From Newsgroup: comp.lang.forth

I have been impressed with one of our frequent contributor's encyclopedic knowledge of 21st century microprocessor architectures during my occasional forays to comp.arch .

This may be slightly off-topic for comp.lang.forth but as this is a more intimate group with a great deal of both knowledge and opinion, let's couch this in terms of microprocessors that can run Forth well on Linux.

Is there a particular CPU that shines in overall architectural elegance?
Which microprocessor is closest to getting it all right? Is it just a
matter of buying the latest available from ARM / Intel / AMD or are
newer models using cores & kludges to gain performance whereas a system based on a 4.4Ghz Intel Pentium Gold 8505 is really the choice for someone who wants a brilliantly executed machine?

--- Synchronet 3.21a-Linux NewsLink 1.2

From Newsgroup: comp.lang.forth

Buzz McCool <buzz_mccool@yahoo.com> writes:

let's couch this in terms of microprocessors that can run Forth well
on Linux... are newer models using cores & kludges to gain
performance ... ?

Users mostly care about performances more than about architectural
brilliance, and so if a vendor can gain performance by adding cores and kludges, they have strong incentive to do so regardless of how much
elegance they started with. And elegant architectures for Forth are
probably not well suited for Linux and vice versa. Succesful Linux
CPU's have quite a lot of registers and while fancy Forth compilers can
make use of them, you end up with an abstraction inversion, simulating a
stack machine that way.

You might like the TI MSP430 which is a 16-bit MCU line inspired by the
PDP-11. It can't run Linux and it's now mostly eclipsed by ARM Cortex-M
and maybe soon Risc-V MCU's, but its CISC-style addressing modes are
convenient for Forth, and its hardware implementations have some neat
features like built in non-volatile (ferromagnetic) RAM in some models.
--- Synchronet 3.21a-Linux NewsLink 1.2

From Newsgroup: comp.lang.forth

In article <87o6ohh02r.fsf@nightsong.com>,
Paul Rubin <no.email@nospam.invalid> wrote:

Buzz McCool <buzz_mccool@yahoo.com> writes:

let's couch this in terms of microprocessors that can run Forth well
on Linux... are newer models using cores & kludges to gain
performance ... ?

Users mostly care about performances more than about architectural >brilliance, and so if a vendor can gain performance by adding cores and >kludges, they have strong incentive to do so regardless of how much
elegance they started with. And elegant architectures for Forth are
probably not well suited for Linux and vice versa. Succesful Linux
CPU's have quite a lot of registers and while fancy Forth compilers can
make use of them, you end up with an abstraction inversion, simulating a >stack machine that way.

You might like the TI MSP430 which is a 16-bit MCU line inspired by the >PDP-11. It can't run Linux and it's now mostly eclipsed by ARM Cortex-M
and maybe soon Risc-V MCU's, but its CISC-style addressing modes are >convenient for Forth, and its hardware implementations have some neat >features like built in non-volatile (ferromagnetic) RAM in some models.

See noforth made by Willem Ouwerkerk, for several SBC MSP430's.
(Nowadays RISCV too.)

https://home.hccnet.nl/anij/nof/noforth.html

I've used it to have a 9600 to 31.5 baud convertor in behalf of midi.
Supports IOT, by wireless transmission (extra hardware needed).

Groetjes Albert
--
The Chinese government is satisfied with its military superiority over USA.
The next 5 year plan has as primary goal to advance life expectancy
over 80 years, like Western Europe.
--- Synchronet 3.21a-Linux NewsLink 1.2

From Newsgroup: comp.lang.forth

On 12/1/2025 10:25 PM, Paul Rubin wrote:

Users mostly care about performances more than about architectural brilliance, and so if a vendor can gain performance by adding cores and kludges, they have strong incentive to do so regardless of how much
elegance they started with. ...

Understood, and thanks for the information about the TI MSP430.

It's a bit silly, but I like to peruse this benchmark site and think about getting a system powered by some forgotten gem. https://www.cpubenchmark.net/cpu_value_alltime.html




--- Synchronet 3.21a-Linux NewsLink 1.2

From Newsgroup: comp.lang.forth

Buzz McCool <buzz_mccool@yahoo.com> writes:

Is there a particular CPU that shines in overall architectural elegance?

Elegance is for tailors:-)

If you look at RISC-V (e.g. the instructions in RV64GC), it shows a
certain minimalism compared to its competitors AMD64 and ARM A64.
That also shows in the size of the RISC-V manual compared to the
manuals of its competitors.

