Rick C <gnuarm.deletethisbit@gmail.com> writes:
> ... The GA144 was a
> technology that required a solution to how to use it before anyone could
> consider what apps it might be used for. In other words, it was a solution
> looking for a problem and never really found a good match.

I briefly, with my network processor hat on, looked at how to morph GA144 to
become a decent solution. The short version was: nothing which the people
with their hand on the GA144 rudder would ever consider.

My memory is that each node, after overhead, is good for less than 100 bytes
of RAM? A GA144 degenerates into a really hard to use 14kb byte part. It
has some true innovations, but new ideas need to morph as they meet reality,
and I never saw that happen.

Andy Valencia
Home page: https://www.vsta.org/andy/
To contact me: https://www.vsta.org/contact/andy.html

Wayne morellini <waynemorellini@gmail.com> writes:
> [Quantum dots] But, you literally could fit in maybe 1 million more
> circuitry for the same peak energy load... These guys are talking
> about 10's of Thz. You max out most applications. But, I see no
> reason that it doesn't suit a stack machine.

What I mean is that this is science fiction technology at the moment.
If it becomes viable for microprocessors, then it would presumably be
fine for stack machines, but also for non-stack machines.

On Friday, December 3, 2021 at 1:40:16 PM UTC-5, Paul Rubin wrote:
> Wayne morellini <waynemo...@gmail.com> writes:
> > [Quantum dots] But, you literally could fit in maybe 1 million more
> > circuitry for the same peak energy load... These guys are talking
> > about 10's of Thz. You max out most applications. But, I see no
> > reason that it doesn't suit a stack machine.
> What I mean is that this is science fiction technology at the moment.
> If it becomes viable for microprocessors, then it would presumably be
> fine for stack machines, but also for non-stack machines.

I agree, but I think the real issue is why chase pie in the sky implementations at 10s of THz when just a few GHz would be a significant improvement? It is normal for technology to proceed in steps rather than great leaps forward.

In general the world has rejected stack machines for many, many years. What stack machines really need is a stack machine application, not a stack machine implementation.

--

Rick C.

-++ Get 1,000 miles of free Supercharging
-++ Tesla referral code - https://ts.la/richard11209

On Saturday, December 4, 2021 at 4:07:09 AM UTC+10, Andy Valencia wrote:
> Rick C <gnuarm.del...@gmail.com> writes:
> > ... The GA144 was a
> > technology that required a solution to how to use it before anyone could
> > consider what apps it might be used for. In other words, it was a solution
> > looking for a problem and never really found a good match.
> I briefly, with my network processor hat on, looked at how to morph GA144 to
> become a decent solution. The short version was: nothing which the people
> with their hand on the GA144 rudder would ever consider.
>
> My memory is that each node, after overhead, is good for less than 100 bytes
> of RAM? A GA144 degenerates into a really hard to use 14kb byte part. It
> has some true innovations, but new ideas need to morph as they meet reality,
> and I never saw that happen.
>
> Andy Valencia
> Home page: https://www.vsta.org/andy/
> To contact me: https://www.vsta.org/contact/andy.html

Thanks Andy.

I've also tried in the past. I've successfully done it with top companies, seeing good changes to ussability of products. But when people get into their own thinking, and misconstrue things, it's time to walk. I was dieing, so why waste my time.

Minimalist equals nick nack. You have to have a balanced level of minimalism to get the maximum benefit. I even came up with minimalist proposals for multi tier memory interface, cross linked auto communications, and simple caching, replacing lots of instruction execution on more advanced work loads, for a lot less energy. If only one chip had a direct execution 640KB address space (the external cache memory chip spoken of decades back), as I hoped, it would have made things a lot better. I've had a problem with brain disease, so I'm not good like I used to be, but have spent a lot of time in the past exploring alternative execution design. But I've considered a 8bit alternative to the chip, with a lot of execution features. But, I can't remember, them, but I think I have a design document somewhere This was accompanied by 16 and 32 bit versions, from memory. It's all rubbish after I tried to design decimal logic gates instead years ago. That's something I've held close to my chest, but the damage was setting in. The way I was trying to do the levels was difficult. However, my collection of design mechanisms, would make a great processor, even today. I still think a 10Ghz+ could be done on silicon.

