• Welcome to Overclockers Forums! Join us to reply in threads, receive reduced ads, and to customize your site experience!

Why are m.2 SSDs so disappointing in real world performance?

Overclockers is supported by our readers. When you click a link to make a purchase, we may earn a commission. Learn More.

SPL Tech

Member
Joined
Nov 28, 2006
So back when we went from HDDs to SATA 6.0 SSDs, loading times improved massively accross the board. It was the single fastest upgrade you could do to a PC, often improving performance even more than upgrading the CPU in many cases. Now that we are in the world of m.2, we again see a massive increase in drive specifications, with m.2. drives easily twice as fast as SSD drives, but for loading time in Windows and in games there is absolutely no improvement whatsoever. Like literately zero improvement. An SSD is quite literately completely identical to an m.2 drive in actual boot speeds for applications. So why is that? Why was the HDD to SSD transition so game-changing but the SSD to m.2 transition completely worthless providing quite literately nothing of value except file transfer time boosts?
 
It feels like every post you make is hating on something...

I think the short answer is, things are a limit until they are no longer a limit. Once it is no longer the limiting factor, making it faster doesn't make much difference. I've seen a significant difference in load speed in one game comparing a SATA SSD and upper end M.2 NVMe model, but do agree for general use the difference isn't really noticeable. The advantage in going HD to SSD was mostly in random performance. Mechanical drives have orders of magnitude worse response there. There are measurable differences between SATA and NVMe SSDs, but if they're both on the "good enough" side, the difference in practice may not be significant.
 
My guess is that humans have become the bottleneck in the system. There has been an industry wide problem for quite some time: not enough users. The chief complaint amongst users could usually be traced back to hardware and every leap in hardware added a surge in users also. We are still using the same types of software though. Web browsers, word processors, accounting, etc.... These softwares have been optimized for hardware and OS, but now the user is the weak point. Users who didnt use a computer because they didnt know how or found slow hardware frustrating still dont know how to use hardware that has no lag. Software designers are not making a whole lot of new products, they are making the ones that exist marketable.

What I am saying is that users and software that need this higher type of performance are in the minority, and that we wont see big leaps forward until people need them. A decade maybe 2 and we will see software (for the end user) that can saturate an nvme drive.
 
A couple of things...

1. it never improved games except in load times.
2. M.2 is the connector. There are m.2 sata based ssds which are just as fast as sata, note. NVMe PCIE3.0 x4 drives are the speedsters. Im assuming you are actually talking about that.. :)
3. I see quicker windows load times using nvme based m.2 drives... slightly quicker load times in games. It depends on the title, etc as to the response.

Is it much or worth it, likely not for most...but there are improvements. a sata ssd would still be fine. :)


Take a few minutes to watch...https://www.google.com/amp/s/www.te...al-world-performance-nvme-vs-sata-vs-hdd.html
 
So back when we went from HDDs to SATA 6.0 SSDs, loading times improved massively accross the board. It was the single fastest upgrade you could do to a PC, often improving performance even more than upgrading the CPU in many cases. Now that we are in the world of m.2, we again see a massive increase in drive specifications, with m.2. drives easily twice as fast as SSD drives, but for loading time in Windows and in games there is absolutely no improvement whatsoever. Like literately zero improvement. An SSD is quite literately completely identical to an m.2 drive in actual boot speeds for applications. So why is that? Why was the HDD to SSD transition so game-changing but the SSD to m.2 transition completely worthless providing quite literately nothing of value except file transfer time boosts?

NVME in of itself is not disappointing the issue is one of expectations and the marketing being used to push them. True NVME drives like dedicated AIB Xpoint and others are the real deal. M2's issues occur when you try cramming an extremely hot drive into a space where there is limited cooling and airflow. It's not a whole lot different from when the first Raptor drives came out, are Raptors THAT much better than a normal drive? No. Why? Because the mechanics and physics of how the drive works hasn't changed.

The speeds you see on the package are only for the Controller, assuming you keep it cool enough and the SLC cache, assuming you do not exhaust it. Once either of those two conditions are violated you end up with a drive that is no better than SATAIII. The current crop of SATAIII drives are no different from NVME drives available other than the two items mentioned previously. Once there becomes a difference between the NAND arrays they use you will see NVME pull ahead but until that happens they are hard to justify. If you have an issue with that there are two ways to fix it and to show it's true performance. Keep it cool and put them in RAID0 and expand it's available cache and look to buy NVME drives with the largest cache possible. The same could be said of SATAIII but you lose the faster cache and controllers assuming money is no object and have to rely on the raw performance of the array itself.

