Getting slow NVME write speeds

[magnomickey]

Thanks, I did have that enabled tho. I really hope samsung releases driver soon and it fixes issues.

I saw this tho... could it be related? Since I have a 500GB drive, i dont have 324 gb free space but 269 GB... my speed goes to around the 1,000 stated tho... any thoughts? Is there any tool i can try to tune performance of the drive?

 

[lexarflash8g]

YEs when I go to Device Manager and right click.. Write caching is enabled.. ill send a screenshot make sure its correct

 

[heartog]

Same issue here with the writes with the 500GB 980 Pro w/ Asus Prime X570-Pro, I currently have around 40% of the drive filled and getting 6xxx MB/s read 1xxx MB/s write. I tested the disk when it was brand new and empty on a B550 board and it was running at full speeds. I suspect it might have to do with X570 or some kind of program/service writing to the disk causing it to not clear the SLC cache properly?

 

[Amby]

Same here with a ASUS TUF GAMING X570-PLUS . Dissapointing

 

[theonek]

apprentely amd chipset driver issue, wait for fixing it, that's only thing for now.... intels don't have this as they still use gen 3.0 pcie for nvme drives.

 

[mhb]

I have built PC with following specs and have same slow write speed with my 980 Pro 500GB Model

ASUS Rog Strix X570-E Gaming
AMD Ryzen 5950X
AMD Radeon 6800 XT

 

[acegutta1]

I got a x570 and a 5900 X rise and processor and I'm getting 6700 read and 2500 write I migrated all my info on the drive I was going to wipe the drive clean and start over but if you're saying they need to have a new update I might just wait

 

[Hardcore Games]

Samsung claims 7000 read, 5000 write so it should be close to that

wipe the SSD clean before installing windows fresh

 

[heartog]

reinstalling and secure erasing the drive will help temporarily, but after a few benchmark tests/few days, it would be slow all again

apprentely amd chipset driver issue, wait for fixing it, that's only thing for now.... intels don't have this as they still use gen 3.0 pcie for nvme drives.

I tested it on a intel board, same issue

 

[hardkod]

Same issue with 2x980Pro Raid0 on Gigabyte x570 Aorus Pro. Reinstalling chipset drivers helps sometimes but not for long. Gigabyte support tell me - "its not our problem. We use amd unchanged code."

 

[igorp]

I am running Ryzen Threadripper 3990X on Asrock TRX40D8-2N2T featuring 2 x Corsair Force MP600 in raid 0. I would also expect better speed then this:

dd if=/dev/zero of=testALT.img bs=100M count=11 conv=fdatasync
1153433600 bytes (1,2 GB, 1,1 GiB) copied, 1,68589 s, 684 MB/s

Single Samsung 970 nvme 2tb on a Pentium D NAS:
1153433600 bytes (1,2 GB, 1,1 GiB) copied, 1,65157 s, 698 MB/s

Running Ubuntu 20.04 with stock kernel 5.4.0-56-generic. Tried also most recent kernels and no change ... but sometimes it boots up in high speed mode. Strange.

 

[bakabek]

Hello.
Faced the same shit. After a week of use the speed dropped to 1000.

MB: Asus Rog Strix B550-F (Wifi)

 

[L3XANDR0]

I'm having the same exact write issue. Can't go over 2100MB/s in sequenital. I have an Asus Dark Hero x570 mobo, Ryzen 5900X CPU, BIOS 3003 (latest), and latest chipset installed. The 980 Pro SSD drivers are also up-to-date. What the heck is going on?

 

[heartog]

I'm having the same exact write issue. Can't go over 2100MB/s in sequenital. I have an Asus Dark Hero x570 mobo, Ryzen 5900X CPU, BIOS 3003 (latest), and latest chipset installed. The 980 Pro SSD drivers are also up-to-date. What the heck is going on?

but 980 pro doesn't have their own "drivers" tho

 

[Vya Domus]

I've noticed something when using CrystalDiskMark, if I set the benchmark to run just 1 loop the write speeds are really low but if run the benchmark for multiple loops, like say 5, I get the expected speeds. When I copy files around manually I also get the right kind of speeds for my drive. So I wonder if there is something in particular about how these benchmarks run.

