[Discuss] I need a recommendation for a Linux-compatible high-end PC builder

[Alan W. Irwin]

Our current principal PC is low end by today's standards, i.e., it has
a 2.4GHz clock rate and only two real cpu cores. However, a much
larger concern with this system is its age which is a decade old. It's
a tribute to ASUS motherboard quality back then that the capacitors
and connectors on this MB are still doing well, but there are limits.
Therefore, I would not be surprised if I have a sudden need for a new
system at any time.  However, I have been saying that for the last 7
years!  So a more important concern is that system is rather low end
for my current needs.  So in anticipation that I will buy a PC
sometime this year I am looking for a Linux-compatible high-end ($1500
price range) PC builder with a good reputation.  So your comments
on that would be much appreciated (other than "build it myself" since
I really don't have that skill set).

Our current system was built by The Edge which was managed by Matt
Speller at that time.  The box he came up with obviously worked out
very well for our Linux-only needs so I would ask him to build me a
box again in a heartbeat, but when I checked with The Edge several
years ago he was no longer with that company. Does anybody here know
whether Matt is still in town and still involved with box builds?

Alternatively, can someone give me a recommendation pro or con for a
Linux-aware local (preferred) or North-American (e.g., zareason if
you have used them recently) Linux box builder?

My high-end PC needs are as follows:

* Must be completely Linux compatible.  I would be willing to pay a
   small premium if Debian was installed on the box since that
   immediately proves its Linux compatibility, but if it is cheaper to
   have the box sold without an OS, then I can install my favorite
   distro, Debian, for myself so long as I can get free replacement
   components if one of those proves not to be Linux compatible.

* The box should be green, i.e., not require excessive power
   consumption. This makes sense just based on the requirement below
   that the box must be quiet, because less power consumption means less
   temperature dissipation required means less fan noise.  Also, we want to
   obtain instant access to our computers at any time, and I also have
   heavy build jobs I schedule each night so we always run our Linux
   boxes 24/7.  Given those always-on needs

   1 watt * 1 KW/1000 watts * 24 hours/day * 365.25 days/year * 0.13 dollars/KW-hour
   = $1.14/year

   Thus, if the box I buy requires 200 watts less than another with
   similar characteristics other than power consumption, that saves me
   ~$230 per year which is a small but still significant fraction of
   the typical ~$1500 cost of a high-end box.

* The box must be extremely quiet because it will be located in a
   small home office environment with two of us typically wanting to
   use that office without a lot of background fan noise to bug us.

* The box should have lots of memory (e.g., 64GB).  Barbara and I
   share the computer (simultaneously running two independent Linux
   desktops on it with direct access for one of us and thin-client
   access for the other). So that configuration takes extra memory.
   Furthermore, the Linux kernel takes advantage of all extra memory to
   reduce latency, Linux applications keep expanding their memory
   needs, and if the box lasts for 10 years I don't want to have to
   replace it prematurely because the memory is inadequate.

* The box will likely have AMD cpu's/chipset/graphics.  See comments
   below about multiple core support and graphics, but I have been
   recently been reading some articles concerning Linux kernel
   developers' anger at Intel's "Meltdown" and "Spectre" current design
   flaws. The jury is still out on whether AMD has similar design
   flaws. Their initial evaluation was "no" for Meltdown and "yes" for
   Spectre, but now there is some question about whether they are also
   have the Meltdown flaw. So the race is on between the two companies
   concerning how soon they can produce chips that are completely free
   of Meltdown and Spectre design flaws.  And I will likely buy the one
   that wins that race subject to the multiple core support and
   graphics issues where AMD appears to be winning right now.

* The box should have a big improvement in parallel power over my
   present system (e.g., 8-10 cores would be really nice). The reason
   for this need is I spend most of my computer time these days
   re-building and re-testing software after some development tweak I do, and
   those needs can take perfect advantage of parallel cpu's through GNU
   make's parallel build capability.  Many low-power cores (as opposed
   to a few high-power ones) would be an ideal design for my parallel
   build but low noise and low power consumption needs.  Finally, I looked at
   the zareason site to see what they supplied, and their best AMD
   desktop system could be expanded from 4 cores to 8 for an additional
   cost of $300 or $75 per core, but a similar test of their best Intel
   system showed the additional cost (in that case from 6 cores to 8
   cores) was $1200 or $600 (!) per additional core.  Note that core
   clock rate is only a rough guide to core quality, but for what it is
   worth the AMD core clock rate was 3.0GHz while the Intel core clock
   rate was "3.4-3.6GHz" (whatever that exactly means).

* The box should have reliable 3D graphics that have good open-source
   support.  I am not too happy with Intel integrated graphics these
   days because their support for our old Intel graphics chip has
   started to become problematic, i.e., I can now reliably crash the
   Intel X server by simply trying to halve the size of an image with
   the ImageMagick display application.  And that never used to be a
   problem for this hardware and older versions of the Intel X driver.

   I can completely understand the commercial motivations of the
   relatively large set of Linux graphics developers paid by Intel to
   support *new* Intel graphics on Linux.  But their official test
   environment typically does not include old Intel graphics hardware
   (they quit testing on my current g33 intel graphics about ~5 years
   ago) so former abilities are inadvertently lost (as above) because
   they don't have good access to old graphics hardware to confirm bugs
   and fix them. Therefore, assuming my next box also lasts 10 years, I
   don't want to have to prematurely throw it out because some graphics
   company is not using due care in supplying old graphics chips for
   testing and debugging purposes.

   The alternatives to Intel open-source graphics are Nouveau (to drive
   NVidia graphics), and the AMD/ATI open-source X driver (to drive
   AMD/ATI graphics).  Both of those open-source software projects have
   been excellent at supporting old graphics hardware according to
   phoronix articles.  In fact, AMD refers users who are still using
   old AMD/ATI chips to the open-source driver they support.  Also, I
   have been impressed with the reliability (I have encountered no X
   crashes at all during my heavy use for the last couple years ) of
   the Nouveau X driver for the NVidia graphics on my thin-client.  So
   either AMD or Nouveau sound good, and I would be tempted to go with
   the Nouveau X driver for driving NVidia graphics because that is the
   one where I already have good experience.  However, recently NVidia
   have become obstructive with the Nouveau project (see
   <https://www.phoronix.com/scan.php?page=news_item&px=Nouveau-XDC2017>).
   I have no idea if that is a temporary glitch in relations or signs
   of active obstruction that will continue for the forseeable future.
   But unless this issue is resolved quickly I will likely never buy
   NVidia again, and many others will likely do the same thing meaning
   overall interest in the Nouveau development will abruptly decrease.
   So we will see how it goes, but likely the choice will be AMD
   graphics.

I am primarily interested in your recommendations for Linux high-end
box builders, but debate over the above design characteristics would
be interesting as well.

Alan
__________________________
Alan W. Irwin

Astronomical research affiliation with Department of Physics and Astronomy,
University of Victoria (astrowww.phys.uvic.ca).

Programming affiliations with the FreeEOS equation-of-state
implementation for stellar interiors (freeeos.sf.net); the Time
Ephemerides project (timeephem.sf.net); PLplot scientific plotting
software package (plplot.sf.net); the libLASi project
(unifont.org/lasi); the Loads of Linux Links project (loll.sf.net);
and the Linux Brochure Project (lbproject.sf.net).
__________________________

Linux-powered Science
__________________________