I want to know your opinion about multi CPUs
ZXNet echo conference «code.zx»
Hello, All
It is possible to create a multiprocessor system, you can read more
here - http://zx.pk.ru/showpost.php?p=71328&postcount=30
["http://zx.pk.ru/showpost.php?p=71328&postcount=30"]
I would like to know your opinion whether this technology will be in demand among
programmers, if an additional mode appears up to 640*480 (on VGA
monitor)?
Hello MegaMyth
I don’t see the need for more than 2 processors,
and this is something else...
Hello MegaMyth
IMHO, multiprocessing support is primarily an OS issue. And the corresponding
The section on the site is the most overgrown with cobwebs. :) So I don’t know, I don’t know...
And 640x480 will make an excellent text screen, which is definitely good.
Hello ASDT
In my opinion, the term multiprocessing means something different than what
via the link...
Hello, ng_dead
By the way - and in principle - what prevents you from making SMP on spec? I think nothing. Yes
ZX-Next - there are already two CPUs and both have equal access to RAM.
Of course, any SMP system has a head CPU, no one denies this.
Hello Sonic
Working on shared memory is one thing, but parallelizing calculations (and executable
program) this is for Babayan :) There is also a feature from the time of HP calculators (and before
of our days, graph-physics-processes) separation of functionality into separate modules.
Central - supervisor.
Hello Sonic
Son> By the way - and in principle - what prevents you from making SMP on spec? I think
Son> nothing. There is ZX-Next - there are already two CPUs and both have equal access to
Son> RAM.
Son>
For several CPUs without a cache, working with single-port memory is
most of the time in line. Those. compared to a single-processor system with
switched OS multitasking, such an N-processor system only benefits
context switching time - 10...15%. But to program it at the lower level -
you'll be tormented.
Hello MegaMyth
Son> what prevents you from making SMP on spec? I think nothing. There is ZX-Next -
Son> there are already two CPUs and both have equal access to RAM.
There is no such thing there.
The VCPU has its own ROM for code, and does not interfere with shared memory.
He himself does not read anything from memory. He's just running around in his ROM. Image
formed by external circuits. Some bits of its address bus are used
to form an address in video memory.
Hello Sonic
Son> There is ZX-Next - there are already two CPUs and both have equal access to RAM.
there is no such thing there, rtfm.
Son> Existing no-wait ZX clones take advantage of the fact that
Son> memory works much faster than Z80.
much faster is greatly exaggerated. in the Pentagon128 the fact is used
that the video controller reads memory in free time from the CPU. those. processor like
would be more important and never waits, there are practically no extra clock cycles of the video controller.
Son> Why not CPU0, CPU1, CPU2, ...CPUn, Video? IMHO it limits us
Son> memory speed only.
everything is possible. you just need to hang each processor with a bunch of registers and take
fairly fast memory.
Iron is not really a problem in this matter. the main thing is who writes this
software for everything to work smoothly and use these resources 100%, you need
good theoretical training and a powerful compiler, which is not available.
Hello, Error404
Err> For several CPUs without a cache, working with single-port memory is
Err> spend most of your time in queue.
Err>
Why?
Existing no-wait ZX clones take advantage of the fact that memory
works much faster than Z80. Z80 and video controller access memory by
queues. Right?
Why only two devices? Why not CPU0, CPU1, CPU2, ...CPUn, Video?
IMHO we are limited only by memory speed.
Or am I doing something wrong?
> But programming it at the lower level will be a pain in the ass.
Why?
Let's imagine such a ZX. At power-up, only CPU0 is active;
the rest are pressed Reset. To start any other CPU you need to enter address 0
instruction JP and release it Reset.
To do this you need two options:
1) At address 0 we have RAM
2) The machine has more than 64 KB of memory and each percent works in its own
segment (may intersect). The base addresses of the segments are assigned to CPU0.
Why not?
Damn, my hands are itching to finish my Pentagon-128...
Hello Vitamin
Vit> And now let's bend our fingers on the existing tasks that require more
Vit> of one processor. How long will we stand with our arms outstretched? :)
Nope. You just named three yourself.
Hello Sonic
Son> And the native system could be 24-bit IS-DOS + EZX-Basic (such
Son> development is underway).
+Basic again? Oh. Guard. Come to your senses before it's too late. :v2_wink2:
Hello hero
her> Well, how do you put a 12 MHz processor + 3.5/7 MHz processor on one bus
her> so that they are SMP and even on memory with bandwidth
her> 1-2Mb/s no more
Not 12 but 14. And the second CPU is 3.5/14. In the latest Profi, the Z80 worked great on
14 MHz as far as I remember.
3.5 is a multiple of 14 so everything fits perfectly. And memory - what, more modern?
Why can’t RU5 take anything?
