960 Pro 1TB performance has lowered/become tempermental

[Hail The Brain Slug]

I received my 960 Pro 1TB about a month ago. When I first received it, it benched exactly as expected (3300-3500 read, 2100-2200 write, and all the randoms were fine). Keep in mind this was in CrystalDiskMark running a 1GiB test size.

Samsung Magician V5.0 had no trouble making the drive look excellent, either. 3500+ read, 2100+ write, 450,000+ random read and 360,000+ random write.

Fast forward to today: Randoms in both test programs seem fine. Sequentials have bombed, though.

CrystalDiskMark gets ~3000 read and ~2000 write (Not so bad), but Samsung Magician gets 2200-2500 read/1600-1800 write now.

If I bump the test size in CrystalDiskMark up to a large amount like 16 or 32 GiB, the sequentials bench reasonably (3400+/2000+), which is more confusing.

I have made zero bios changes, zero driver changes, and am running the same installation of Windows. The only difference I can figure is I have installed ~380 GiB worth of software and programs on the drive. This still leaves 570GiB of free space, so it should not affect it.

Is it a bad drive controller? Something wrong with my PC? Even more confusing that huge samples test fine, but the Samsung Magician test and smaller CDM samples are testing significantly lower than when the drive was new.

The fact that large samples test fine and small samples don't (I'm sure magician uses tiny sample sizes, which is why the test is so quick) makes me think somehow there's some latency occurring that didn't used to. That would affect small sample sizes but not so much on large ones.

Edit: A side note, I have tested it as the OS drive and booted into windows on another drive as a non-OS drive, which should eliminate any software or background process overhead on the drive time. I am also testing with the latest version of Samsung Magician and have the latest version 2.1 of Samsung's NVMe driver.

System specs:
5960X @ 4.5/4.375
Asus X99-A/USB 3.1
32GB Corsair DDR4 3000 /14-15-15-32
980ti Classified
960 Pro 1TB
850 Evo 500GB
SP550 240GB
2x HE8 8TB HDD
EVGA 1000PS

I've done the testing both in the M.2 slot and in a PCI-e riser card in all PCI-e slots with and without a heatsink, eliminating a problem with the M.2 slot, with a specific PCI-e slot, or with the riser card.

I've got half a mind to toss the drive in the riser card in my old Z87 computer and test it, to eliminate my MB as a culprit. I'll update if I do that.

Edit: Tried that but it would appear due to a chipset or NVMe support limitation the sequentials only reach ~2200MB/s on the Z87 computer. Too bad, I was hoping for some more data to point to the drive or my X99 system as the culprit.

 

[Elixer]

Which version of windows are you running exactly?

I wouldn't be concerned over benchmarking, and especially wouldn't use such huge data sets for a benchmark.

Have you noticed day to day operations slower at all? If not, don't worry about it at this time.

 

[UsandThem]

Yup, that is normal behavior as you begin to fill up a SSD (even though there is still a lot of free space). As data is added, the drive needs to extra 'maintenance'. Once they get above 75% filled, you should notice even a further decline in overall performance.

http://www.howtogeek.com/165542/why-solid-state-drives-slow-down-as-you-fill-them-up/

http://www.anandtech.com/show/6489/playing-with-op

If I owned one of those drives, I would set aside additional OP to see the difference it made as it made a difference with the 850 PRO and 850 EVO:

http://www.anandtech.com/show/9451/the-2tb-samsung-850-pro-evo-ssd-review/2

Finally, the 960 PRO with additional OP:

http://www.anandtech.com/show/10754/samsung-960-pro-ssd-review/3

After the initial period of very high performance, the 960 Pro enters a steady state with very good short-term consistency but gradual long-term variation in performance. This is more similar in character to the behavior of the Intel SSD 750 than Samsung's earlier SSDs, though it's interesting to note that the 960 Pro is more twice as fast during the initial phase before transitioning to steady state.

 

[Hail The Brain Slug]

Which version of windows are you running exactly?

I wouldn't be concerned over benchmarking, and especially wouldn't use such huge data sets for a benchmark.

Have you noticed day to day operations slower at all? If not, don't worry about it at this time.