Which microprocessor is closest to getting it all right? Is it just a
matter of buying the latest available from ARM / Intel / AMD or are
newer models using cores & kludges to gain performance whereas a system based >on a 4.4Ghz Intel Pentium Gold 8505 is really the choice for someone who wants >a brilliantly executed machine?

The Pentium 8505 is an Alder Lake U with half of it's P-cores and half
of its E-Cores disabled, and some of the cache, leaving 1 P-Core and 4
E-cores. If you want the best performance, there are certainly a
number of microarchitectures by AMD, Apple, ARM, Intel, and Qualcomm
that compete for that spot and show fascinating feats in clock rate
and instructions per cycle (IPC), and the cores in the Pentium 8505
are not far behind; newer microarchitectures are better. If you want
something small, maybe look for the ARM Cortex M0 or one of the small
RISC-V cores.

- anton
--
M. Anton Ertl http://www.complang.tuwien.ac.at/anton/home.html
comp.lang.forth FAQs: http://www.complang.tuwien.ac.at/forth/faq/toc.html
New standard: https://forth-standard.org/
EuroForth 2025 CFP: http://www.euroforth.org/ef25/cfp.html
EuroForth 2025 registration: https://euro.theforth.net/
--- Synchronet 3.21a-Linux NewsLink 1.2

From Newsgroup: comp.lang.forth

Paul Rubin <no.email@nospam.invalid> writes:

Succesful Linux
CPU's have quite a lot of registers and while fancy Forth compilers can
make use of them, you end up with an abstraction inversion

That's what you have claimed repeatedly. That would be true if the
programmers actually wanted a register machine and simulated that in
some way with the stack. That's not what is happening. The stack is
a good abstraction for passing operands and parameters. That's why it
is used in Forth, in the JavaVM, in WebAssembly, and elsewhere. The
stack is not a good abstraction for dealing with lots of values,
that's why Forth has locals, the Java VM has local variables, and
WebAssembly has locals.

- anton
--
M. Anton Ertl http://www.complang.tuwien.ac.at/anton/home.html
comp.lang.forth FAQs: http://www.complang.tuwien.ac.at/forth/faq/toc.html
New standard: https://forth-standard.org/
EuroForth 2025 CFP: http://www.euroforth.org/ef25/cfp.html
EuroForth 2025 registration: https://euro.theforth.net/
--- Synchronet 3.21a-Linux NewsLink 1.2

From Newsgroup: comp.lang.forth

On 12/2/2025 10:12 AM, Anton Ertl wrote:

The Pentium 8505 is an Alder Lake U with half of it's P-cores and half
of its E-Cores disabled, and some of the cache, leaving 1 P-Core and 4 E-cores. If you want the best performance, there are certainly a
number of microarchitectures by AMD, Apple, ARM, Intel, and Qualcomm
that compete for that spot and show fascinating feats in clock rate
and instructions per cycle (IPC), and the cores in the Pentium 8505
are not far behind; newer microarchitectures are better. If you want something small, maybe look for the ARM Cortex M0 or one of the small
RISC-V cores.

Thanks Anton. May I ask if you have a particular CPU that you find
pleasurable to use in your daily work?

--- Synchronet 3.21a-Linux NewsLink 1.2

From Newsgroup: comp.lang.forth

On 12/2/25 12:25 AM, Paul Rubin wrote:

You might like the TI MSP430 which is a 16-bit MCU line inspired by the PDP-11. It can't run Linux and it's now mostly eclipsed by ARM Cortex-M
and maybe soon Risc-V MCU's, but its CISC-style addressing modes are convenient for Forth, and its hardware implementations have some neat features like built in non-volatile (ferromagnetic) RAM in some models.

I ported eForth to an FRAM MSP430:

http://davesrocketworks.com/electronics/msp430/eforth/index.html


Having fiddled with ARM (an eForth for that too) I can safely say that
the MSP430 is better for low power applications. You can put an ARM into something resembling a low power mode but it takes forever to start up
again. At least in comparison to the MSP430.
--
http://davesrocketworks.com
David Schultz
"The cheaper the crook, the gaudier the patter." - Sam Spade
--- Synchronet 3.21a-Linux NewsLink 1.2

From Newsgroup: comp.lang.forth

In article <2025Dec2.193708@mips.complang.tuwien.ac.at>,
Anton Ertl <anton@mips.complang.tuwien.ac.at> wrote:

Paul Rubin <no.email@nospam.invalid> writes:

Succesful Linux
CPU's have quite a lot of registers and while fancy Forth compilers can >>make use of them, you end up with an abstraction inversion

That's what you have claimed repeatedly. That would be true if the >programmers actually wanted a register machine and simulated that in
some way with the stack. That's not what is happening. The stack is
a good abstraction for passing operands and parameters. That's why it
is used in Forth, in the JavaVM, in WebAssembly, and elsewhere. The
stack is not a good abstraction for dealing with lots of values,
that's why Forth has locals, the Java VM has local variables, and
WebAssembly has locals.