It has been the easiest time for GA to design a successful processor. All they needed to do was to target android and JavaScript, and sell into that market. That market covers many devices. So, while they might not have been producing top mobile phone chips, there are lesser products that use java/JavaScript, even cheap phone chipsets After sales in those areas, they could have afforded to move up to better mobile chipsets.

I can tell you one thing. For the workloads they are looking at with the glasses. They should be dusting off the advanced 32 bit processor proposal. Which would be something you guys are interested in!

Wayne.

On Saturday, December 4, 2021 at 7:30:47 AM UTC+10, gnuarm.del...@gmail.com wrote:
> On Friday, December 3, 2021 at 1:40:16 PM UTC-5, Paul Rubin wrote:
> But, I see no
> > > reason that it doesn't suit a stack machine.
> > What I mean is that this is science fiction technology at the moment.
> > If it becomes viable for microprocessors, then it would presumably be
> > fine for stack machines, but also for non-stack machines.
> I agree, but I think the real issue is why chase pie in the sky implementations at 10s of THz when just a few GHz would be a significant improvement? It is normal for technology to proceed in steps rather than great leaps forward.
>
> In general the world has rejected stack machines for many, many years. What stack machines really need is a stack machine application, not a stack machine implementation.
>
> --

I'm not saying that they should do a Thz processor. I'm saying that QDCA is a descent bet. That they should look at the 500mhz+ advances and do a few GHz version. When the Thz, if it ever does, gets worked out,, they can move onto that. No science fiction involved, it's what scientists are actually working towards. It's business, you plan for the future and take a bet on which direction to start taking steps into. It's often not clear when using external innovations.

Stack machines aren't the problem. Shboom, and rtx went on to have success.. It's implementation we are concerned about.

Rick C <gnuarm.deletethisbit@gmail.com> writes:
> In general the world has rejected stack machines for many, many years.
> What stack machines really need is a stack machine application, not a
> stack machine implementation.

Yeah the main attraction of stack machines (if there is one at all) is
very tiny implementation for a useful amount of processing. That can be
helpful for deeply embedded micropower things, or on resource limited
FPGA's. For example there's a new FPGA with 1000 LUT4's that will
supposedly cost around 50 cents per unit. I don't think a RISC-V soft
cpu will fit in there, but a single GA or b16-like cpu node might fit.
That would let you compute some stuff and also have enough left over for
some logic functions.

On Friday, December 3, 2021 at 6:35:47 PM UTC-5, Wayne morellini wrote:
> On Saturday, December 4, 2021 at 7:30:47 AM UTC+10, gnuarm.del...@gmail.com wrote:
> > On Friday, December 3, 2021 at 1:40:16 PM UTC-5, Paul Rubin wrote:
> > But, I see no
> > > > reason that it doesn't suit a stack machine.
> > > What I mean is that this is science fiction technology at the moment.
> > > If it becomes viable for microprocessors, then it would presumably be
> > > fine for stack machines, but also for non-stack machines.
> > I agree, but I think the real issue is why chase pie in the sky implementations at 10s of THz when just a few GHz would be a significant improvement? It is normal for technology to proceed in steps rather than great leaps forward.
> >
> > In general the world has rejected stack machines for many, many years. What stack machines really need is a stack machine application, not a stack machine implementation.
> >
> > --
> I'm not saying that they should do a Thz processor. I'm saying that QDCA is a descent bet. That they should look at the 500mhz+ advances and do a few GHz version. When the Thz, if it ever does, gets worked out,, they can move onto that. No science fiction involved, it's what scientists are actually working towards. It's business, you plan for the future and take a bet on which direction to start taking steps into. It's often not clear when using external innovations.

