Surface Book 2 SSD write speeds abysmal for some. Lots of questions, no answers.

Topic

New Reply

For the past six months, customers have been asking Microsoft about the excruciatingly slow sequential write speeds on their new Surface Book 2 PCs. I
[See the full post at: Surface Book 2 SSD write speeds abysmal for some. Lots of questions, no answers.]

1 user thanked author for this post.

satrow

Reply | Quote

Replies

Edit: is TRIM actually active (not just according to fsutil – it’s useless) on these rigs?

Reply | Quote

How’s your SSD write speed? Tell us on the AskWoody Lounge.

Blistering speed here why? I put it in my self wouldnt buy or let M$ do it for me that could be the difference. Even Win10 and Win7 hasn’t messed it up…… yet!

Reply | Quote

Just wanted to make it clear that the reports I’ve read claim this is a problem with the Surface Book 2. Some Surface Pro 4 machines ship with the same Samsung SSD, but I haven’t seen any reports of ultra-slow writes on the SP4 machines.

Reply | Quote

I am 98% certain that automatic trim is not happening on these devices.

On recent releases of Windows 10, you can set this on a schedule, or run a forced trim manually using our old friend “Defragment and Optimize Drives” application, included in the OS.

Some may not realize that this old app now supports force trimming on demand for SSD’s.

Open the app, note that if the tool does NOT see your SSD’s as solid state drives, this procedure won’t work.  (That also means that Windows thinks you’re running a spinner, which is a deeper problem that should be fixed.)

See image for an example:  the C: and E: drives are SSD’s, the D: drive in this example is an older spinner.  Defrag on Windows 1703 and up treats these drives VERY differently.  It will TRIM solid state drives, not defrag, and it will DEFRAG spinners.

 

 

This is relevant to know because in the old days the common advice was to disable the Defrag schedule for SSD’s, and I find today that many self helpers still routinely do this based on outdated advice on modern systems.

 

~ Group "Weekend" ~

1 user thanked author for this post.

johnf

Reply | Quote

When I first installed a SSD (Samsung 850 Evo 1TB) into my Core 2 Duo laptop, which of course was never designed with SSDs in mind, I was not sure that the older system’s ICH (Southbridge), firmware, and driver would automatically pass the TRIM command from the OS to the device itself.  I used Trimcheck to test it and find out whether it was.

The result was that it was, in fact, TRIMming the drive after every write.  The program (which uses the command prompt) is really simple… it writes a bunch of digits to a file on a known area on the SSD.  It then deletes the file.  Wait a bit (the author suggests about a minute; I doubt I waited anything close to that) and run the program again.  It will attempt to find the data it wrote to the SSD.  If all it finds are 0s in the place that it expected to find its data, it concludes that TRIM must have taken place.  It was happening reliably even on this old system.  I think at the time I was running Windows 7.  The SATA drivers would have been the stock Microsoft drivers.

If TRIM works transparently on such an unlikely system, it would have to be a defect of some magnitude to keep it from working on a far newer, more modern system.  That’s not to mean that can’t happen… in the Windows 10 era, you never know.  It’s one possible explanation for the reported slowing, but not the only possible one.

FWIW, my SSD in my Core 2 Duo laptop reports about 260 MB/s sequential throughput on reads and writes, about half what it would be on a SATA 3 connection, but the sequential read numbers don’t do the SSD justice in terms of getting a feel for what kind of performance increase you’d get in daily use.  It’s the 4k random read numbers where a SSD really struts its stuff, factoring in the time it takes to seek to a random sector on the drive, simulating data reads or writes to lots of small files rather than one huge one.  The throughput numbers don’t look impressive compared to the sequential 550MB/s SATA3 numbers (or the even more impressive NVMe numbers), but the 26 MB/s is nearly two orders of magnitude (100x) better than what I got with my WD Black 7500 RPM HDD (around 0.31 MB/s).

Dell XPS 13/9310, i5-1135G7/16GB, KDE Neon
XPG Xenia 15, i7-9750H/16GB & GTX1660ti, KDE Neon

3 users thanked author for this post.

Microfix, satrow, Elly

Reply | Quote

The stock W7 > W10 MS SATA drivers allow Trim but watch out for MS’s Windows Updates, they can, and do, break Trim by installing duff Intel driver packages – if you ‘lose’ Trim because of this, rollback the SATA drivers from the Device Manager and disallow MS/Windows Updates from updating hardware drivers.