In riscv you can spare 25+ registers to hold LOCAL to-values.
As long as you take care of the chance of recursion,
can be very efficient.

- anton

Groetjes Albert
--
The Chinese government is satisfied with its military superiority over USA.
The next 5 year plan has as primary goal to advance life expectancy
over 80 years, like Western Europe.
--- Synchronet 3.21a-Linux NewsLink 1.2

From Newsgroup: comp.lang.forth

In article <%OJXQ.110578$mEQ5.8101@fx37.iad>,
David Schultz <david.schultz@earthlink.net> wrote:

On 12/2/25 12:25 AM, Paul Rubin wrote:

You might like the TI MSP430 which is a 16-bit MCU line inspired by the
PDP-11. It can't run Linux and it's now mostly eclipsed by ARM Cortex-M
and maybe soon Risc-V MCU's, but its CISC-style addressing modes are
convenient for Forth, and its hardware implementations have some neat
features like built in non-volatile (ferromagnetic) RAM in some models.

I ported eForth to an FRAM MSP430:

http://davesrocketworks.com/electronics/msp430/eforth/index.html

Having fiddled with ARM (an eForth for that too) I can safely say that
the MSP430 is better for low power applications. You can put an ARM into >something resembling a low power mode but it takes forever to start up
again. At least in comparison to the MSP430.

Small wonder. The PSP430 was especially designed for low power.
Willem Ouwerkerk has made applications that uses a sub uA current drain.

David Schultz

Groetjes Albert
--
The Chinese government is satisfied with its military superiority over USA.
The next 5 year plan has as primary goal to advance life expectancy
over 80 years, like Western Europe.
--- Synchronet 3.21a-Linux NewsLink 1.2

From Newsgroup: comp.lang.forth

Buzz McCool <buzz_mccool@yahoo.com> writes:

Thanks Anton. May I ask if you have a particular CPU that you find >pleasurable to use in your daily work?

AMD64 implementations by AMD and Intel are fast and run the OS of my
choice (GNU/Linux), which cannot be said of ARM A64 or RISC-V
implementations (either much slower or lack GNU/Linux support).

- anton
--
M. Anton Ertl http://www.complang.tuwien.ac.at/anton/home.html
comp.lang.forth FAQs: http://www.complang.tuwien.ac.at/forth/faq/toc.html
New standard: https://forth-standard.org/
EuroForth 2025 CFP: http://www.euroforth.org/ef25/cfp.html
EuroForth 2025 registration: https://euro.theforth.net/
--- Synchronet 3.21a-Linux NewsLink 1.2

From Newsgroup: comp.lang.forth

David Schultz <david.schultz@earthlink.net> wrote:

On 12/2/25 12:25 AM, Paul Rubin wrote:

You might like the TI MSP430 which is a 16-bit MCU line inspired by the
PDP-11. It can't run Linux and it's now mostly eclipsed by ARM Cortex-M
and maybe soon Risc-V MCU's, but its CISC-style addressing modes are
convenient for Forth, and its hardware implementations have some neat
features like built in non-volatile (ferromagnetic) RAM in some models.

I ported eForth to an FRAM MSP430:

http://davesrocketworks.com/electronics/msp430/eforth/index.html

Having fiddled with ARM (an eForth for that too) I can safely say that
the MSP430 is better for low power applications. You can put an ARM into something resembling a low power mode but it takes forever to start up again. At least in comparison to the MSP430.

AFAIK if you are satisfied with default frequency from RC oscillator,
then mainstream ARM-s start up pretty fast. MSP430 is nice because
afer startup you can quickly switch to a different frequency, but
for many purposes default frequency is OK. I did a silly LED blinker
where LED is powered from separate source and MCU is responsible for
timing and that needed about 10 uA for MCU. If you really care
about low power you should use low-power MCU-s, like STM32L series
or MSP432. Old STM32L could operate at about 4uA (measured current
including LCD display).