As of right now unless you overclock, have money to burn and sufficient space and power going with a RAID-ed SATAIII will provide more consistent performance across the board.

When you read or write to an NVME drive the controller pulls the information from the array and puts it into it's SLC cache, that cache can range from a GB or two to as much as 50GB. Once it's there it stays there until it decides it's no longer needed and puts it back into the array. The idea is that the controller can manage the space and not have to use it's pagefile which is the array directly. Some drives can write to the array itself others can't again each causes their own respective problems but I hope that helps to illustrate what is happening.
 
Like ED posted, NVMe vs SATA controllers show performance differences for sure. One advantage not mentioned, which could be argued isn't specifically performance related, is the form and fit. An m.2 drive with any controller has a much smaller footprint and allows for portable devices to become more compact. The design engineers of lappys and tablets must be having a good time right about now.
 
Yeah, SPL Tech. I think you are failing to differentiate between SATA M.2 SSD and PCI-e (NVMe) SSD drives. The SATA interface of the SATA M.2 is the limiting factor. They just have different package than a conventional SATA SSD but will perform the same. The PCI-e interface of the NVMe SSD drives eliminates the bottleneck of the SATA interface so that their boot times are much faster. Once in Windows, however, you won't notice much difference most of the time between M.2 and NVMe I understand. I have not owned either one to this point and have confined myself to using conventional SSD drives.
 
Last edited:
I went from HDD to m.2 NVMe SSD and it's like a world of difference. A few years ago, I tried a SATA SSD and it was very fast booting up too, as long as it lasted, which wasn't long. But, my current Samsung 960 Pro 4X PCI-e is noticeably faster. I have a CrystalDiskMark screenshot in another thread. I have to warn people though, if you go with the m.2 drives for performance, make sure you add extra cooling (heatsinks and good air flow). The m.2 drives are faster at a cost and that cost is heat. I think we'll see a lot of premature drive failures at first with these m.2 drives until adequate cooling is addressed by manufacturer and user.
 
Yeah, SPL Tech. I think you are failing to differentiate between SATA M.2 SSD and PCI-e (NVMe) SSD drives. The SATA interface of the SATA M.2 is the limiting factor. They just have different package than a conventional SATA SSD but will perform the same. The PCI-e interface of the NVMe SSD drives eliminates the bottleneck of the SATA interface so that their boot times are much faster. Once in Windows, however, you won't notice much difference most of the time between M.2 SATA and NVMe I understand. I have not owned either one to this point and have confined myself to using conventional SSD drives.
Fixed... :)

I also have to disagree. There are differences in Windows as well. If you game, load times are improved. Most won't notice the split second it takes MS Office to load, however, look at the vid I posted and Photoshop. There are plenty of tangible results which show a difference. Whether or not that difference is worth the price premium these fetch, is up to the person deciding. Hell, I double blinded my wife and she picked out the NVMe based system. ;)

Also, while the bandwidth limitations are part of this issue, the 4K and IOPS is where these shine and a part of the reason they are so fast. 4k randoms are not terribly different, but, enough so that it is a part of the those boot times. Its how fast the drive is in other places that aslo play a large role in boot times. ;)
 
One thing not mentioned yet is the cost ratio of m.2 NVMe SSD vs. 2.5" SATA SSD. The SATA SSDs cost much less and while are slower than the NVMe drives, they are way faster than HDD (and don't have the heat issues of the m.2 drives). This is definitely a good compromise option for those on a budget.
 
Sustained transfers on nvme drives are leaps and bounds ahead of SATA based SSDs...sometimes six fold. Just there aren't a whole bunch of gains to be had from launching one game after another or loading one level or another because the bottleneck with rust drives was random access and that was removed with 2.5" SSDs.
 
on my two game rigs with nvme drives and 4790K's clocked to 5 ghz, I don't even see a splash screen.
on my 6800K rig at 4 ghz and nvme drive it's not a heck of a lot faster boot than a sata ssd, I guess a lot plays into it.
I haven't put a stop watch on them.
 
