What is IOPS and does it matter?

[kanker777]

Hi there,

I was gonna buy a m.2 ssd and started by comparing all kinds of different drives.

But got kinda confused because I don't know what IOPS means and if I will benefit from a drive with a higher IOPS number, if I'm just gonna play games of of it.

If someone knows allot about tihs kind of stuff plz let me know

(btw it's about choosing between a Crucial mx300 m.2 525GB and a WD blue m.2 500 GB)

[jj9987]

Input/Output Operations Per Second, shows how many operations can that storage device carry out in one second. It does not influence when moving one big file, but when dealing with many small files, this will come into account a lot. HDDs usually have this in the range of 50 to 100, SATA SSDs have it up to 100k and M.2 NVMe drives have it even bigger, that depends on the device itself.

[Alex Kovacs]

You could do a little google search: https://en.wikipedia.org/wiki/IOPS

[ItsTheDuckAgain]

IOPS stands for input and ouput operations per second and basically tells you how much one can anticipate of the drive. It's a theoretical value, in general more IOPS = better and faster drive. Choosing between the two you decided to get, I'd go with the crucial, because WD should stay away from flashchips and do their spinning stuff :-)

[typographie]

http://www.anandtech.com/show/10741/the-western-digital-blue-1tb-ssd-review/6

http://www.anandtech.com/show/10741/the-western-digital-blue-1tb-ssd-review/8

If it helps, here are some benchmarks that include both the Crucial MX300 and WD Blue SSD. My (admittedly loose) understanding of storage benchmarks is that random read/write at low queue depth is what most closely approximates a consumer workload. The Crucial drive appears to be the clear winner there.

[JoostinOnline]

If you are buying a drive for programs or your OS (as opposed to storing large files), what matters is how fast it can read small files. Check Amazon reviews for the 4K read speeds. People upload screenshots of Crystal Disk Mark. Look for a drive that has 4K read speeds of at least 25MB/s. 

[DocSwag]

2 minutes ago, ItsTheDuckAgain said:

IOPS stands for input and ouput operations per second and basically tells you how much one can anticipate of the drive. It's a theoretical value, in general more IOPS = better and faster drive. Choosing between the two you decided to get, I'd go with the crucial, because WD should stay away from flashchips and do their spinning stuff :-)

WD acquired SanDisk, plus just because they make HDDs doesn't mean their SSDs are bad...

 

OP, comparing IOPs isn't a great metric though. You should look at benchmarks, since the IOPs numbers are measuring peak performance, which you'll never seen in the real world. Instead, look at benchmarks from places like tech report (for boot times and app launch times) or Anandtech (for more prosumer scenarios).

 

Question though, why are you looking at the m.2 versions of those SSDs? They're basically sata SSDs stuck in the m.2 form factor, so they aren't any faster than sata SSDs. Also, if the ssd is just for storing os, files, and apps nvme SSDs aren't worth it over sata ones since they don't improve boot times or app launch times.

[DocSwag]

3 minutes ago, JoostinOnline said:

If you are buying a drive for programs or your OS (as opposed to storing large files), what matters is how fast it can read small files. Check Amazon reviews for the 4K read speeds. People upload screenshots of Crystal Disk Mark. Look for a drive that has 4K read speeds of at least 25MB/s. 

When buying for os that doesn't matter at all, since SSDs aren't the bottleneck anymore.

[typographie]

2 minutes ago, DocSwag said:

They're basically sata SSDs stuck in the m.2 form factor, so they aren't any faster than sata SSDs.

I think it's just a question of build convenience. The M.2 form factor lets you get rid of a couple of cables, and at least for these two products the M.2 version doesn't seem to cost any extra.

[DocSwag]

20 minutes ago, typographie said:

I think it's just a question of build convenience. The M.2 form factor lets you get rid of a couple of cables, and at least for these two products the M.2 version doesn't seem to cost any extra.

Sometimes the m.2 version is a bit more expensive and I was too lazy to check if that was the case for these drives  

[orangecat]

More = Faster.