You talk about planning as if it were inevitable these things will get designed and built. Where does the money come from? With no track record to speak of the hard part is finding someone who wants to start spending millions and millions of dollars on totally unproven design ideas.

> Stack machines aren't the problem. Shboom, and rtx went on to have success. It's implementation we are concerned about.

What success? They may have found a few design wins. I think the RTX gets used in space apps because it is rad hard (very hard to come by in general). There's nothing about this pedigree that would attract the sort of investor who will pay for such grandiose chips.

--

Rick C.

+-- Get 1,000 miles of free Supercharging
+-- Tesla referral code - https://ts.la/richard11209

On Saturday, December 4, 2021 at 11:31:55 AM UTC+10, gnuarm.del...@gmail.com wrote:
> On Friday, December 3, 2021 at 6:35:47 PM UTC-5, Wayne morellini wrote:
> > On Saturday, December 4, 2021 at 7:30:47 AM UTC+10, gnuarm.del...@gmail..com wrote:
> > > On Friday, December 3, 2021 at 1:40:16 PM UTC-5, Paul Rubin wrote:
> > > But, I see no
> > > > > reason that it doesn't suit a stack machine.
> > > > What I mean is that this is science fiction technology at the moment.
> > > > If it becomes viable for microprocessors, then it would presumably be
> > > > fine for stack machines, but also for non-stack machines.
> > > I agree, but I think the real issue is why chase pie in the sky implementations at 10s of THz when just a few GHz would be a significant improvement? It is normal for technology to proceed in steps rather than great leaps forward.
> > >
> > > In general the world has rejected stack machines for many, many years.. What stack machines really need is a stack machine application, not a stack machine implementation.
> > >
> > > --
> > I'm not saying that they should do a Thz processor. I'm saying that QDCA is a descent bet. That they should look at the 500mhz+ advances and do a few GHz version. When the Thz, if it ever does, gets worked out,, they can move onto that. No science fiction involved, it's what scientists are actually working towards. It's business, you plan for the future and take a bet on which direction to start taking steps into. It's often not clear when using external innovations.
> You talk about planning as if it were inevitable these things will get designed and built. Where does the money come from? With no track record to speak of the hard part is finding someone who wants to start spending millions and millions of dollars on totally unproven design ideas.

Wherever it:s conventional silicon or not, it requires money. It's about where they will be in time, even 5. It's about commerce survival, and having the edge. They have to have something in offer for people to buy.

> > Stack machines aren't the problem. Shboom, and rtx went on to have success. It's implementation we are concerned about.
> What success? They may have found a few design wins. I think the RTX gets used in space apps because it is rad hard (very hard to come by in general). There's nothing about this pedigree that would attract the sort of investor who will pay for such grandiose chips.

Those where significant designs of their day. Again, there is only normal level design being talked about at this stage. If they earn money, then they can advance to better designs. At the moment, its not optimal, so something has to change to continue.

On Saturday, December 4, 2021 at 12:53:02 AM UTC+1, Paul Rubin wrote:
[..]
> Yeah the main attraction of stack machines (if there is one at all) is
> very tiny implementation for a useful amount of processing. That can be
> helpful for deeply embedded micropower things, or on resource limited
> FPGA's. For example there's a new FPGA with 1000 LUT4's that will
> supposedly cost around 50 cents per unit. I don't think a RISC-V soft
> cpu will fit in there, but a single GA or b16-like cpu node might fit.
> That would let you compute some stuff and also have enough left over for
> some logic functions.

By the time you have finished that, that FPGA has a successor with
16,000 LUT4's and as a company you have no follow-up to fend off
the RISC-V guys. There might be a possibility when a time comes
that it is impossible to increase instructions/Watt or the number
of LUTs/mm^2.