Although, of course, if we’re talking about such conceptual things, then it’s better
first make a scalable architecture on the Z180/eZ80 that has a mode
compatibility with the old Spec...
CP/M works great on these processors. I think that Spectrum OS is also possible. A
the native system could be 24-bit IS-DOS + EZX-Basic (such a development
is underway).
Hello Vitamin
Vit> I named those tasks where multiprocessing is possible
Vit> but how far from being the most effective option, I therefore indicated possible ones
Vit> solutions
Circuit design SMP in this case is the most universal and simple solution.
Let's remember the sophisticated DMA USC circuit, but almost everything is replaced
one CPU. And the CPU has more capabilities.
Hello Vitamin
Vit> Two. But with a caveat.
Vit>
1) Sound
2) Graphics
3) DMA. Why is the CPU bad for you for transferring data and at the same time implementing
any operations with them? At one time, the blitter helped a lot on the Amiga. However
(3) combines with (2), so let there be two...
It’s a pity that each pixel doesn’t have its own color. It would be possible without special
At least have some problems with 3D...
> Yes, and I clarified the existing tasks.
So you named the missing tasks? :)
Hello, Sonic
Son> Hе 12 а 14.
Hе 14 а 12 ;)
Hello Sonic
Well, how do you put a 12 MHz processor + 3.5/7 MHz processor on one bus so that they
there were SMP and even on memory with a bandwidth of 1-2Mb/s no more
Hello, The Exploited
Now let’s get our fingers crossed for the existing tasks that require more than one
processor. How long will we stand with our arms outstretched? :)
The only least hemorrhoidal application of multiprocessing is devices with
its memory and communication with the host through ports (GS for example). It may also
be the same video adapter, but this is already esoteric...
On the good side, DMA and an interrupt controller are missing. Because they (especially DMA)
have much greater applicability and usefulness.
Hello Sonic
Son> However (3) is combined with (2), so let there be two...
(3) is also combined with (1) and in the most direct way! Even closer
than with (2).
Son> So you named the missing tasks?
I named those tasks where it is possible to use multiprocessing, but how
is far from the most effective option, therefore I indicated possible solutions
Hello Sonic
Son> Nope. You just named three yourself.
Two. But with a caveat. Yes, and clarified the existing tasks. But we essentially have no tasks
There are those where multiprocessing as such is desperately needed.
Hello, Error404
Err> +Basic again? Oh. Guard. Come to your senses before it's too late.
Well... It's just that some people argue that "Spec should allow you to learn
program immediately after turning on the power."
And by the way, there is something about a BASIC-based OS... A kind of spirit
retro-home-computing...
You can make it more exotic - Firmware based on Forth. Maybe ieee1275
make fun :)
Hello Sonic
Son> Let's remember the sophisticated DMA USC circuit, but that's almost all
Son> is replaced by one CPU
her> Well, how do you put a 12 MHz processor on one bus
In GS 12MHz percent (DMA USC was without a processor)
Hello hero
her> Not 14 but 12
Why?
Hello, Error404
Err> In principle, even BASIC has a right to exist. The main thing is that
Err> it was not a necessary evil - so that it would be possible to turn off the ROM from
Err> address space to have all 64k as RAM. And somehow
Err> at the start do not immediately fall into BASIC, but first, for example, try
Err> boot from disk. Or have a custom bootlist.
If I'm not mistaken, the first personal computers of the IBM PC (XT?) had just such a basic
in ROM, which started in the absence of boot disks.
Hello Sonic
Son> Well... It's just that some people argue that "the spec should allow
Son> learn to program immediately after turning on the power."
Son> And by the way, there is something in an OS based on BASIC... A kind of spirit
Son> retro-home-computing...
Son> You can make it more exotic - Firmware based on Forth. You can finally
Son> ieee1275 make fun. :)
OpenFirmware. :)
In principle, even BASIC has a right to exist. The main thing is that he is not
inevitable evil - to be able to turn off the ROM from the address
space to have all 64k as RAM. And somehow at the start it doesn’t fall out
straight into BASIC, and first, for example, try to boot from disk. Or have
custom bootlist.
Hello,GriV
Gri> I'll do a little off-topic.
Gri> In the current discussion, technical problems have so far been discussed
Gri> how memory will work, etc., but what about support again
Gri> such a system? Why would it be possible to use two
Gri> processor? Purely theoretically - every processor should have
Gri> with a separate address-page space, as expected
Gri> implement? What will this do for the existing software park?
This will do nothing for existing software. Software that supports new features is needed
will do it on his own.
Hello, Error404
Err> In principle, even BASIC has a right to exist. The main thing is that
Err> it was not a necessary evil - so that it would be possible to turn off the ROM from
Err> address space to have all 64k as RAM.
Err>
There are such cars. In theory, existing ones can be upgraded very easily - just
disabling the ROM is added - and that’s it.