Windows 10, latest build fresh install. I stopped with the benchmarking as I'm certain my results won't change unless I clear the drive, and I don't want to do that just for testing. I doubt I would be able to notice a change in usage since the randoms are doing well, it's just the Q32 sequentials that are slowing down.

Yup, that is normal behavior as you begin to fill up a SSD (even though there is still a lot of free space). As data is added, the drive needs to extra 'maintenance'. Once they get above 75% filled, you should notice even a further decline in overall performance.

http://www.howtogeek.com/165542/why-solid-state-drives-slow-down-as-you-fill-them-up/

http://www.anandtech.com/show/6489/playing-with-op

If I owned one of those drives, I would set aside additional OP to see the difference it made as it made a difference with the 850 PRO and 850 EVO:

http://www.anandtech.com/show/9451/the-2tb-samsung-850-pro-evo-ssd-review/2

Finally, the 960 PRO with additional OP:

http://www.anandtech.com/show/10754/samsung-960-pro-ssd-review/3

I have studied this information, but I was under the impression the no-OP tests were done with a full drive and the OP tests were with a full drive (But with 25% overprovisioned). My drive is approximately 43% full, so I have 53% (oops, 57%) open space.

The drive is filled before the start of the test, and the test duration is one hour. Any spare area will be exhausted early in the test and by the end of the hour even the largest drives with the most overprovisioning will have reached a steady state.

I admit I was used to my 950 Pro, which never faltered even filled to 80% capacity, so I was very confused at first about this whole situation.


I likely would have never even known except I purchased a KryoM.2 Add-in card heatsink for the 960 Pro, and I ran benches to make sure it didn't affect performance.

Thanks for the prompt replies, I will just assume this is status-quo for the 960. I will be keeping an eye on it periodically as I fill it up more.

 

[UsandThem]

Thanks for the prompt replies, I will just assume this is status-quo for the 960. I will be keeping an eye on it periodically as I fill it up more.

If the performance gets too unbearable, I'll trade my 850 EVO for it. 😉

Seriously, that is really nice performance, so enjoy it. I think I have finally settled on a 960 EVO for my next PC upgrade. Just gonna let the dust (and pricing and availability) settle a bit from all the new NVMe drives that have recently launched.

 

[BonzaiDuck]

If the performance gets too unbearable, I'll trade my 850 EVO for it. 😉

Seriously, that is really nice performance, so enjoy it. I think I have finally settled on a 960 EVO for my next PC upgrade. Just gonna let the dust (and pricing and availability) settle a bit from all the new NVMe drives that have recently launched.

I thought I borked my dual-boot system today using a 250GB 960 EVO in PCIE for a caching drive from an SSD and an HDD. Primocache wasn't "living up" to expectations for SSD-caching. I made a mistake in "logical volume maintenance" in "disk management." It fouled things up. Luckily, with Primocache, all I had to do was physically remove the drive, then boot into Windows (Praise GEE-zus), and delete the caching tasks, or recreate them without the 960 EVO.

XabanakFanatic -- My benchies on the EVO didn't vary much between Magician and other benchmarks. Always about 3,200 read and between 1,500 and 1,900 write. I've already chewed up 600 GB of writes to the disk. Call it "testing the water" for $150 -- $130 for the 960 EVO and ~$20 for the PCIE card.

You're better off putting these drives in a PCIE card for an x4 slot, or that was my understanding of it.

Further, the EVO may get you past 400 TBW; there were indications that the PRO would get to 1.2 PB. So if you have to Secure-Erase or otherwise re-format the drive and write piles of files to it, you have a pretty long string.

The PRO drive could be a worthwhile investment. But I was looking at Egg and BH Photo today, and there's been a chronic shortage of inventory. The price last month for pre-order was closer to $630. Now, I think they're asking over $800. It's gonna happen eventually, but I have to think about it. That's a lotta cheddar . . .

A person might be able to get something like a 600P and then use a small part of 32GB to accelerate it. But I have "Techno-Lust" as another member put it posting to some thread in the last month or two. While I continue to rock my own SP550, I'm going to watch the market for these NVMe's -- for a while. Ugh. Dis-ci-pline! Argghh!!