-marcel

Marcel Hendrix <mhx@iae.nl> writes:
> By the time you have finished that, that FPGA has a successor with
> 16,000 LUT4's

It's not a hugely complicated thing to put a small cpu and some logic
into an fpga. There will not be a 16000 LUT, 50 cent FPGA anytime soon.
There has not been such a thing as a 50 cent FPGA at all in the past, as
far as I know. When there is one with 16000 luts for 50 cents, the one
with 1000 luts will be 5 cents. This is still important for some kinds
of products. Yes you can buy 50 cent microprocessors with 10s of
kilobytes of ram, but if your application only needs a few dozen bytes,
you can use a 3 cent processor.

https://jaycarlson.net/2019/09/06/whats-up-with-these-3-cent-microcontrollers/

On Saturday, December 4, 2021 at 9:53:02 AM UTC+10, Paul Rubin wrote:
> Rick C <gnuarm.del...@gmail.com> writes:
> > In general the world has rejected stack machines for many, many years.
> > What stack machines really need is a stack machine application, not a
> > stack machine implementation.
> Yeah the main attraction of stack machines (if there is one at all) is
> very tiny implementation for a useful amount of processing. That can be
> helpful for deeply embedded micropower things, or on resource limited
> FPGA's. For example there's a new FPGA with 1000 LUT4's that will
> supposedly cost around 50 cents per unit. I don't think a RISC-V soft
> cpu will fit in there, but a single GA or b16-like cpu node might fit.
> That would let you compute some stuff and also have enough left over for
> some logic functions.

Paul, I like to look at things from different angles to get different modalities. To get results differently. For instance how fast can they make these things go these days? I remember the resistive alternative, which I was waiting to use, before it was taken off market, where 5Ghz decade before last, and it would seem to indicate it had possibly be head room to 20Ghz (how hot, I don't know). Such small dies allows massive heat sinking where you could overclock to get a basic 32 bit processor design up beyond the equivalent of gigahertz. Which is a much quicker way or achieve similar results to a doing a custom processor. (product with fpga to profit, to implement in the custom fpga processor in silicon). Many ways to skin a ... I had been waiting for the magneticQDCA FPGA instead.

On Friday, December 3, 2021 at 4:30:47 PM UTC-5, gnuarm.del...@gmail.com wrote:
> On Friday, December 3, 2021 at 1:40:16 PM UTC-5, Paul Rubin wrote:
> > Wayne morellini <waynemo...@gmail.com> writes:
> > > [Quantum dots] But, you literally could fit in maybe 1 million more
> > > circuitry for the same peak energy load... These guys are talking
> > > about 10's of Thz. You max out most applications. But, I see no
> > > reason that it doesn't suit a stack machine.
> > What I mean is that this is science fiction technology at the moment.
> > If it becomes viable for microprocessors, then it would presumably be
> > fine for stack machines, but also for non-stack machines.
> I agree, but I think the real issue is why chase pie in the sky implementations at 10s of THz when just a few GHz would be a significant improvement? It is normal for technology to proceed in steps rather than great leaps forward.
>
> In general the world has rejected stack machines for many, many years. What stack machines really need is a stack machine application, not a stack machine implementation.
>
> --
>
> Rick C.
>
> -++ Get 1,000 miles of free Supercharging
> -++ Tesla referral code - https://ts.la/richard11209

Does Java qualify as a stack machine application?

I remember that Sh Boom CPU was capable of running Java at near native speed.
Nobody cared or...
the entire goto market strategy was flawed and/or underfunded. (my guess)

Having lived on both the engineering dept. side and the "make $XX million with this
product or you are fired!" side, I find I have to remind the tech side that a better
product only gives you permission to begin doing business.
Business success comes from money and psychology
ie:the "art" of marketing (one to many)
and sales (one to one) communication.