> And somehow at the start it doesn’t fall straight into BASIC, but first, to
> for example, try to boot from disk. Or have a custom one
> bootlist.
Or an alternative is to intelligently integrate loading into BASIC. RANDOMIZE USR
15619:REM:RUN - I agree this is inconvenient. Look at the imported disk interfaces
everything rules. In PlusD you simply type RUN if there is no program in memory - and
order. In D80 you don’t need to type anything at all - just BASIC expands and
everything. Begins to understand constructs like LOAD"A:Program". As indeed it should
be.
Apparently, the creators of the beta disk were guided by the same considerations as
and the creators of Soviet Speck clones: “it doesn’t matter how well, the main thing is how
"As little money as possible."
Hello,GriV
Why does Spectrum need two (three, four, one hundred) processors? It's better to install one, but
good!
Hello Sonic
I'll do a little offtopic.
I'm looking at the existing x86 and x64 architecture. I look at them
multiprocessor solutions (2, 4 or more processors on one motherboard
board). And I can draw this conclusion: for this extremely supported
platform money, while multi-core is a kind of “bonus”, due to
which can slightly increase the performance of some applications.
The only exception is the server market, but its cost (as
probably any server solution) makes you want to howl at the moon at night
sky.
That. even in the PC platform against the background of multitasking/multithreaded OSes yet
multi-core is just a bonus.
Now I am returning from the sinful earth under the clouds of our specks.
Currently, I would really like to develop the platform, but I don’t see the point
develop it in such a revolutionary way. Personally, I am NOT against multiprocessing, but 1)
on which platforms this configuration will be implemented 2) how it will be
a multiprocessing manager is organized 3) how to manage these managers 4)
assumes symmetric peer-to-peer multiprocessing system or something
different?
In the current discussion, technical problems have so far been discussed, how will
work memory, etc., but what about supporting such a system again? for
why would it be possible to use two processors? Purely theoretically -each processor must have a separate address-page space,
How is this supposed to be implemented? What will this do for the existing software fleet?
Hello, Jukov
One good processor costs n bad ones and vice versa.
In general, it's not about the processors. The sound of the ZX is ALREADY not bad (for the capabilities of the core
Spectrum). Graphics... haven't seen any solid standards yet. There are interruptions, there are
procedures, but multi-core is not an indicator.
Hello TomCaT
Tom> One good processor costs n bad ones and vice versa.
Tom> In general, it's not about the processors. The sound of the ZX is ALREADY good (for
Tom> capabilities of the Spectrum core). Graphics... haven't seen any solid ones yet
Tom> standards. There are interrupts, there are procedures, but multi-cores are not
Tom> indicator.
As for the spec, you should under no circumstances change the Z80 IMHO, this is quite
a powerful and modern processor that allows the platform to avoid the effect
present on the pussy - "increasing computing resources
compensated by the degradation of the developers." Current Spectrum graphics mode
is also quite digestible.. you need to think in general, and not stick powerful “stones” on
each function and copy the solutions from the pussy..
Hello, Error404
Err> Well, “powerful and modern” is, of course, unlikely - now anyone
Err> the refrigerator will plug the Z80 with its built-in controller
Maybe a modern controller will shut up, of course... but I have a writing one
a sophisticated Philips DVD player... I suspect that its controller is no worse
than in an ultra-modern refrigerator .. but it has a turn-on time
shows what the crooked eccentric wrote in a BASIC interpreter (written
crooked by another eccentric) a program for it, and together these eccentrics brought together
performance up to one megahertz 4004... Powerful and modern is when
devices with similar performance are used in military, space and
medical industries and perform tasks quite satisfactorily :) and the Z80 is
satisfies the requirement
Err> But the standard graphics mode is already below the minimum. Nothing but games
Err> doesn’t look good on such a screen - it’s just not enough at the same time
Err> of the information present.
graphic mode is yes.. a sore spot of the platform...
Hello Raydac
Ray> Regarding the spec, you should under no circumstances change the Z80 IMHO, this
Ray> is quite a powerful and modern processor,
Ray>
Well, “powerful and modern” is, of course, unlikely - now any refrigerator
its built-in controller will shut up the Z80 [exaggerating :cool; ]; The point is different:
For a hobby you don’t need “powerful and modern”. It's just the opposite that's interesting -
create something interesting in a minimalist (within reasonable limits) environment.
Ray> The current Spectrum graphics mode is also quite digestible.. think
Ray> needs to be done, and not stick powerful “stones” into each function and
Ray> copy solutions from pussy..
Ray>
But the standard graphics mode is already below the minimum. Nothing but games on this
It doesn’t look good on the screen - there’s just not enough simultaneous presence
information. And games... well, I personally have never been interested in them, let alone
modern environment, when any phone with a toad has more content in this
I mean, even more so.