Generally, if your SSD’s firmware can be updated, Trim will be active. Hard Disk Sentinel (there’s a trial version) will also detect whether Trim is actually active. As with TrimCheck and the SSD maker’s toolboxes, no Trim test software is likely to be 100% reliable, best to check with a number of methods.

NVMe SSD’s are a ‘special’ case, they will probably require drivers from the SSD maker (or perhaps a more ‘generic’ driver/toolbox like that from Crucial) to enable all ‘features’.

Do not rely on FSUTIL, it doesn’t check whether Trim is active, it only checks whether it’s enabled in the OS.

1 user thanked author for this post.

Microfix

Reply | Quote

So in essence what you are saying @Satrow is that MSFT’s own branded device drivers are killing off trim within W10?

If so, this was one of the main reasons I’ve NEVER installed MSFT device drivers unless absolutely necessary (last resort) on XP, W7 and W8.

Keeping IT Lean, Clean and Mean!

1 user thanked author for this post.

satrow

Reply | Quote

No, their own W7 era W10 SATA drivers work fine with Trim, it’s only some of the 3rd party SATA driver packages tested and supplied through MS/WU have done this. (MS tests them for what, not crashing Windows? Unless they’ve changed tactics recently, they’ve certainly not tested them for full SSD compatibility.)

Even though you’ve set W10 not to auto-upgrade hardware drivers, it pays to check at least once after each major update (6-monthly), to ensure MS hasn’t done a reset along with the upgrade.

2 users thanked author for this post.

Ascaris, Microfix

Reply | Quote

This tool is only useful for HDDs. For SSDs, VMs, et cetera, you have to use the PowerShell cmdlet Optimize-Volume.

Reply | Quote

My SB2 15″ 256GB has 2800mb/s reads and 1500mbtd writes… Its a samsung pm951 inside

1 user thanked author for this post.

woody

Reply | Quote

For reference, SSDs are defragmented by Windows, but only to keep the file fragments under control.
http://www.hanselman.com/blog/TheRealAndCompleteStoryDoesWindowsDefragmentYourSSD.aspx

cheers, Paul

1 user thanked author for this post.

NetDef

Reply | Quote

Very informative article!  Thanks for sharing the link.  I wonder if anything has changed since it was written (2014) in W10 1709/1803?

~ Group "Weekend" ~

Reply | Quote

I have a Surface Pro 4 with a SAMSUNG MZFLV256HCHP-000MV and I get the same 1500ish MB/s read and 300+ MB/s write. So this is not a SB2 only problem.

2 users thanked author for this post.

Microfix, satrow

Reply | Quote

I have a surface book 2 1tb and I get about 1300 write on mine. Though I have reinstalled and reconfigured windows on it since I got it. It is possible that through Windows updates, something might have gotten messed up. I would check trim for sure and that there are no weird settings on the controller driver.

Reply | Quote

Upgrading to 1803 should resolve the issue. As mentioned elsewhere, 9re-1803 versions, especially 1709, have serious I/O performance issues Microsoft is not going to fix.

Reply | Quote

I’ve got a Hynix SC308 256GB SSD in my Dell and ran Crystal disk app and it did not do well with write speed either. Came in at around 250 vs 560 for read. Which according to review tests falls within specs. Maybe Microsoft used a cr***y SSD in some Surfaces?

Reply | Quote

These people have no idea what they’re talking about!  “Slow” write-speed is two orders of magnitude (a factor of 100 times) slower than what they’re crying about. Try 3MB/s on a random-write to a 5400 RPM laptop drive, that’s slow! 300+ MB/s is screaming fast for writing a MLC/TLC SSD, it’s just not the artificially inflated RAM-cache speeds they are seeing for reads and not even up to the 450-ish+ speeds some drives get by operating portions of their flash as SLC write caches.

Reply | Quote

Slow is all relative. If your SSD is capable of write speeds that are nearly as fast as the read speeds, you reasonably expect your computer to achieve those speeds.

cheers, Paul

1 user thanked author for this post.

Ascaris

Reply | Quote

So whats the difference to a ssd with for example ~1500mb write speed instead of 350mb? Most of the things are reading.

Reply | Quote