On Friday, December 3, 2021 at 11:26:41 PM UTC-5, Wayne morellini wrote:
> On Saturday, December 4, 2021 at 11:31:55 AM UTC+10, gnuarm.del...@gmail.com wrote:
> > On Friday, December 3, 2021 at 6:35:47 PM UTC-5, Wayne morellini wrote:
> > > On Saturday, December 4, 2021 at 7:30:47 AM UTC+10, gnuarm.del...@gmail.com wrote:
> > > > On Friday, December 3, 2021 at 1:40:16 PM UTC-5, Paul Rubin wrote:
> > > > But, I see no
> > > > > > reason that it doesn't suit a stack machine.
> > > > > What I mean is that this is science fiction technology at the moment.
> > > > > If it becomes viable for microprocessors, then it would presumably be
> > > > > fine for stack machines, but also for non-stack machines.
> > > > I agree, but I think the real issue is why chase pie in the sky implementations at 10s of THz when just a few GHz would be a significant improvement? It is normal for technology to proceed in steps rather than great leaps forward.
> > > >
> > > > In general the world has rejected stack machines for many, many years. What stack machines really need is a stack machine application, not a stack machine implementation.
> > > >
> > > > --
> > > I'm not saying that they should do a Thz processor. I'm saying that QDCA is a descent bet. That they should look at the 500mhz+ advances and do a few GHz version. When the Thz, if it ever does, gets worked out,, they can move onto that. No science fiction involved, it's what scientists are actually working towards. It's business, you plan for the future and take a bet on which direction to start taking steps into. It's often not clear when using external innovations.
> > You talk about planning as if it were inevitable these things will get designed and built. Where does the money come from? With no track record to speak of the hard part is finding someone who wants to start spending millions and millions of dollars on totally unproven design ideas.
> Wherever it:s conventional silicon or not, it requires money. It's about where they will be in time, even 5. It's about commerce survival, and having the edge. They have to have something in offer for people to buy.

Sorry, I don't understand what you are saying, literally.

> > > Stack machines aren't the problem. Shboom, and rtx went on to have success. It's implementation we are concerned about.
> > What success? They may have found a few design wins. I think the RTX gets used in space apps because it is rad hard (very hard to come by in general). There's nothing about this pedigree that would attract the sort of investor who will pay for such grandiose chips.
> Those where significant designs of their day. Again, there is only normal level design being talked about at this stage. If they earn money, then they can advance to better designs. At the moment, its not optimal, so something has to change to continue.

Significant to whom? They were tiny blips on the RADAR screen. I'm still waiting for someone to show any real world advantages to stack processor chips in the real world of today.

The GA144 attempted to be the universal peripheral laden MCU, but failed in being an MCU at all. The proponents talked about how low power the individual processors were and the low power when not processing, but the programming was so complex they came up with a virtual machine implementation that negated the power savings. In fact, everything they provided was a dollar short of being useful to a user.

Partly, the problem of the GA144 was overcoming the entrenchment of conventional processors, but that is the world at this point. Talking about some power savings or speed advantage or even the flexibility of peripherals is not of much use if it solves a problem the users don't have while creating problems users don't have with conventional solutions.

So how would money be made from such designs?

--

Rick C.

+-+ Get 1,000 miles of free Supercharging
+-+ Tesla referral code - https://ts.la/richard11209

On Saturday, December 4, 2021 at 4:42:46 AM UTC-5, Marcel Hendrix wrote:
> On Saturday, December 4, 2021 at 12:53:02 AM UTC+1, Paul Rubin wrote:
> [..]
> > Yeah the main attraction of stack machines (if there is one at all) is
> > very tiny implementation for a useful amount of processing. That can be
> > helpful for deeply embedded micropower things, or on resource limited
> > FPGA's. For example there's a new FPGA with 1000 LUT4's that will
> > supposedly cost around 50 cents per unit. I don't think a RISC-V soft
> > cpu will fit in there, but a single GA or b16-like cpu node might fit.
> > That would let you compute some stuff and also have enough left over for
> > some logic functions.
> By the time you have finished that, that FPGA has a successor with
> 16,000 LUT4's and as a company you have no follow-up to fend off
> the RISC-V guys. There might be a possibility when a time comes
> that it is impossible to increase instructions/Watt or the number
> of LUTs/mm^2.