I think we need to distinguish between the heat issues with m.2 nvme and m.2 SATA. I think the heat issues are mostly with the nvme, correct? I wouldn't think an m.2 SATA would get any hotter than a plain old SSD sata.
 
I don't have any heat issues with any of the three 960 evo drives i use.
I have read a lot about the issue but can't speak to it.
 
I think we need to distinguish between the heat issues with m.2 nvme and m.2 SATA. I think the heat issues are mostly with the nvme, correct? I wouldn't think an m.2 SATA would get any hotter than a plain old SSD sata.

Correct. Heat is mostly with NVMe because of the high speeds. It's apparently the controller on the NVMe drives that gets the hottest. The SATA SSD run cooler, but I wouldn't be too complacent, they can get warm with a lot of usage, at least make an effort to get some case ventilation to the drives.
 
I think we need to distinguish between the heat issues with m.2 nvme and m.2 SATA. I think the heat issues are mostly with the nvme, correct? I wouldn't think an m.2 SATA would get any hotter than a plain old SSD sata.

Correct, I'll give two examples like-for-like

Samsung 850PRO said:
Environment
AVERAGE POWER CONSUMPTION
(System Level)3)
Active Read: Max. 3.5W (2 TB)
Active Write: Max. 3.0W (1 TB)
POWER CONSUMPTION (IDLE)3)
Max. 0.07W
ALLOWABLE VOLTAGE
5V ± 5%
RELIABILITY (MTBF)
Max. 2 Million Hours
OPERATING TEMPERATURE
0 - 70 ℃
Shock
1,500G, duration 0.5ms


Samsung 960PRO said:
Environment
AVERAGE POWER CONSUMPTION
(System Level)3)
Typ. 5.1 Watts (512 GB)
Typ. 5.3 Watts (1024 GB)
Typ. 5.8 Watts (2048 GB)
POWER CONSUMPTION (IDLE)3)
.04W
RELIABILITY (MTBF)
1.5 Million Hours Reliability (MTBF)
OPERATING TEMPERATURE
0°C to 70°C (Measured by SMART Temperature. Proper airflow recommended)
Shock
1500G, duration 0.5m sec, 3 axis

So if Samsung's specs are to be believed the 960 pulls twice the power of the 850 while in use. Also keep in mind the 850 was on an older fabrication process so it *should* be more than 2x and Samsung is one of the cooler running NVME drives. This is more typical of a current generation SATA3 drive

Intel Pro5400S said:
Product Collection
Intel® SSD Pro 5400s Series
Code Name
Products formerly Loyd Star Pro
Capacity
180 GB
Status
Launched
Launch Date
Q1'16
Lithography Type
16nm
Recommended Customer Price
$89.00
Performance
Sequential Read (up to)
560 MB/s
Sequential Write (up to)
475 MB/s
Random Read (8GB Span) (up to)
71000 IOPS
Random Write (8GB Span) (up to)
85000 IOPS
Latency - Read
50 µs
Latency - Write
50 µs
Power - Active
.09W Typical
Power - Idle
.05W Typical

So while in use a 960PRO pulls SIXTY FIVE TIMES the power or my current m2 sata3 drive but matches at idle. This explains why when polled NVME drives get red hot because the split in active vs inactive is immense!
 
Last edited:
STR was never the bottleneck in boot times. It was access time switching between the hundreds of files accessed on startup. A 2007 Samsung 64 GB SSD had < 0.5 ms seek time, compared to around 20 ms for 2006 SATA spinning rust. For every 50 files accessed, that was an extra second of boot time (20 ms * 50 seeks), hence why defragging was so important to limit the seeking as much as possible. When every single application relies on 20+ DLLs and you're running 10 different things on startup and Windows itself runs 40-50 different services by default, that's a lot of seeking.
 
Last edited:
There are things that can be done to optimize and speed up Windows start times. If you do both, get a SSD and optimize windows, well... it's great.
 
No, I am talking about PCI-e 4x m.2 NVMe drives vs. SATA 6.0 SSDs. Like for example the Evo 850 SSD vs the 960 Pro NVMe m.2 drive. All the tests I've seen show little to no difference in real-world performance for application loading times.

example:



 
Back