For me important limitation of low power modes was that in typical old
MCU-s (and that includes MSP430 that I have) the only hardware active
in low-power mode is real time clock and a bunch of wakeup pins.
Modern MCU-s are much better in this aspect, there are everal low-power
devices which draw very little power and can be active while main clock
is switched off.
--
Waldek Hebisch
--- Synchronet 3.21a-Linux NewsLink 1.2

From Newsgroup: comp.lang.forth

In article <2025Dec3.154202@mips.complang.tuwien.ac.at>,
Anton Ertl <anton@mips.complang.tuwien.ac.at> wrote:

Buzz McCool <buzz_mccool@yahoo.com> writes:

Thanks Anton. May I ask if you have a particular CPU that you find >>pleasurable to use in your daily work?

AMD64 implementations by AMD and Intel are fast and run the OS of my
choice (GNU/Linux), which cannot be said of ARM A64 or RISC-V
implementations (either much slower or lack GNU/Linux support).

This is changing fast. I have the Orange pi RV2 , (riscv)
and it has a 8 core over 1 Ghz CPU, and run several brands of GNU Linux,
e.g.

Welcome to Orange Pi 1.0.0 Noble with Linux 6.6.63-ky

I do development work on ciforth RISCV 64, and it generates the
html postscript pdf and info files allright.

It doesnot compare in speed to my 25 Kg HP workstation, but fits in the
palm of you hand and does a decent job of all mundane tasks, such as
showing videos.
It has support for AI engines build in, however I have not succeeded
in trying that out.

- anton

Groetjes Albert
--
The Chinese government is satisfied with its military superiority over USA.
The next 5 year plan has as primary goal to advance life expectancy
over 80 years, like Western Europe.
--- Synchronet 3.21a-Linux NewsLink 1.2

From Newsgroup: comp.lang.forth

albert@spenarnc.xs4all.nl writes:

In article <2025Dec3.154202@mips.complang.tuwien.ac.at>,
Anton Ertl <anton@mips.complang.tuwien.ac.at> wrote:

AMD64 implementations by AMD and Intel are fast and run the OS of my
choice (GNU/Linux), which cannot be said of ARM A64 or RISC-V >>implementations (either much slower or lack GNU/Linux support).

This is changing fast. I have the Orange pi RV2 , (riscv)
and it has a 8 core over 1 Ghz CPU, and run several brands of GNU Linux,
e.g.

Welcome to Orange Pi 1.0.0 Noble with Linux 6.6.63-ky

Looking at <http://www.orangepi.org/html/hardWare/computerAndMicrocontrollers/details/Orange-Pi-RV2.html>, it says:

|Provides 40K+DMIPS CPU arithmetic and 2TOPS AI arithmetic, single-core
|CPU arithmetic is more than 130% of ARM A55.

And that's pretty much all about the CPU capabilities it says. It
does not even say which "ARM A55" it is comparing to, especially not
the clock rate. It also does not specify the clock rate of their own
CPU.

Concerning the "40K+DMIPS" number, <https://www.weathink.com/news/hangye/26.html> says:

Cortex-A55 3.0 DMIPS/MHz
Cortex-A76 10.7-12.4 DMIPS/MHz

So according to this claim, the Orange pi RV2 performs like one
13333MHz Cortex-A55 core, or maybe the 8 cores perform like 8 1663MHz Cortex-A55 vores. Meanwhile, the Radxa Rock5b that we have had for a
while has 4 2200MHz Cortex-A76 cores and 4 1800MHz Cortex-A55 cores,
with an overall 115K DMIPS (Using the smaller number for the DMIPS/MHz
of the Cortex-A76.

The Rock5b is nice compared to the other SBCs we have, but it does not
compare to the AMD and Intel machines we have (the numbers are times
in seconds, i.e., smaller numbers are better):

sieve bubble matrix fib fft release; CPU; gcc
0.020 0.021 0.012 0.027 0.015 20250115; AMD Ryzen 8700G 5000MHz; gcc-12.2.0 0.027 0.033 0.011 0.043 0.017 20250115; Golden Cove 3.8GHz (Core i3-1315U); gcc-10.2.1
0.046 0.086 0.025 0.077 0.031 20250115; Gracemont 2.6GHz (Core i3-1315U); gcc-10.2.1
0.103 0.120 0.040 0.094 0.039 20250219; Rock 5B RK3588 A76 2220MHz; gcc-9.2.1 0.205 0.234 0.136 0.270 0.134 20250219; Rock 5B RK3588 A55 1800MHz; gcc-9.2.1

So a 5000MHz Zen 4 core is about 10 times as fast as a 1800MHz
Cortex-A55. If a core on the Orange Pi RV2 performs like a 1663MHz
Cortex-A55, the Zen4 core is more than 10 times as fast as such a
core, and the 8 cores of the Ryzen 8700G are also more than 10 times
as fast as the 8 cores of the Orange Pi RV2.