I think you missed something. He is not talking about arbitrary devices. He specifically mentioned a $0.50 FPGA. I don't know who's chip this is, but I vaguely recall someone announcing that. You won't find a 16,000 LUT FPGA at that price for more than a decade, even if then. Even when you put a RISC-V into a low priced FPGA you will find it runs much, much slower than a small stack machine and uses a lot more power. I believe there have been some impressive implementations of RISC-V in FPGAs that use very little space. However, they run very slowly.

--

Rick C.

++- Get 1,000 miles of free Supercharging
++- Tesla referral code - https://ts.la/richard11209

On Sunday, December 5, 2021 at 12:40:43 AM UTC+10, Brian Fox wrote:
> On Friday, December 3, 2021 at 4:30:47 PM UTC-5, gnuarm.del...@gmail.com wrote:
> > On Friday, December 3, 2021 at 1:40:16 PM UTC-5, Paul Rubin wrote:
> > > Wayne morellini <waynemo...@gmail.com> writes:
> > > > [Quantum dots] But, you literally could fit in maybe 1 million more
> > > > circuitry for the same peak energy load... These guys are talking
> > > > about 10's of Thz. You max out most applications. But, I see no
> > > > reason that it doesn't suit a stack machine.
> > > What I mean is that this is science fiction technology at the moment.
> > > If it becomes viable for microprocessors, then it would presumably be
> > > fine for stack machines, but also for non-stack machines.
> > I agree, but I think the real issue is why chase pie in the sky implementations at 10s of THz when just a few GHz would be a significant improvement? It is normal for technology to proceed in steps rather than great leaps forward.
> >
> > In general the world has rejected stack machines for many, many years. What stack machines really need is a stack machine application, not a stack machine implementation.
> >
> > --
> >
> > Rick C.
> >
> > -++ Get 1,000 miles of free Supercharging
> > -++ Tesla referral code - https://ts.la/richard11209
> Does Java qualify as a stack machine application?
>
> I remember that Sh Boom CPU was capable of running Java at near native speed.
> Nobody cared or...
> the entire goto market strategy was flawed and/or underfunded. (my guess)
>
> Having lived on both the engineering dept. side and the "make $XX million with this
> product or you are fired!" side, I find I have to remind the tech side that a better
> product only gives you permission to begin doing business.
> Business success comes from money and psychology
> ie:the "art" of marketing (one to many)
> and sales (one to one) communication.

You are right, but it includes process of steps and strategy
A good product only gives you something to sell. I remember years ago, when I was told about paradigm shifts requiring a 10x improvement to be accepted. This makes the product a lot more marketable. Some of the issues with shboom, was changes. Was HP and power PC taking away the target workstation company Apollo. Then, the whole Java thing itself. It didn't really take. Now. JavaScript has taken somewhat. Java is still ok, but Android and JavaScript had taken the light. I think JavaScript is, more or less, ok for many consumer side things. The biggest is the phone tablet market. On the lower end, is JavaScript phones based upon off a fork of the Firefox OS. Kaios. However, for a lot of third world gadgets, JavaScript is ok. This leads to opportunity for an common architecture for an android like phone chipset with or without phone features (2+ chips) where the chips are suitable for phones, gadgets, and internet of things. Of course, you don't have to use JavaScript, or C, you can use misc programming to get more out of it. Maybe a third more basic chip or two might also be helpful. But, these are common architecture with some features left out or included, and fabbed on the same die. A good place to start with one or two basic chips, and work towards better chips off of that. It can fit in there. For us, we can use it as misc, and be hired as the assembler like performance programmers. Live translation of JavaScript to native code is probably going be good, but misc better. Anyway, so we are talking about full screened $10 phone market with billions of potential customers. $30 phones, even better, and superior to other options, and $100 phones, which might be 100 million potential customers. Just using one chip. If you can reach 10% of those sales, then that pays for a future chips generation. Where you can aim at cheap good mid range phones and game phones for normal games. The problem is that they have run out their patent advantage, without taking advantage of the now costly hot phone chipset market, so anybody can do it. However, they are the experts to go to and hire for this, so they still have potential in the area. Low end phones could be a lot better, and even mid range phones are too high price, that is a market vulnerability. I'm not talking about giant leaps here, I'm talking about reading the land and taking small steps working up to giant leaps. If only they had done this 20 years ago, we would have had the leap, but there wasn't, so now there is small steps again, and maybe a multi GHz version of an enhanced array with some 32 but chips, when funded. The current prices or high end phones in a being driven in part, by the person be of the top chipsets. This elevator, has room under it, and as much as they might try to reverse it's direction, it's architecture can't get to the lowest floors well. I think.

