[GTALUG] war story: buying RAM

[Lennart Sorensen]

On Wed, May 10, 2023 at 11:12:44AM -0400, D. Hugh Redelmeier via talk wrote:
> This is an M75s gen 2 SFF unit.
> I installed the DIMMs and then could not replace the tray with 2.5"
> and 3.5" bays.  It bottomed out on the DIMMs.
> 
> I suspect that the DIMMs themselves would fit but their flamboyant
> heat spreaders would not.  The heat spreaders had stickers saying "warranty
> void if removed" so I didn't want to try.  I have no idea how
> important heat spreaders are for RAM.

Oh OK.  Yeah the ones crucial was selling had no such nonsense on them.
They are just regular DDR4 DIMMs.  So it does not need low profile,
it just needs normal sized.

> Speaking of which, the NVMe drives I buy come without heat sinks.  Are
> they important?  My guess: perhaps for NVMe gen 4 because they seem to
> take considerably more power.

Some of the faster NVMe drives can get hot, especially when writing, and
if they get too hot they tend to throttle and loose speed, so the heatsink
can help keep the performance up.  Of course with normal use you probably
aren't accessing it that much so it probably doesn't matter that much.

> If heatsinks are needed, why don't they come with the drives?  Are
> they supposed to come with the motherboard?

It varies.  Some motherboards include them, some don't, some drives do,
some don't.  Of course a heatsink from a drive might not even fit with
a certain board or case, depending on the size and shape.

> In the case of the M75s, the NVe mounting kit that I cannot get from
> Lenovo does include a heat sink.
> <https://www.lenovo.com/ca/en/p/accessories-and-software/thinkcentre-and-thinkstation/thinkcentre-and-thinkstation-hard-drives/4xf1c39743>
> 
> There is no such heat sink in the M75q.

I suppose the case being so small they figure having a heatsink to help
cool the drive is a good idea.

> I may well be misremembering the number.  Perhaps it was $200 for each
> of the two 32MiB SDRAM DIMMs.  Or even some other price.  It was
> probably early fall in 1997, but I'm not sure.  (It was before Win 98
> was released.)

Yeah I am thinking February of 1997.  $200 each could have made sense
unless ram prices took a dive in 1997.  I don't remember.

> I do remember 64M being extravagant at the time.  I bought it with two
> 32MiB SDRAM DIMMs I think the motherboard was an Asus TX97.  That used
> an Intel TX chipset.  The TX had a limitation that it could not cache
> more than 64M of physical memory.  The CPU was an AMD K6 200 or 233.

Yeah I was working with my dad's new PPro 200 at the time.  It was his
CAD system, and it was maxed out.

> It was a while before I switched over to the K6/Linux box from the
> SPARCClassic.  Solaris was still better than Linux and the
> Sun's display was better.  The PC's CPU was a lot faster -- 200MHz vs
> 32Mhz.
> 
> At that time, "workstations" were still a higher tier than PCs.  PC
> stores didn't understand workstation users.  But there was almost no
> remaining technical reason so this disappeared fairly quickly.  Linux
> was one of the factors. It is hard to remember how bad the PC display
> situation was compared with workstations.

I thought PCs were able to have pretty decent displays, but they were not
cheap.  We used a 20" Mitsubishi Diamond Pro 20 with a 486DX2-66 which
managed 1280x1024 at 60Hz, which was then replaced with the PPro 200 with
a 21" Hitachi CM803U in 1997.  1600x1200 at 85Hz seemed quite acceptable.
The 486 ran a Mach32 2MB VLB card, the PPro was a Rage II 8MB PCI card.
There was a 486DX50 before that with a #9 GXi Level 25 video card
(2MB VRAM + 1MB DRAM ISA card with a TI34020 processor on it), but it
was stolen and the other 486 was the replacement.  The #9 card was
displaylist based, which gave some amazing performance even though it
was ISA.  The software just sent commands to the card, and you kept the
actual data in the DRAM on the card and the TI chip executed the commands
to generate the framebuffer.  Where machines with standard Trident or
other VGA cards would need seconds to do a refresh in a CAD program,
this thing took miliseconds since the card did all the line drawing
and scaling.  It was also able to implement the entire GDI stack of
windows 3.1, so all window drawing, text rendering, fills, etc, were
all done in hardware.  Video playback is what killed TIGA since tat
required transfering new framebuffer data all the time, and ISA was no
good for that.

And I have no idea why I can still remember all that crap. :)

-- 
Len Sorensen