[JoostinOnline]

1 hour ago, DocSwag said:

When buying for os that doesn't matter at all, since SSDs aren't the bottleneck anymore.

Umm...yes it does.  You'll see a very noticeable difference between a 20MB/s and 30MB/s drive.  SSD's still go to 100% usage, especially at boot time.

[DocSwag]

36 minutes ago, JoostinOnline said:

Umm...yes it does.  You'll see a very noticeable difference between a 20MB/s and 30MB/s drive.  SSD's still go to 100% usage, especially at boot time.

Evidently not really, since the difference between an 850 evo and 960 Pro is .2 seconds.

 

http://techreport.com/review/30813/samsung-960-pro-2tb-ssd-reviewed/5

[JoostinOnline]

1 minute ago, DocSwag said:

Evidently not really, since the difference between an 850 evo and 960 Pro is .2 seconds.

 

http://techreport.com/review/30813/samsung-960-pro-2tb-ssd-reviewed/5

You're comparing two of the highest end drives on the market.  They have very close 4K read speeds.  Around 40MB/s if memory serves.

[DocSwag]

1 minute ago, JoostinOnline said:

You're comparing two of the highest end drives on the market.  They have very close 4K read speeds.  Around 40MB/s if memory serves.

Wat? 850 evo is highest end on the market?

[JoostinOnline]

Just now, DocSwag said:

Wat? 850 evo is highest end on the market?

When it comes to 4K read speeds (and therefore boot times), yes.  That's why there is no benefit to NVMe for boot times.   There are NVMe drives that load Windows slower than SATA ones.  The Intel 600p series, for example.

[DocSwag]

2 minutes ago, JoostinOnline said:

When it comes to 4K read speeds (and therefore boot times), yes.  That's why there is no benefit to NVMe for boot times.   There are NVMe drives that load Windows slower than SATA ones.  The Intel 600p series, for example.

The X25 is also in that list, and it was released like 5 years ago.

[JoostinOnline]

Just now, DocSwag said:

The X25 is also in that list, and it was released like 5 years ago.

So?

 

[NumLock21]

SSD with low IOPs feels like using a standard hard drive and they suck.

The Crucial SSD software is crap. WD's is better.

[DocSwag]

2 hours ago, JoostinOnline said:

So?

 

Stuff was slower back then, sequentials, IOPs, everything. Indicating that there obviously is a bottleneck elsewhere.

[JoostinOnline]

33 minutes ago, DocSwag said:

Stuff was slower back then, sequentials, IOPs, everything. Indicating that there obviously is a bottleneck elsewhere.

Only partially true.  The only thing that was slower was sequential reads, because SATA was a bandwidth limitation.  NVMe eliminated that.  However, the 4K read speeds are nowhere near the SATA cap, so NVMe doesn't benefit there.

[DocSwag]

1 hour ago, JoostinOnline said:

Only partially true.  The only thing that was slower was sequential reads, because SATA was a bandwidth limitation.  NVMe eliminated that.  However, the 4K read speeds are nowhere near the SATA cap, so NVMe doesn't benefit there.

Evidently you're not entirely correct, since from here:

You said that the 4k results indicate boot times, but the 960 pro has considerably higher 4k speed than any other SATA SSD. Yet boot times aren't better.

[JoostinOnline]

28 minutes ago, DocSwag said:

Evidently you're not entirely correct, since from here:

You said that the 4k results indicate boot times, but the 960 pro has considerably higher 4k speed than any other SATA SSD. Yet boot times aren't better.

That's strange.  If you look on Amazon reviews, you'll find lots of people (here's one, for example) showing far higher 4K read speeds.  Maybe they got a dud? The reviews for the 960 Pro were more consistent though.

 

Those graphs do show a heavy correlation between 4K read speeds and boot times.  It would appear that there are diminishing returns though.

