Cool.
I have 10 of the 4Gbs. Should have waited???

Actually 4 is a lot of memory for single purpose computers. I am going
to run ISP functions and microservices.
Since I currently have Postfix-Dovecot, SVN, Web Gateway, and a few
other things running on a 10 year-old 2gb server.
I am pretty sure that I can split this over a couple of 4Gb Pis and come
out ahead.

As an OpenSCAD rendering accessory for a notebook, it should have memory
to spare with 8GB.

Have you run OpenSCAD on it?
What OSs have you tried? I would like to have CentOS on it but so far as
I know there is no official distribution for the ARM-72.
I have tried a few others.
A headless Raspbien should be fine as a basis for a rendering machine.

Ron

On 2020-06-16 2:44 p.m., Torsten Paul wrote:

--
Ron Wheeler
Artifact Software
438-345-3369
rwheeler@artifact-software.com

On 16.06.20 22:23, Ron Wheeler via Discuss wrote:

Unless you have some extra heavy work load it should
not matter that much. More RAM is always nice, of cause,
but it does come with the higher price as well so it's
not directly replacing the 4Gb model.

Just a little bit for testing the AppImage and running
the speed comparison.

Only Raspberry PI OS (64-bit) Beta so far. Just got
the new PIs for a work project. I have a 4Gb model
running headless as podcast downloder + NAS for a
while though and that works just fine.

Yep, I hope once the release happens it will also
come in that lite version which is great for that.

ciao,
Torsten.

On Tue, 16 Jun 2020 14:15:11 -0400
Ron Wheeler via Discuss discuss@lists.openscad.org wrote:

You believe wrongly. That stopped being true around the time of the AMD
K6. RISC designs were at one point doing more per clock. Often they did
one instruction per clock or per two clocks when the 486 or Pentium took
several to do something. Today performance processors are quite different
to either historic RISC or CISC and in fact are much alike internally.
They do multiple instructions per clock regardless of instruction set.

Today's great battle is a three way battle between massively parallel
simple units, lots of conventional small processors, and larger
processors with more parallel instructions. Smaller transistors but no
huge clock rate/thermal improvement effectively means the question is 'how
do you do parallelism best ?'

The first is GPU, the second is AMD x86 and most performance ARM, the
third is stuff like Intel with AVX512 and the like plus systems like
Power.

The Pi is however none of these. It's a slower, cheaper and smaller
processor design fabricated on what is now a very old but reliable
and low cost process. It's designed for something quite different and
fills a set of market spaces that performance processors do not. It's
very low cost, it interfaces easily to 3.3v parts, it has very low
cost supporting glue and PCB design - hence the way it makes the Pi work.

Phone processors are different again being designed for very high
integration, burst performance and very low power.

There are also then processors designed to be good at continually having
work to do slowly but efficiently - eg in the telco space.

Alan

Great info.

Shows my age!!!
I thought that 2015 was just yesterday!
It is appalling to hear 2015 decribed as "very old".

I run my little datacenter on CPUs built as far back as 2007.

No wonder I think that the Pi4 is fast!

Ron

On 2020-06-16 6:27 p.m., Alan Cox wrote:

--
Ron Wheeler
Artifact Software
438-345-3369
rwheeler@artifact-software.com

Thanks for sharing.

The price is pretty low for the extra 4Gb, if you have an application
that can use caching or a lot of in-memory temporary data.

I purchased a Bitscope rack that will hold 20 Pi4s and have designed a
power supply to support it but COVID hit just about the time I was ready
to do some 3D printing and laser cutting for an acrylic case at my local
Fablab (Public Library) to mount these in my rack.

Given that a lot of my current servers are 4Gb or less and with
Kubernetes/Docker I can build a multi-CPU platform for microservices, I
am looking forward to migrating to a physical cluster where physical
Pi4s replace VMs that are made from dividing up larger more expensive
servers.
At $55 per physical VM, why bother with virtual VMs.
In a 4U rack, one can actually get 40 Pi4s. That is a lot of power in a
4U rack for less than $3,000 USD.
Would make on heck of a rendering farm for an office of designers.

With 8GB, one could actually break a Pi4 into 2 or 4 VMs, so that is
even more flexibility if your applications are not CPU-bound.

Ron

On 2020-06-16 5:15 p.m., Torsten Paul wrote:

--
Ron Wheeler
Artifact Software
438-345-3369
rwheeler@artifact-software.com

On 16.06.2020 15:59, nop head wrote:

It should indeed matter, because OpenSCAD is to my knowledge not running
booleans in parallel threads. If so it will onbly be using 1 out of 4
CPUs in the PI 4. AngelCAD on the other hand should be able to use all 4
CPUs, like on any PC. But still, this is not

Carsten Arnholm