If I look at what happens in task manager when CrystalDiskMark runs indeed it never actually reaches the peak speeds in the first loop but it does in subsequent ones, I bet there is either some kind of caching involved or some timing issue that causes this.

These NVME drives are so fast that they can now move multiple GB/s of data, I think these benchmarks either move too little data to accurately measure write speeds or the intervals are too small to account for the time it takes for them to reach their peak speeds. Because the thing is writing always has a much bigger latency that reading from memory.

 

[pierce3215]

I have noticed the same issue with no resolution. Coincidentally, I captured a video recording of the performance right after I purchased it, and now have posted a video with the performance loss.
The time in between the advertised speeds and now much slower write speeds is about 2.5 months.

Initial Release - Advertised Speeds Screen Capture

12/28/2020 - Performance Drop in Write Speeds

One major thing I noticed when running CrystalDiskMark today was the huge response time differential in write speeds vs. read speeds. You can see this in the second video at the bottom left of the screen capture (task manager).

 

[5950xSlowNvme]

Created a new account to chime in.

Exactly the same problem with my 980 Pro 500GB as the OS drive. However, my 980 Pro 1TB and WD Black SN850 have no such issues. They're all on the onboard M2 slots.

I've also seen a massive drop on my Sabrent Rocket 2TB, running off an M2 Expander card. However, my Samsung 970 1TB on the same M2 Expander card has no problems and is consistent with prior checks.

I'm running an MSI MEG Godlike 570 mobo, with PCI gen settings set.

Whatever is causing this, if it is software related, appears to be targeting specific models only rather than a blanket issue across all NVMEs. I'm about to reformat the Sabrent and see how that looks after.

Edit: Regards the Sabrent, it was a about 60gb off its cap when the above issue was found. I've since cleared it back and its now back to running normal speeds. Might be coincidence but I'm loading it back up now to see how it goes.

However, the 980 is way off its cap, at 329gb free.

Cant seem to edit my prior post but I've now gone through a secure erase of the 980 and redeployed an image, its no different.

So either the hardware itself is failing or there's a chipset driver issue that's only affecting the 980 500GB model.

 

[Zach_01]

Cant seem to edit my prior post but I've now gone through a secure erase of the 980 and redeployed an image, its no different.

So either the hardware itself is failing or there's a chipset driver issue that's only affecting the 980 500GB model.

You know, if there is a software (OS, driver) issue, applying an image will not erase it. Maybe you need a clean install of windows. Unless the image was made when SSD was fast.

 

[5950xSlowNvme]

You know, if there is a software (OS, driver) issue, applying an image will not erase it. Maybe you need a clean install of windows. Unless the image was made when SSD was fast.

Yeah, I realise that but I'm working through potential issues. Secure erase gets the drive back as close to factory as possible, at the hardware level, so I wanted to see if that could make a difference.

Next steps are a new 980 arriving tomorrow. I'll swap out, reimage and see what that looks like. If that's no different, it's an OS reinstall as a last ditch.

I do suspect its the drive that's at fault though, because I also have a 980 1TB as a secondary and that's consistently performant. Windows is flagging the driver used for both the 980s as identical.

 

[yotano211]

Just return it, the ssd is broken.

 

[5950xSlowNvme]

You know, if there is a software (OS, driver) issue, applying an image will not erase it. Maybe you need a clean install of windows. Unless the image was made when SSD was fast.

Interestingly, I just checked back through my backups and I do have an image of the drive when it was working properly (8th December). So before the drive swap and OS reinstall, I'll be trying that too.