I'll bet you preordered that sucker at the reasonable price, though. "Preorder" and "Out of stock, 2 to 4 weeks" just turned me off while counting my money on one system and shopping on another.

 

[Hail The Brain Slug]

I swear, XavierMace is going to find this and chide me. I thought I borked my system today with carelessness over a 960 EVO 250GB I bought for an experiment.

XabanakFanatic -- My benchies on the EVO didn't vary much between Magician and other benchmarks. Always about 3,200 read and between 1,500 and 1,900 write. I've already chewed up 600 GB of writes to the disk. Call it "testing the water" for $150 -- $130 for the 960 EVO and ~20 for the PCIE card.

You're better off putting these drives in a PCIE card for an x4 slot, or that was my understanding of it.

Further, the EVO may get you past 400 TBW; there were indications that the PRO would get to 1.2 PB. So if you have to Secure-Erase or otherwise re-format the drive and write piles of files to it, you have a pretty long string.

The PRO drive could be a worthwhile investment. But I was looking at Egg and BH Photo today, and there's been a chronic shortage of inventory. The price last month for pre-order was closer to $630. Now, I think they're asking over $800. It's gonna happen eventually, but I have to think about it. That's a lotta cheddar . . .

I got the 1TB for $510 with the special discount that was available the end of October. And I am using it in a riser card. Search KryoM.2 and I have the passive heatsink version.

 

[BonzaiDuck]

I got the 1TB for $510 with the special discount that was available the end of October. And I am using it in a riser card. Search KryoM.2 and I have the passive heatsink version.

That was a shrewd move, Bro. About $32 bucks? I was in too much of a hurry with my little $150 experiment. I came to the idea of the PCIE configuration late in the game, and was managing my SATA ports. To get an x4 slot, I had to give up two. I added a cheap Marvell SATA-III PCIE x1 controller to compensate -- only needed for eSATA and an extra hot-swap bay.

But the heatsink -- wow. I think there was a white-paper Samsung published, showing the effect of lower temperature on the performance. Can you read the temperature of that drive?

 

[Hail The Brain Slug]

That was a shrewd move, Bro. About $32 bucks? I was in too much of a hurry with my little $150 experiment. I came to the idea of the PCIE configuration late in the game, and was managing my SATA ports. To get an x4 slot, I had to give up two. I added a cheap Marvell SATA-III PCIE x1 controller to compensate -- only needed for eSATA and an extra hot-swap bay.

But the heatsink -- wow. I think there was a white-paper Samsung published, showing the effect of lower temperature on the performance. Can you read the temperature of that drive?

Yeah. In a 1GiBx3 crystaldiskmark run, the bare drive got to 56C. In the KryoM.2, max 32C.

You can only really find it on websites that charge shipping, so its more like $40. It's quite well made, and it's very compact as far as heatsink cards go. The primary competition is the Angelbird Wings PX1, but that card looks huge and is covered with LEDs. The heatsink might have more mass, but the size of this one has proven to be more than sufficient for a 950 or 960 Pro.

I missed out on a sale Amazon had via a coupon to get a Plextor M8PE bare M.2 1TB for $315, otherwise I would have picked one up and thrown it into another KryoM.2.

 

[BonzaiDuck]

Yeah. In a 1GiBx3 crystaldiskmark run, the bare drive got to 56C. In the KryoM.2, max 32C.

You can only really find it on websites that charge shipping, so its more like $40. It's quite well made, and it's very compact as far as heatsink cards go. The primary competition is the Angelbird Wings PX1, but that card looks huge and is covered with LEDs. The heatsink might have more mass, but the size of this one has proven to be more than sufficient for a 950 or 960 Pro.

I missed out on a sale Amazon had via a coupon to get a Plextor M8PE bare M.2 1TB for $315, otherwise I would have picked one up and thrown it into another KryoM.2.

Somebody who bought the Plextor (or was the M8PE the one with the red/black "heatspreader?") -- said that it ran hot.

I've also paid attention to OEM Samsung "Pro" drives, in partnership with outfits like Lenovo. These are the SM 961 and the PM 961 -- the PM variety being the OEM. This latter is billed as a "960 Pro PM961" IIRC. And it doesn't work with Magician; it's not bundled with anything, because it's an OEM part. Still -- the 960 Pro drives -- probably already said so -- were priced the other day hitting about $830 and change. The PM961 is about $620.

