Bill wrote:
Flasherly wrote:

And I'd need the real experience of running both PCIe Gen3x2 M.2
against SATA SSD for an impression of applicable benefits.

Since you brought it up, and since I am curious. I would like to ask if
there a difference between running these 2 technologies? My MB has the
capability of running both (I am using SSD Sata 6.0). I don't process
video, play games, or do anything else that involves a very large number
of GB all at once. I just enjoy a responsive system.

Bill

Let's take an example.

If I run a Macrium backup, if the storage is infinitely
fast, the backup still only runs at 200-250MB/sec.
Why is that ? The software computes a checksum while
it is backing up, and that is the rate limiting step.

I have little MD5 and SHA1 programs here of my own,
and the speed ranges between 100MB/sec and 300MB/sec.
And these are far from optimal programs.

Now, given these speeds for some existing tasks,
how much difference would a 600MB/sec or a 2400MB/sec
storage device make.

The answer is... none.

Once the things your computer does are "accelerated
enough", perhaps there is a benefit from a whizzy
storage device. For example, if the CPU had MD5 or
SHA1 acceleration, the Macrium backup could run much
faster, in which case a faster storage device might
help.

7ZIP can do a CRC32 calc on a file, at 1500MB/sec. But
no serious uses of CRC32 exist for GB sized files. It
was original invented to detect errors in 1500 byte packets.
If you "need a hash calc in a hurry" though, it's
one option, and a case where your fast storage device
pays off.

The "saddest day" for me, was having a fast storage
device, and owning some software that could only
achieve 1MB/sec performance from it. The same level
of performance it would have got, from an old
hard drive.

The fastest I/O I've seen on my machine, was spotted
in perfmon.msc, while something similar to Secure Erase
was being done at driver level on a RAMDisk. That
ran at around 10GB/sec for a couple seconds.

But in Windows, if any activity traverses full
storage stacks over and over again, no steaming
hot transfer numbers result. If you can "stay
in a storage driver" (which almost never happens),
it could go fast.

And this is really nothing different than we saw
25 years ago. We had desktop computers, where the
operating system could not use a disk drive at full
speed. And we would have a Unix box, which could
drive the disk at the limit of its speed (which
at the time might have been around 40MB/sec).
That impressed me the first day I saw it.
"Woah, this thing knows how to use a disk."
We need to slim down the OS design, to get
that feeling back again.

I keep my eye peeled, for some standout example
of a whizzy disk paying off, but haven't run into
that example yet. While CRC32 is fun, there's not
a lot of point to it.

Paul