon UK for 69GBP

I think I haven't seen such low price in pounds..lol

I noticed it too on Amazon.ca.
Was $115 CDN last week.
$119 CDN this week.
Still a good savings.

 

[phybersplice]

20H2 keeps BSOD at 93% then gets rolled back. Trying everything in the toolkit to update. Running out of options. If something works I will post an update.
this is proving to be quite challenging as the BSOD flashes up for a second and reboots. I can’t see why it failed. Anyone know where the crash logs are kept?
Once I figure out how to update Windows I will then figure out which drives to send back to Amazon.
I have on hand:
MP510
P5
8200 pro is coming today.
Samsung 970 Evo Plus (sending this back due to drawing too much power).

 

[ftaok]

Seems to be fine, I would do it the same way. As for the latest Boot-ROM version number, you can check this website:

Which firmware should your Mac be using? (version 5)

This article lists the firmware versions of Macs which have been successfully updated to run macOS 11.7.10. Apple doesn’t provide an official list of the current firmware versions which shoul…

eclecticlight.co

For the moment it looks that installing Mojave will give you the latest Boot-ROM for your Mac as well. Just in case.

I hope Updating to Mojave will get me to the latest BootROM. I just tried creating a Big Sur USB installer and it failed miserably. I don't know what happened, but downloading Big Sur (and Catalina as well) from the Mac App Store does not seem to download the full installer. I was expecting a 12.6 GB app, but instead I got a 28MB app. Opening the app, it goes right to the Big Sur installer, but the Terminal command doesn't seem to recognize it as an app that can used to create a bootable drive.

 

[macpro_mid2014]

I hope Updating to Mojave will get me to the latest BootROM. I just tried creating a Big Sur USB installer and it failed miserably. I don't know what happened, but downloading Big Sur (and Catalina as well) from the Mac App Store does not seem to download the full installer. I was expecting a 12.6 GB app, but instead I got a 28MB app. Opening the app, it goes right to the Big Sur installer, but the Terminal command doesn't seem to recognize it as an app that can used to create a bootable drive.

You can use the link on Mr. Macintosh's website to download the installer: https://mrmacintosh.com/macos-big-sur-full-installer-database-download-directly-from-apple/

Save it to Downloads, run it and you will get the Install macOS Big Sur.app on /Applications folder.
You can now use createinstallmedia from the Install macOS Big Sur.app subfolder

 

[Audit13]

20H2 keeps BSOD at 93% then gets rolled back. Trying everything in the toolkit to update. Running out of options. If something works I will post an update.
this is proving to be quite challenging as the BSOD flashes up for a second and reboots. I can’t see why it failed. Anyone know where the crash logs are kept?
Once I figure out how to update Windows I will then figure out which drives to send back to Amazon.
I have on hand:
MP510
P5
8200 pro is coming today.
Samsung 970 Evo Plus (sending this back due to drawing too much power).

How to Stop Windows From Restarting After a BSOD

1. Open the Advanced System Settings menu. The easiest way to get there is by searching for "View advanced system settings" and clicking the icon which appears. ...
2. Click Settings under "Start Up and Recovery" on the Advanced tab. ...
3. Uncheck Automatic Restart. ...
4. Click Ok.

 

[kieu]

I'm kinda confuse. There's 2 charts for each of the comparison categories for SSDs(2020-02 and 2020-12) in the original post. The 2020-12 is the latest one right?

 

[ftaok]

OK, so I got my MBA6,2 onto the latest BootROM and I'm still going to run High Sierra for the time being. I'd like to change my Hibernate/Standby modes back to the default setting, but I'm running into a snafu. I don't know what the default settings were.

Using pmset -g |grep hibernatemode, I get
hibernatemode 0

Using pmset -g | grep standby, I get
standbydelay 10800
standby 0

Using pmset -g | grep autopoweroff, I get
autopoweroffdelay 28800
autopoweroff 0

Can anyone point me to where I can find the default values?

Thanks.