And a 5000MHz Zen4 core is still 2.5-5 times faster than a 2220MHz
Cortex-A76 on a Rock 5B.

While reading up on the Orange Pi RV2, I came across an advert for the
Radxa Orion O6, which appears to be more powerful than the Rock5b, but
still quite a bit below what Intel and AMD offer, and and relatively
expensive.

It doesnot compare in speed to my 25 Kg HP workstation, but fits in the
palm of you hand

The NUC that produced the Golden Cove results also fits in your hand,
although it's not quite as small as a Raspi.

- anton
--
M. Anton Ertl http://www.complang.tuwien.ac.at/anton/home.html
comp.lang.forth FAQs: http://www.complang.tuwien.ac.at/forth/faq/toc.html
New standard: https://forth-standard.org/
EuroForth 2025 CFP: http://www.euroforth.org/ef25/cfp.html
EuroForth 2025 registration: https://euro.theforth.net/
--- Synchronet 3.21a-Linux NewsLink 1.2

From Newsgroup: comp.lang.forth

On 12/4/2025 9:26 AM, Anton Ertl wrote:

sieve bubble matrix fib fft release; CPU; gcc
0.020 0.021 0.012 0.027 0.015 20250115; AMD Ryzen 8700G 5000MHz; gcc-12.2.0 0.027 0.033 0.011 0.043 0.017 20250115; Golden Cove 3.8GHz (Core i3-1315U); gcc-10.2.1

In comparing these two CPUs, it appears to me that single thread performance is the
key factor.

https://www.cpubenchmark.net/cpu.php?cpu=AMD+Ryzen+7+8700G&id=5836 https://www.cpubenchmark.net/cpu.php?cpu=Intel+Core+i3-1315U&id=5300




--- Synchronet 3.21a-Linux NewsLink 1.2

From Newsgroup: comp.lang.forth

Buzz McCool <buzz_mccool@yahoo.com> writes:

On 12/4/2025 9:26 AM, Anton Ertl wrote:

sieve bubble matrix fib fft release; CPU; gcc
0.020 0.021 0.012 0.027 0.015 20250115; AMD Ryzen 8700G 5000MHz; gcc-12.2.0 >> 0.027 0.033 0.011 0.043 0.017 20250115; Golden Cove 3.8GHz (Core i3-1315U); gcc-10.2.1

In comparing these two CPUs, it appears to me that single thread performance is the
key factor.

These are single-thread benchmarks, so of course single-thread
performance is the key factor for performing well on them.

- anton
--
M. Anton Ertl http://www.complang.tuwien.ac.at/anton/home.html
comp.lang.forth FAQs: http://www.complang.tuwien.ac.at/forth/faq/toc.html
New standard: https://forth-standard.org/
EuroForth 2025 CFP: http://www.euroforth.org/ef25/cfp.html
EuroForth 2025 registration: https://euro.theforth.net/
--- Synchronet 3.21a-Linux NewsLink 1.2

From Newsgroup: comp.lang.forth

On 12/5/2025 9:53 AM, Anton Ertl wrote:

These are single-thread benchmarks, so of course single-thread
performance is the key factor for performing well on them.

Got it.

So in my case an elegant CPU might boil down to something high on
this list with a low price tag.

https://www.cpubenchmark.net/single-thread/



--- Synchronet 3.21a-Linux NewsLink 1.2

From Newsgroup: comp.lang.forth

Buzz McCool <buzz_mccool@yahoo.com> writes:

So in my case an elegant CPU might boil down to something high on
this list with a low price tag.

https://www.cpubenchmark.net/single-thread/

If your goal is to have a good single-thread CPU Mark score, yes.

In general, look for benchmarks that reflect what you are doing.

- anton
--
M. Anton Ertl http://www.complang.tuwien.ac.at/anton/home.html
comp.lang.forth FAQs: http://www.complang.tuwien.ac.at/forth/faq/toc.html
New standard: https://forth-standard.org/
EuroForth 2025 CFP: http://www.euroforth.org/ef25/cfp.html
EuroForth 2025 registration: https://euro.theforth.net/
--- Synchronet 3.21a-Linux NewsLink 1.2