Wretched Excess

Have you ever made one of those impulse purchases and then wondered what
you could have been thinking about immediately after? This time I bought
16gB of DDR4 for no other reason than that the price seemed good and the
parts matched what was already in the machine. In this case the machine is
a Shuttle XPC Cube SH170R6 with a Skylake I7-6700 that I just finished
building early last month to replace an elderly i7 system build in 2009.

I guess that there is no problem with having 32gB in a machine which is
going to be doing _really_ heavy-duty stuff but this machine is just my
personal daily-driver. It runs eight threads under the BOINC client 24X7
but that load works fine with only 8gB so there is no advantage there. I
remember that the first machine, if you can even call it that, I built in
1975 had 2kB of memory and the first 'real' machine a few years later had
64kB and I was really proud of that being so far ahead of everyone else's.
I guess I can take solace in the fact that there will definitely be no swap
file activity going on but that isn't much to brag about. Certainly it
won't make the machine any faster.

There were thoughts about making a self-clearing ram-disk from part of the
memory but I don't know what that would be useful for in my applications --
having an SSD as the system disk means that I/O speed isn't much of an
issue. Guess I probably should have saved the money and put it toward a
fast M.2 SSD which might have tripled the I/O speed over the SATA SSD.

Oh, and I scared the crap out of myself during the upgrade. After
installing the new memory the system wouldn't boot at all. I found that, in
the tight confines of the little Shuttle case, I hadn't _quite_ seated one
of the new strips - probably no more than 1mm of error. Even after 40+
years building computers things do go wrong with alarming frequency.

John McGaw wrote:
Have you ever made one of those impulse purchases and then wondered what
you could have been thinking about immediately after? This time I bought
16gB of DDR4 for no other reason than that the price seemed good and the
parts matched what was already in the machine. In this case the machine
is a Shuttle XPC Cube SH170R6 with a Skylake I7-6700 that I just
finished building early last month to replace an elderly i7 system build
in 2009.

I guess that there is no problem with having 32gB in a machine which is
going to be doing _really_ heavy-duty stuff but this machine is just my
personal daily-driver. It runs eight threads under the BOINC client 24X7
but that load works fine with only 8gB so there is no advantage there. I
remember that the first machine, if you can even call it that, I built
in 1975 had 2kB of memory and the first 'real' machine a few years later
had 64kB and I was really proud of that being so far ahead of everyone
else's. I guess I can take solace in the fact that there will definitely
be no swap file activity going on but that isn't much to brag about.
Certainly it won't make the machine any faster.

There were thoughts about making a self-clearing ram-disk from part of
the memory but I don't know what that would be useful for in my
applications -- having an SSD as the system disk means that I/O speed
isn't much of an issue. Guess I probably should have saved the money and
put it toward a fast M.2 SSD which might have tripled the I/O speed over
the SATA SSD.

Oh, and I scared the crap out of myself during the upgrade. After
installing the new memory the system wouldn't boot at all. I found that,
in the tight confines of the little Shuttle case, I hadn't _quite_
seated one of the new strips - probably no more than 1mm of error. Even
after 40+ years building computers things do go wrong with alarming
frequency.

I use RAMDisks, but they're not very fast.

https://s12.postimg.org/al2uxmjxp/HDTune_RAM.gif

The performance on the newer OSes varies a lot.
The "record low" on Win10 was 1GB/sec, on the very
first release of 10240. Later, I would see patches
of the disk dropping to 1.5GB/sec. Sometimes, it
manages 3.5GB/sec.

So definitely within range of some expensive NVMe drive.

My record highs, were running the version that still had
PAE support, on Win7 x32. I got 7GB/sec there. And for a
brief instant once, using some kind of "Secure Erase"
experiment, I got a burst of 10GB/sec in Performance Monitor
plugin.

But for most average work, if you write your own C code,
you get to use it at around 300MB/sec. Even a cheap SSD
can do that.

I'd like to compare with the Linux OS, which offers
TMPFS when you boot the LiveCD, but I didn't manage
to find a nice HDTune equivalent test program. The distros
vary as to their inode allocation. One was properly tuned
and had 8 million inodes on a 32GB TMPFS, and that one
was suitable for comparison to regular hard drives.
In a file creation test, TMPFS could create files
at a rate of 186,000 files per second. Which is
why I expect great things, with a decent benchmark
program.

Paul