You know, if there is a software (OS, driver) issue, applying an image will not erase it. Maybe you need a clean install of windows. Unless the image was made when SSD was fast.

Well, well, well!

Without any further changes made beyond a secure erase, reimage and then just leaving the PC running for a while, its now back to normal. It didn't work immediately after the reimage (no history, as I cancelled the test before it ended).

Edit: Ran a couple more and grabbed the temp data from the run too. I did have the side off the case at the point when it started to run correctly but put it back on for the subsequent two runs, to ensure it wasn't a temperature issue (and it seemingly is all fine going by hwinfo).

 

[pierce3215]

@5950xSlowNvme

So you solely performed a secure erase and all is now back to factory performance. I have no idea what the culprit would be then...

 

[yotano211]

Just return the piece of junk back to the reseller, no need to waste more time on this.

 

[ThrashZone]

Hi,
Yep those sammy's are toasty aren't they
Too toasty for me frankly I added a blower heatsink on mine now a lot better

 

[5950xSlowNvme]

@5950xSlowNvme

So you solely performed a secure erase and all is now back to factory performance. I have no idea what the culprit would be then...

Yep but I think there's more to it than that, also.

Earlier today, after installing a load of content to the 980 1TB drive, gaming felt sluggish. I checked it and lo and behold, it too had dropped in write speed, as per the 500gb unit. However, shortly after that, it was fine again. See attached.

So it's not just the 500gb 980 that I've now had the write issue with. Looking further into it, I found this 980 review...

Samsung 980 Pro M.2 NVMe SSD Review: Redefining Gen4 Performance (Updated)

Powered by a new 8nm NVMe SSD controller and the company’s V6 V-NAND.

www.tomshardware.com

And found this...

Sustained Write Performance and Cache Recovery
Official write specifications are only part of the performance picture. Most SSDs implement a write cache, which is a fast area of (usually) pseudo-SLC programmed flash that absorbs incoming data. Sustained write speeds can suffer tremendously once the workload spills outside of the cache and into the "native" TLC or QLC flash. We use iometer to hammer the SSD with sequential writes for 15 minutes to measure both the size of the write cache and performance after the cache is saturated. We also monitor cache recovery via multiple idle rounds.

You have to be careful when filling up the 980 Pro because it will do so quicker than you might expect. Samsung's 980 Pro sustained 100GB of writes to its TurboWrite cache at a rate of 4.5 GBps before performance degraded. Once its cache filled, write speed averaged 1 GBps until full, outwriting the competition. Cache recovery is on the slower side, however. While the static 6GB cache recovered immediately, the 'Intelligent' dynamic cache did not recover within a half-hour idle window.

See bolded areas, particularly.

This is pretty consistent with what I've seen today, with the 980 1TB. I spent about 2 hours downloading a bunch of games to it, about 400gb worth, which then saw the write performance drop. After quitting the games and leaving the PC for 5 mins or so, writes were back to normal.

With the 980 500gb (OS), a lot of what I've tried to date has revolved around reformatting and reimaging the drive, which showed no difference. I went through a regular format, a few times, and then a secure erase and format, which showed no improvement after the PC was back up and running. Obviously this all involved a bucketload of write activity, to the tune of about 350gb each time. However, leaving the PC idle for about 2 hours or so after the secure erase and reformat/reimage, saw the write performance back to normal (see my post above).

I believe that after a sustained amount of write activity to the drive, the cache becomes full which reduces write speeds down to about 1/3 or 1/4 of expected, until the cache is cleared. If you look back through this thread, you'll see that someone else mentioned they had the problem and then it mysteriously fixed itself after a while.

I'm going to test this again tomorrow and see if I can replicate the behaviour but if you have the problem currently, I'd suggest just leaving your PC on and idle for a while, give it a couple of hours and see what it looks like.

More to follow.

Edit: No further instances of write speeds dropping today. No major writes performed, just reboots and playing a few games.