On Sunday, December 5, 2021 at 1:30:00 AM UTC+10, gnuarm.del...@gmail.com wrote:
> On Saturday, December 4, 2021 at 4:42:46 AM UTC-5, Marcel Hendrix wrote:
> > On Saturday, December 4, 2021 at 12:53:02 AM UTC+1, Paul Rubin wrote:
> > [..]
> > > Yeah the main attraction of stack machines (if there is one at all) is
> > > very tiny implementation for a useful amount of processing. That can be
> > > helpful for deeply embedded micropower things, or on resource limited
> > > FPGA's. For example there's a new FPGA with 1000 LUT4's that will
> > > supposedly cost around 50 cents per unit. I don't think a RISC-V soft
> > > cpu will fit in there, but a single GA or b16-like cpu node might fit..
> > > That would let you compute some stuff and also have enough left over for
> > > some logic functions.
> > By the time you have finished that, that FPGA has a successor with
> > 16,000 LUT4's and as a company you have no follow-up to fend off
> > the RISC-V guys. There might be a possibility when a time comes
> > that it is impossible to increase instructions/Watt or the number
> > of LUTs/mm^2.
> I think you missed something. He is not talking about arbitrary devices. He specifically mentioned a $0.50 FPGA. I don't know who's chip this is, but I vaguely recall someone announcing that. You won't find a 16,000 LUT FPGA at that price for more than a decade, even if then. Even when you put a RISC-V into a low priced FPGA you will find it runs much, much slower than a small stack machine and uses a lot more power. I believe there have been some impressive implementations of RISC-V in FPGAs that use very little space. However, they run very slowly.

Ok, basically simple. Just overclocking potential to get to a higher speed, despite the power. Means, you have an interim alternative step to try a stack design. It's business.

>
> --
>
> Rick C.
>
> ++- Get 1,000 miles of free Supercharging
> ++- Tesla referral code - https://ts.la/richard11209

On Saturday, December 4, 2021 at 5:25:03 AM UTC-5, Paul Rubin wrote:
> Marcel Hendrix <m...@iae.nl> writes:
> > By the time you have finished that, that FPGA has a successor with
> > 16,000 LUT4's
> It's not a hugely complicated thing to put a small cpu and some logic
> into an fpga. There will not be a 16000 LUT, 50 cent FPGA anytime soon.
> There has not been such a thing as a 50 cent FPGA at all in the past, as
> far as I know. When there is one with 16000 luts for 50 cents, the one
> with 1000 luts will be 5 cents. This is still important for some kinds
> of products. Yes you can buy 50 cent microprocessors with 10s of
> kilobytes of ram, but if your application only needs a few dozen bytes,
> you can use a 3 cent processor.
>
> https://jaycarlson.net/2019/09/06/whats-up-with-these-3-cent-microcontrollers/

No, there will never be 5 cent FPGAs. The low end price of FPGAs is limited by testing costs. FPGAs have very large numbers of interconnections (contrary to what Hugh thinks) and many bitstream loads are required to test them all. That's why you can save money on quantity buys by specifying your design and allowing the chips you buy to be tested only on the fabric in use. This detracts from the flexibility of after sale updates, but that's often not used anyway and this allows the FPGA cost to approach the cost of an ASIC.