UPDATE - thanks to @kvic for the solution of using System Preferences/Energy Saver. Hadn't even thought of it. Anyway the default settings are as follows.

hibernatemode 3

standbydelay 10800
standby 1

autopoweroffdelay 28800
autopoweroff 1

 

[olfo]

This has been spinning in my head for a long while. Finally I solved the riddle - why random 4k write is so miserable in macOS i.e. way below what the hardware is capable of. In fact, SN550 is innocent here.

I believe the issue applies to all 3rd-party NVMe drives in macOS and the problem is two folds - the filesystem and the NVMe driver in macOS.

The 50k IOPS (~200MB/s) was run on SN550 formatted as APFS. If formatted as HPF+, the same test should achieve double speed ~100k IOPS (~400MB/s). But that's still only about a quarter of what SN550 is capable of. Here comes the suspicion in the NVMe driver. I believe Apple does (perhaps deliberately) not optimise the NVMe driver for 3rd-party m.2 drives (as Apple's SSDs don't use it). I believe for majority users of this thread, it's "non-issue" as everyday laptop workloads will hardly hit Apple's cap.

Worth pointing out forum members saw the same issue and did some digging in the following thread: https://forums.macrumors.com/threads/cmp-slow-afps-4k-write-performance.2166560/. The key message is very clear: under the same test condition (CPU, SSD, benchmark s/w), HPF+ performs better than APFS on 3rd-party NVMe SSDs.

So about SN550, one con's less.

Peace

Not sure if this relates, but I have just compared APFS performance to HFS+ performance. If you have some space on your SSD for these tests, it is really easy to create a HFS+ partition alongside with APFS.

Instructions here:

APFS & HFS Partitions on same Disk? - Apple Community

discussions.apple.com

Then I ran AmorphousDiskMark on the APFS and on the HFS+ partition.

APFS partition:

HFS+ partition:

The SSD is the 2TB Crucial P2.

As an interesting side note: I could even create a dual boot setup. My Macbook Pro ran MacOS 10.13 on the HFS+ partition and MacOS 10.14 on the APFS volume. Booting from MacOS 10.12 wouldn't complete, though.

The results were more or less the same with either High Sierra and Mojave as boot volume.
APFS was faster here. But when I did a real-world test, measuring the speed with a huge file in Photoshop, the speed difference wasn't really noticeable.

 

[kvic]

Can anyone point me to where I can find the default values?

The results were more or less the same with either High Sierra and Mojave as boot volume.
APFS was faster here. But when I did a real-world test, measuring the speed with a huge file in Photoshop, the speed difference wasn't really noticeable.

I did the SN550 tests in Catalina. Recently did it in Big Sur with ADATA SX8200 Pro and same observation. Actually that was the moment I started rethinking what's going on. I believe ppl in the thread that I linked to above did it in Mojave & Catalina.

Btw, do you think your HPF+ random 4k write is exceptionally low? I couldn't find official P2 specifications on random 4K. Perhaps you need to boot into Windows To Go or Linux Live USB to estimate P2's limits on random 4k.

One thing can't emphasise once more: it's macOS that limits random 4K write performance to its fullest.

 

[olfo]

I did the SN550 tests in Catalina. Recently did it in Big Sur with ADATA SX8200 Pro and same observation. Actually that was the moment I started rethinking what's going on. I believe ppl in the thread that I linked to above did it in Mojave & Catalina.

Btw, do you think your HPF+ random 4k write is exceptionally low? I couldn't find official P2 specifications on random 4K. Perhaps you need to boot into Windows To Go or Linux Live USB to estimate P2's limits on random 4k.

One thing can't emphasise once more: it's macOS that limits random 4K write performance to its fullest.

Hmm, if I understand you correctly:
You suppose that 4K write performance with NVMe drives is lower on MacOS than on Windows 10?
I think a tool to compare Windows with MacOS is needed. The only tool that I know of is AmorphousDiskmark, which aims to make the results comparable to CrystalDiskmark in Windows.

