NVMe SSD's - Not all they're cracked up to be...

See:
http://www.tomshardware.com/reviews/samsung-sm951-nvme-raid,4418.html

If you want to boast of storage subsystem 'scores', NVMe based drives are what you need to throw your money away on.

If you need to run SQL type databases (and ones that fit inside a 0.5TB or 1TB (in RAID0) capacity, they'll do a much better job for a while (the warranted endurance ratings makes them ill suited for enterprise level workloads with high write requirements - especially high 4K random writes).

For anything in between? Well, you're more towards the 'throwing your money away' tilt of the above.


See:
http://www.tomshardware.com/reviews/samsung-sm951-nvme-raid,4418-3.html

I've always stated that artificial BM's are next to useless in determining how effectively a product will increase my productivity in my real world workflows and workloads. With 'useless', being rated higher than most of the BM's available for storage subsystems today.

There is one that is available today that goes against simply reporting bigger numbers (or higher differences between SSD's) and that is 'PCMark 8 Real-World Software Performance'. Not exactly what I need, but much more useful than any other BM out there for when I need to make a buying decision on nothing other than what is available 'online'. Of course, whatever I pick is still tested for suitability for my uses... and compared against whatever component I was previously running in my 'complete' workloads and workflows (i.e. I don't test for a single metric - the overall performance benefit or loss is what is important. And I don't lose sight of that for an instant).

With the above in mind, and following the second link above, we can see where the top NVMe drives fall within each other in as close to real world usage as possible. (Note; I ignore the gaming tests - not a reflection of my uses at all).

With the PS Light workload, we can see that the 950 Pro is the highest rated NVMe drive. Yeah; even beating out the RAID0 contenders also (as I've always stated too).

In the PS Heavy workload, there is a tie between the 950 Pro and the 951 NVMe drives - but both still beat any RAID0 configuration shown.

As we click 'next' until we reach the MS Powerpoint graph, we see the 951 and the 950 tie or trade positions slightly for each test.

This comparison of M.2 drives is pretty useless on it's own though as it doesn't answer the 'compared to what previously running component' question I ask of each new component I test before a purchase decision is reached.


See:
http://www.tomshardware.com/reviews/samsung-850-evo-850-pro-2tb-ssd,4205.html

If we scroll down to the 'PCMark 8 Real-World Software Performance' graphs on the link above, we can compare to SATAIII drives as best we can 'online'.

Starting again with our PS Light workload graphs on both the M.2 and the SATAIII links above, we see the following:

(Note: I'm not picking certain drives to compare here; just the 'best' SATAIII's vs the 'best' M.2's, whichever they may be from test to test).

3.8 seconds (~3%) faster M.2 drive for PS Light on a less than 2 minute test.
8.6 seconds (~2.5%) faster M.2 drive for PS Heavy on a less than 6 minute test.
1.7 seconds (~2.5%) faster M.2 drive for InDesign for a less than 1 minute test.

Uh... are the others even worth typing out?


What is interesting about the first and second links about M.2 drives is how temperature and throttling problems (especially in a 'slim type' notebook) are skirted and not even mentioned in the least (these issues will degrade the M.2's drive performance to not only below SATAIII levels, but potentially below HDD levels too.

What did we learn from the above exercise?

First:
RAID0 in up to workstation class workloads is not only NOT beneficial, it is a detriment to performance too when real world workloads are compared and not the unimportant 'synthetic' BM's that are thrown at us by manufacturers and most of the online rags alike.

That is also ignoring the reliability issues of RAID0 too - with each added drive in an array, the reliability of the system as a whole is decreased dramatically - not just with potential hardware failures, but also software, O/S and driver failures too which cannot always be foreseen and accounted for. Drive image backups won't help when a project is due in hours and Amazon promises next day shipping for free...

Second:
M.2 drives can be faster as a group vs. SATAIII drives, but at what cost?

That cost is not just the one time cost of a drive (which may or may not be less than an similar capacity 2.5" SATAIII drive). No, to me it is the cost of the drive, the cost of a much smaller capacity that will affect my workflow (negatively), the cost/time of implementing that drive into a given system for a given lifecycle and also the higher maintenance (in time) that managing a smaller drive will entail over that lifecycle (at least monthly, if not weekly or even more often as the much smaller capacity becomes a deadweight that needs to be tossed overboard to save the platform it serves.

Third:
With the heat/throttling issues actively being avoided (still) by the online rags, I can only conclude that unless these type of drives are used in a chassis (notebook or desktop) that takes M.2 cooling seriously, M.2 is still a fail for any system that has the potential to use an 2.5" SATAIII drive of 1TB or larger.

Fourth:
Yeah; there are some workloads that demand the highest performing drive you can buy. Workstation class workloads are not in that category though.

The overall benefits, including the performance aspects, of a good/great 2.5" SATAIII drive has not been met yet in the M.2 world. The downsides are plenty and include only a quarter of the capacity, untamed heat/throttling issues and an average of 2% raw performance increase on real world workflows that can be negated easily with the throttling issues and/or the capacity restrictions on a light or heavy, but sustained and daily dependency on such a storage subsystem.

If the above was the sole information I had to rely on as to whether to deploy M.2 drives or not; the answer is (still) clear to me.

No.

Still waiting for the true successor to 2.5" SATAIII SSD's for mobile use...

 

Well you know I am a light user, just browse the net and play games sometimes. To me, the 950 PRO NVMe 512GB SSD x2 that I have is a small downgrade? Yes true the benchmarks are higher than the normal 2.5" SSDs I owned before, but the boot time is much slower, so from what I experienced as a user, I lost that snappiness feel of an SSD that I felt the first time I upgraded from an HDD to an SSD. All these 2GB/S+ speeds the get in sequential benchmarks mean nothing to me using my workflow.

 

Don't know about you but my SM951 definitely feels like an SSD and those transfer speeds are insane when copying files.. Sure sustained performance for hours isn't that good but then again I don't copy more than 50GB at one time..

 

That was one of my points. Phoenix is running RAID0 and you're running a single drive (I assume).

 

Yup single SM951... See my sig lol..

 

Yeah, and even though the 950 was the top drive in many of the tests, in RAID0 it still was slower than any single drive...

 

See:
http://www.tomshardware.com/news/micron-3d-nand-flash,31193.html

June, ah June... (2TB+ SSD's from other than Samsung with s****y TLC ... Win10 Redstone RTM... and at least news about Optane and 3D-XPoint tech...).

Can't wait...

And, one more reason why M.2 drives are a stillborn tech except for the 'sexy' thin/anorexic notebooks that don't have a GbE LAN port, let alone a proper 2.5" (7mm) drive bay.

 

You think?

 

^What boot time you are getting vs trad SSD? I am look into NVMe Based stuff but it seems the price of entry is still too high.

 

Yeah, I do.

Not in synthetic 'scores' (yawn), in real world workflows...

 

Getting closer...

See:
http://www.legitreviews.com/intel-s...d-xpoint-memory-that-fits-in-ddr4-slot_176826

While the '7x' faster than SATAIII SSD's may not seem like much, it seems it will still be faster than any current NVMe drive.

Note that Kaby Lake platforms will be (possibly?) required. That means Windows 10 x64 Pro is a mandatory requirement then too.