[Mira Yurizaki]

I'm just going to leave this here because shameless self plug:

But if you don't want to read the post, here's the most relevant snippet:

Spoiler

Storage Specs and what they mean

IOPS and seek time

IOPS is short for input/output operations per second. This is how many times per second a storage drive can service requests, usually under worst case scenarios. i.e., IOPS are measured when doing 4K random tests. Seek time is the average time each request is serviced. IOPS and seek time are reciprocals of each other, much like hertz and cycle time. So if you have an IOPS rating of 300, it's reasonable to assume the average seek time is 3ms. IOPS is impacts two things:

• How quickly data can be read or written
• How the number of requests impacts the time it takes to complete an operation.

Bandwidth

This is how much data in a second the storage can transfer over, and usually the most visible or at least, the spec lots of people see and give weight to.

 

 

Queue Depth (QD)

Storage drives have queues to store commands from the system. In performance tests, it's usually done with 1 command per request and 32 commands per request. The former number is for worst case performance and the latter number is to create a fair comparison between AHCI and NVMe, as AHCI supports a maximum of 32 commands while NVMe supports 65536 commands per queue with 65536 queues. Though another reason may simply be that software hasn't really found a use for that many commands and the firmware of SSDs are tuned for lower QD values.

 

Sequential vs. Random read/write

Sequential represents the best case scenario when getting data off a storage device. In the absolute best case, it's simply something like "I want x number of bytes from location y" and the drive is able to do so. Random represents the worse case scenarios, with the worst being "I want 1 byte from [really large number] locations". You can think of sequential access like being able to read a book from start to finish with random access needing to flip back and forth between pages.

 

Random performance is more of a problem in hard drives. However, AHCI protocols can smartly queue up how the drive seeks data such that it's as sequential as possible, even if the data is random (it can re-arrange it on the controller card)

 

4K performance

The smallest unit of storage for the purposes of addressing is the sector. That is, in modern drives, you cannot access a single byte. If you want a single byte, you have to access the sector, transfer it over to RAM, and then have the application access a single byte. Sectors today are 4 kilobytes, or 4096 bytes, hence the "4K." 4K performance represents the worst case scenario, meaning addressing the smallest amount of data across random locations.
 

How all of this plays together

Imagine you have two scenarios:

1. Transferring 1,000 files of 1 MB each, totaling 1GB
2. Transferring 10 files of 100 MB each, totaling 1 GB

And you have two drives with the following performance:

1. 100 MB/s with an IOPS rating of 10,000
2. 500 MB/s with an IOPS rating of 100

You might think that in either case, the 500 MB/s drive should beat the 100 MB/s drive. However, this is the actual performance between the two:

• Transferring 1,000 files of 1MB each
  • The 100 MB/s drive takes a total of (1GB/100MB/s + 1000 * seek time) = 10 + (1000 * 0.0001) = 10.01 seconds.
  • The 500 MB/s drive takes a total of (1GB/500MB/s + 1000 * seek time) = 2 + (1000 * 0.01) = 12 seconds
• Transferring 10 files of 100MB each
  • The 100 MB/s drive takes a total of (1GB/100MB/s + 10 * seek time) = 10 + (10 * 0.0001) = 10.001 seconds.
  • The 500 MB/s drive takes a total of (1GB/500MB/s + 10 * seek time) = 2 + (10 * 0.01) = 2.1 seconds

Knowing how these two interact is key when figuring out if your use case will benefit from a storage drive or not. Of course, there hasn't been a case where bandwidth and seek time are disproportionate like this, but this was an example to demonstrate how one aspect affects things.

 

[DocSwag]

1 hour ago, JoostinOnline said:

That's strange.  If you look on Amazon reviews, you'll find lots of people (here's one, for example) showing far higher 4K read speeds.  Maybe they got a dud? The reviews for the 960 Pro were more consistent though.

 

Those graphs do show a heavy correlation between 4K read speeds and boot times.  It would appear that there are diminishing returns though.

It seems that most tests peg it more around 40 MB/s, so I'm not entirely sure, but most put the 960 pro at around 60 MB/s.

 

I do agree, there is a bit of a correlation, but the difference in boot times is much smaller, most likely, as I said, because of bottlenecks elsewhere.