[GTALUG] Build critique request and the story behind it.

[D. Hugh Redelmeier]

| From: Russell via talk <talk at gtalug.org>

| Adding in the cost of the LAN case I'm $250 over my price point.

What's a LAN case?  Local Area Network?

| However 
| that point didn't figure in the cost of driving from store to store and 
| checking out the deals in 1994 and then ordering and driving back to 
| pick it up. This was all time and money for me back in the day. In 1994 
| it cost me $50 dollars a day to start my truck, not counting gas.

It's not very meaningful to compare 1994 and 2017 prices.  And the
retail marketplace is way different.

Compared with 2012 prices, current prices seem high to me.

- Moore's law is flagging so progress is slower.  This make the
  replacement market a lot smaller.

- fewer stores trying to grow marketshare in a no-longer growing market.

- Intel completely dominates the processor market and prices
  accordingly (with a slight dip due to AMD's Ryzen)

- Cryptocurrency Miners have bid up the price of AMD video cards

- RAM prices are high.  Supposedly due to fabs switching to producing
  flash memory.  But I suspect lack of competition due to
  consolidation or collusion.

- HDD prices are static.  I paid less for 3T Seagates from Newegg
  three years ago (Black Friday 2014 price: $89.99).

| One thing that confuses me is manufacturing reporting conventions. For 
| instance the DDR4-4000 modules are described as a Column Access Strobe 
| latency of 19. This CPU reports support for DDR4-2666 CAS 15. This 
| number refers to the onboard cache, I think?

No, it is not about the on-chip CPU cache.

The signals going in and out of RAM modules are complicated.  You could 
look at a timing diagram to understand them (I have, a few times, but it 
doesn't stick in my head).  What follows is from possibly defective memory 
(in my head).

Latency is about delay from sending a signal until getting a result.  
Except that for RAM it is broken into stages.  CAS tells the module that 
the computer has placed the column number portion of the address on the 
address pins (addresses are broken into half (column and row addresses), 
and each half is presented to the module's address pins in turn.

CAS latency says how many clock cycles come between the computer asserting 
CAS and the result showing up.  Your CPU can support as few as 15.  Your 
RAM actually takes 19.  No problem.

Memory module characteristics are available in a little ROM on the module 
itself.  The computer firmware queries all those ROMs and configures the 
computer's memory controller(s) to match.  Sometimes firmware setup 
screens let you override those setting but I would not do so.

| I went with a single 8 gig DDR4-2666 module, just to be safe. 

No need.  In fact, your computer would probably be faster with two.  But 
probably not enough to matter.

Modern memory systems are quite a lot more flexible than old ones.  
Memories don't need to be matched (but a slow one will probably slow down 
a fast one).

| I wish I was better at math.

This doesn't seem like math to me.  But then my math environment isn't 
normal.

Apparently "math phobia" is a thing that one can absorb from the 
environment.  There's a whole literature about this and the damage is 
does.

| I attribute some of my confusion to having 
| been frightened off math by my having viewed the machine math room at my 
| vocational school. All those heavy welded tables sorting punch cards, 
| all that noise concentrated in a small room brrrr, gave me the willies.

That sounds like "unit record" equipment.  Not really good for math.  
Generally used for accounting up until the early or mid 1960s.  The few 
classrooms that had them probably kept them a long time since they had 
been quite expensive.  Keypunches lasted perhaps 15 years longer than the 
other unit record equipment.

Band saws and open-hearth forge fans would be louder.

(In my shop class, the forge used coal which we had to coke before using.)

| Offhand I'm planning to put /home and /opt on the 3TB drive but if you 
| have any recommendations for an optimal partitioning scheme for the SSD, 
| in order to reduce writes as an effort to prolong service life, I'd 
| appreciate it.

If you do the arithmetic (necessarily guestimates) SSDs will probably last 
as long as you'll want to use them, and longer.

In my limited experience, SSDs typically fail more abruptly and
completely than HDDs.  So backups are even more important.

Since I'm lazy about backups, I generally put the OS on the SSD and my 
stuff on the HDD.  At least when I have both SSD and HDD.  But fewer of my 
computers have HDDs at all.  And I really should do more backups.

I find modern desktops (Windows and Linux) cry out for SSD performance.

I'm setting up CentOS 7 on a little computer that came with a 32G SSD.  
That's fine for the OS + 10G swap (I added a 2T HDD for /var and /home).  
Your SSD is 512G so it would be a waste to use it just for the OS.