Most of my caching experience so far with this 960 EVO 250GB is not responsible for a great deal of writes to the drive. Instead, in the process of tweaking the cache and the drive allocations, deleting and creating volumes will rack up a lot of TBWs. I've had this EVO for about 2 weeks, and because of my trials and errors, it has 800GB of writes already. They only guarantee the EVO for about 400 TBW, but that's a long way to go.

Thanks to you, when I spring for a Pro or a competing drive from another manufacturer, I'm going to get one of those KryoM.2 cards.

But it stands to reason, doesn't it? EVen if you got a Pro at a fire-sale price, nothing would be too good for it. It's just a good investment. I saw it on Amazon for about $80! Supply and demand -- they're squeezing what they can out of enthusiast techno-lust.

 

[Hail The Brain Slug]

Most of my caching experience so far with this 960 EVO 250GB is not responsible for a great deal of writes to the drive. Instead, in the process of tweaking the cache and the drive allocations, deleting and creating volumes will rack up a lot of TBWs. I've had this EVO for about 2 weeks, and because of my trials and errors, it has 800GB of writes already. They only guarantee the EVO for about 400 TBW, but that's a long way to go.

My 960 Pro has 2.4TBW already, but a lot of that was benchmarking I really didn't need to do. I'm not worried, though. I put about 10TBW on my 950 Pro in 1 year of use including a similar amount of benchmarking. At that rate of use, my 950 pro would have hit the 400 TBW rating in 40 years.

I expect a similar story with the 960 Pro.

Thanks to you, when I spring for a Pro or a competing drive from another manufacturer, I'm going to get one of those KryoM.2 cards.

IMO heat is a major source of electronic component failure. I think keeping the temperature nice and low should greatly reduce the chance of a component failure over time. Definitely worth the $40.

But it stands to reason, doesn't it? Even if you got a Pro at a fire-sale price, nothing would be too good for it. It's just a good investment. I saw it on Amazon for about $80! Supply and demand -- they're squeezing what they can out of enthusiast techno-lust.

What you saw might have been the water-block versions. They make them with two different styles of water block in case you're nutter enough to actually think water cooling it will be more effective than the already fairly sizable passive heatsink option. Maybe for particularly hot drives, but I would never go that far.

For funsies, here's my crappy photo album of installing it. I eventually took it back out over paranoia of the bending and rolled the lower thermal pad flatter (They're silly putty consistency) and got the drive to seat nice and flat.

 

[BonzaiDuck]

My 960 Pro has 2.4TBW already, but a lot of that was benchmarking I really didn't need to do. I'm not worried, though. I put about 10TBW on my 950 Pro in 1 year of use including a similar amount of benchmarking. At that rate of use, my 950 pro would have hit the 400 TBW rating in 40 years.

I expect a similar story with the 960 Pro.



IMO heat is a major source of electronic component failure. I think keeping the temperature nice and low should greatly reduce the chance of a component failure over time. Definitely worth the $40.



What you saw might have been the water-block versions. They make them with two different styles of water block in case you're nutter enough to actually think water cooling it will be more effective than the already fairly sizable passive heatsink option. Maybe for particularly hot drives, but I would never go that far.

For funsies, here's my crappy photo album of installing it. I eventually took it back out over paranoia of the bending and rolled the lower thermal pad flatter (They're silly putty consistency) and got the drive to seat nice and flat.

Yeah, I caught that when I saw the difference between what you paid and what I initially found. A Florida outfit known to many enthusiasts on this side of the puddle (Performance PCs) had the expansion card and the heatsink as two different items totaling approximately what you claimed as a Euro price.

Now, the price at Newegg has moderated back to $700 for the 1TB Pro.

Also, on the photo album, it reveals the design of the card offering both high promise and uncertainty for me, because it sits vertically in the PCIE slot as opposed to "low profile." It could actually interfere with one of my side-panel fans, mounted in the inner plastic door of a Stacker 830. But if there is clearance, there would also be maximum cooling for the heatsink. I need to get my little metric ruler and open up the case to see.