The only Macbook Pro that offers a true chance to do a fair SSD comparison between Windows and MacOS seems to be the 15" 2015 model aka the MacBookPro11,4.

I only have the late 2013 15" MacBookPro11,3 – which has a slower SSD performance per se.
I can offer some humble thoughts about that Mac at least.
So – when I look at the numbers on the screenshots from the AmorphousDiskmark website (showing my original 500GB Apple AHCI SSD), compared to my screenshots above, I can clearly see that the Crucial P2 is faster anyhow.

AmorphousDiskMark for macOS - measures storage read/write performance

AmorphousDiskMark measures storage read/write performance in MB/s and IOPS.

www.katsurashareware.com

On the UserBenchmark website you may find a lot of data from different NVMe SSDs in different Mac models.
All tests are only done on Windows 10 though.
But anyway, maybe that's a good point to start?
For example, I can look at the original MacBookPro11,3 Apple SSD:

Please click the green human to continue - UserBenchmark

And compare it to the Apple 500GB SSD pulled from the MacBookPro11,4 now put inside the MacBookPro11,3:

Please click the green human to continue - UserBenchmark

Compare this to the MVMe ADATA SX8200PNP 1TB:

Please click the green human to continue - UserBenchmark

The ADATA performs a lot better in the MacBook Pro than all Apple SSDs.
Except 4K write.
This is not typical for this SSD under Windows 10 in PCs:

UserBenchmark: Nvme ADATA SX8200PNP 1TB

ssd.userbenchmark.com

But the Crucial P2 performs better in a PC, so it will probably do so in the MacBookPro11,3 :

UserBenchmark: Crucial P2 3D NVMe PCIe M.2 2TB CT2000P2SSD8

ssd.userbenchmark.com

I think it should be possible to find UserBenchmark tests for the MacBookPro11,4 – and get closer to an answer to your initial question.

P.S.:
Out of curiosity, I have checked on some UserBenchmark results for the MacBookPro11,4.
It should score highest of all Macbook Pros, because it has the fastest SSD lane.
Here are two benchmarks with the WD SN550 1TB:

Apple MacBookPro11,4 Performance Results - UserBenchmark

www.userbenchmark.com

Apple MacBookPro11,4 Performance Results - UserBenchmark

www.userbenchmark.com

I had expected the sequencial read and write to be higher, but maybe that eGPU has slowed it down a bit?
Compared to other SSDs in this Macbook Pro:
OWC Aura Pro X2 960GB:

Apple MacBookPro11,4 Performance Results - UserBenchmark

www.userbenchmark.com

Adata XPG SX8200 Pro NVMe PCIe M.2 1TB:

Apple MacBookPro11,4 Performance Results - UserBenchmark

www.userbenchmark.com

Compared to a PC with an i7 processor:

Asrock Z590-C/ac Performance Results - UserBenchmark

www.userbenchmark.com

Compared to an iMac19,1 from 2019 with original Apple SSD:

Apple iMac19,1 Performance Results - UserBenchmark

www.userbenchmark.com

 

[Marco_2k7]

Hi there!

I've read the OP but I still have some questions about upgrading my macbook (pro retina 13" 2015) to a third party ssd.

- I'm running Big Sur 11.6, should I be ok with sleeping/hybernation issues?
- Is there any problem with importing a Time Machine backup after the upgrade?
- I seem to understand that NVMe drives use more battery...why that happens?
- My idea is to get a 500gb/1Tb Samsung (I had good experience with their ssd) NVMe drive, any suggestion on the model? I'm looking into the 970 Evo Plus or the 980, any major difference? Or if you do not reccomend Samsung, any suggestion in the 150€ price range?

Thank you so much for your help.

 

[phybersplice]

It’s a WHEA UNCORRECTABLE ERROR for Win10 when I move the drive to the 2013. Everything works in the 2015.

 

[kvic]

Hmm, if I understand you correctly:
You suppose that 4K write performance with NVMe drives is lower on MacOS than on Windows 10?
I think a tool to compare Windows with MacOS is needed. The only tool that I know of is AmorphousDiskmark, which aims to make the results comparable to CrystalDiskmark in Windows.

MacOS significantly handicaps 4k random write performance on 3rd-party m.2 NVMe drives. The problem is more serious when formatted as APFS than HPF+ (N.B. your data point is an exception to this). However, even with HPF+, 4k random write on these 3rd-party NVMe drives are less than 50% of what hardware is capable of. Also, it's unrelated to PCIe bandwidth.

Crucial doesn't publish specification on P2. As I know Linux and Windows do not have such handicaps, I was suggesting if you want to estimate your P2 random 4k specs. Run it on Windows or Linux.

The problem applies to all 3rd-party NVMe drives as I said in my previous post. Here let's use P2 for illustration again.

But the Crucial P2 performs better in a PC, so it will probably do so in the MacBookPro11,3 :
UserBenchmark: Crucial P2 3D NVMe PCIe M.2 2TB CT2000P2SSD8

Didn't read test methodologies there. We could do an estimation from the above data point regarding P2's random 4k performance

• random 4k write: 1350MB/s or 330k IOPS
• random 4k read: 1100MB/s or 270k IOPS

Compare to your result. P2 formatted as APFS: random 4k write only about 26% of what hardware is capable of. random 4k read is about 78%.

 

[olfo]

MacOS significantly handicaps 4k random write performance on 3rd-party m.2 NVMe drives. The problem is more serious when formatted as APFS than HPF+ (N.B. your data point is an exception to this). However, even with HPF+, 4k random write on these 3rd-party NVMe drives are less than 50% of what hardware is capable of. Also, it's unrelated to PCIe bandwidth.

Crucial doesn't publish specification on P2. As I know Linux and Windows do not have such handicaps, I was suggesting if you want to estimate your P2 random 4k specs. Run it on Windows or Linux.

The problem applies to all 3rd-party NVMe drives as I said in my previous post. Here let's use P2 for illustration again.



Didn't read test methodologies there. We could do an estimation from the above data point regarding P2's random 4k performance

• random 4k write: 1350MB/s or 330k IOPS
• random 4k read: 1100MB/s or 270k IOPS

Compare to your result. P2 formatted as APFS: random 4k write only about 26% of what hardware is capable of. random 4k read is about 78%.

Well, if it is legitimate to compare the numbers between AmorphousDiskMark and the Userbenchmark, I see the Crucial P2 in the MacbookPro11,3 well in the average PC Windows 10 range. Or isn't it?
I don't have windows or Linux, so I can't offer a test in these OSes. Are you sure your observations aren't hardware related? Sorry, maybe I just don't get it.
But if you want numbers that are better comparable, why don't you test your own SSD in AmorphousDiskmark on MacOS, and CrystalDiskmark on Windows and share the results with us?
After all, my main point was that the 2TB Crucial P2 is still way better than the Apple SM0512F 500GB or the Apple SM0512G PCIe 500GB. Like many other NVMe SSDs. Or so it seems to me.

 

[kvic]

My journey and conclusion on this topic actually ended with my post in #9453. I had a pretty clear grasp of the problem already. I believe I explained well there too.

I saw you seemed to be interested in exploring this topic and appeared a bit confused, I was trying to offer you some help and further clarification.

 

[luryxyt]

what is the best SSD drive for macbook pro 2017?

 

[ftaok]

what is the best SSD drive for macbook pro 2017?

What concerns you most? Speed? Price? Power consumption?

Are you intending to use BootCamp/Windows?

Which MBP 2017? If you have a Touchbar MBP, I'm pretty sure that the SSD is not removable.

 

[luryxyt]

What concerns you most? Speed? Price? Power consumption?

Are you intending to use BootCamp/Windows?

Which MBP 2017? If you have a Touchbar MBP, I'm pretty sure that the SSD is not removable.

low power consumption, and if it would have the same speed as the original apple ssd

No i'm not intending to use Bootcamp/windows, maybe only Prallax.

I have mpb a1708 128ssd.

I want to replace original SSD.

Min capacity 512Gb.

 

[ftaok]

low power consumption, and if it would have the same speed as the original apple ssd

No i'm not intending to use Bootcamp/windows, maybe only Prallax.

I have mpb a1708 128ssd.

I want to replace original SSD.

Min capacity 512Gb.

I watched a video on Youtube regarding the non-Touchbar MBP. The form factor on the SSD used is not the same as older MBP and MBA. The adapter is also different.

You should probably do a search for a forum thread that addresses the specific concerns of the 2017 MBP models. The information in this thread probably doesn't apply to your MBP.

EDIT - here's the MR thread for your MBP.

Found this adaptor for 2016/2017 non-touch bar for ssd upgrades

Has anyone heard or tried these out with a 42mm length pci-e ssd? 1587411306 Ive found 512,1tb,2tb nvme ssd that fit this with 2000+mbps read and write too. Makes the cost of making my Mac 1tb roughly 220 dollars.

forums.macrumors.com

 

[bradbomb]

Does anyone have experience with the Kingston NV1? I had ordered one last week on sale directly from Kingston to try and got an adapter from Amazon (non-sintech). I'm now waiting for a sintech adapter to see if that is at fault, but with this adapter, the computer does not detect it at all while booted from a Mojave Installer USB Drive. The computer's BootROM is updated as I was running Big Sur previously on my Apple SSD in my Mid-2014 13" MacBook Pro.

 

[kvic]

Does anyone have experience with the Kingston NV1? I had ordered one last week on sale directly from Kingston to try and got an adapter from Amazon (non-sintech). I'm now waiting for a sintech adapter to see if that is at fault, but with this adapter, the computer does not detect it at all while booted from a Mojave Installer USB Drive. The computer's BootROM is updated as I was running Big Sur previously on my Apple SSD in my Mid-2014 13" MacBook Pro.

Interesting pick.

Kingston NV1 looks like a direct competitor to Crucial P2, even down to the massaging of their specifications [1]. Using the same Phison E13 controller. Perhaps similar power consumption characteristics. So it should work.

[1] Both Kingston and Crucial conveniently hide random 4k ratings on NV1 and P2. From NV1's spec, we could guess it uses QLC chips. Let's see how fast it's after passing the fast cache. So please report back.

 

[Marco_2k7]

Hi there!

I've read the OP but I still have some questions about upgrading my macbook (pro retina 13" 2015) to a third party ssd.

- I'm running Big Sur 11.6, should I be ok with sleeping/hybernation issues?
- Is there any problem with importing a Time Machine backup after the upgrade?
- I seem to understand that NVMe drives use more battery...why that happens?
- My idea is to get a 500gb/1Tb Samsung (I had good experience with their ssd) NVMe drive, any suggestion on the model? I'm looking into the 970 Evo Plus or the 980, any major difference? Or if you do not reccomend Samsung, any suggestion in the 150€ price range?

Thank you so much for your help.

Anyone with any advice?

 

[phybersplice]

So here's my update on my venture to upgrade the 11,1 2013 Late MBPr with 1TB of capacity.

I got the Sintech short adapter from Amazon.ca for $1.99 (had a 20$ credit on my account).

I've tried the Samsung 970 Evo Plus (Battery drain too high) $189.99
Crucial P5 (Complete Disaster) $124.99
Corsair MP510 (not a bad drive, but it's an odd capacity at 960GB) $169.99
ADATA / XPG SX8200 Pro $149.99

My goal was to get a proper Bootcamp partition of Windows 10 on the drive as well.
I tried all these drives, 3 gave me various issues, all gave me BSOD during the Win10 install.
The only one that allowed me to continue with the install by doing the regedit trick in post #1 was the ADATA / XPG.
I didn't use the included heatsink.
I've managed to update Win10 to what Win10 Updater says is current 20H2.
I could manually do 21H1 but right now, it's not that important.
At the end of the day, it was one hell of an experiment.
Keeping the XPG SX8200 Pro.