I didn't see anything in the Padauk article about how they achieve the price, but it is entirely possible they do it by not testing the parts, or maybe by testing them in parallel while on the wafer.

--

Rick C.

+++ Get 1,000 miles of free Supercharging
+++ Tesla referral code - https://ts.la/richard11209

On Saturday, December 4, 2021 at 10:40:58 AM UTC-5, Wayne morellini wrote:
> On Sunday, December 5, 2021 at 1:30:00 AM UTC+10, gnuarm.del...@gmail.com wrote:
> > On Saturday, December 4, 2021 at 4:42:46 AM UTC-5, Marcel Hendrix wrote:
> > > On Saturday, December 4, 2021 at 12:53:02 AM UTC+1, Paul Rubin wrote:
> > > [..]
> > > > Yeah the main attraction of stack machines (if there is one at all) is
> > > > very tiny implementation for a useful amount of processing. That can be
> > > > helpful for deeply embedded micropower things, or on resource limited
> > > > FPGA's. For example there's a new FPGA with 1000 LUT4's that will
> > > > supposedly cost around 50 cents per unit. I don't think a RISC-V soft
> > > > cpu will fit in there, but a single GA or b16-like cpu node might fit.
> > > > That would let you compute some stuff and also have enough left over for
> > > > some logic functions.
> > > By the time you have finished that, that FPGA has a successor with
> > > 16,000 LUT4's and as a company you have no follow-up to fend off
> > > the RISC-V guys. There might be a possibility when a time comes
> > > that it is impossible to increase instructions/Watt or the number
> > > of LUTs/mm^2.
> > I think you missed something. He is not talking about arbitrary devices.. He specifically mentioned a $0.50 FPGA. I don't know who's chip this is, but I vaguely recall someone announcing that. You won't find a 16,000 LUT FPGA at that price for more than a decade, even if then. Even when you put a RISC-V into a low priced FPGA you will find it runs much, much slower than a small stack machine and uses a lot more power. I believe there have been some impressive implementations of RISC-V in FPGAs that use very little space. However, they run very slowly.
> Ok, basically simple. Just overclocking potential to get to a higher speed, despite the power. Means, you have an interim alternative step to try a stack design. It's business.

I'm sorry, but I only understand about 10% of what you write. I often can't identify the verbs in your sentences. I don't get any of this.

--

Rick C.

---- Get 1,000 miles of free Supercharging
---- Tesla referral code - https://ts.la/richard11209

On Sunday, December 5, 2021 at 1:24:59 AM UTC+10, gnuarm.del...@gmail.com wrote:
> On Friday, December 3, 2021 at 11:26:41 PM UTC-5, Wayne morellini wrote:
> > On Saturday, December 4, 2021 at 11:31:55 AM UTC+10, gnuarm.del...@gmail.com wrote:
> > > On Friday, December 3, 2021 at 6:35:47 PM UTC-5, Wayne morellini wrote:
> > > > On Saturday, December 4, 2021 at 7:30:47 AM UTC+10, gnuarm.del...@gmail.com wrote:
> > > > > On Friday, December 3, 2021 at 1:40:16 PM UTC-5, Paul Rubin wrote:
> > > > > But, I see no
> > > > > > > reason that it doesn't suit a stack machine.
> > > > > > What I mean is that this is science fiction technology at the moment.
> > > > > > If it becomes viable for microprocessors, then it would presumably be
> > > > > > fine for stack machines, but also for non-stack machines.
> > > > > I agree, but I think the real issue is why chase pie in the sky implementations at 10s of THz when just a few GHz would be a significant improvement? It is normal for technology to proceed in steps rather than great leaps forward.
> > > > >
> > > > > In general the world has rejected stack machines for many, many years. What stack machines really need is a stack machine application, not a stack machine implementation.
> > > > >
> > > > > --
> > > > I'm not saying that they should do a Thz processor. I'm saying that QDCA is a descent bet. That they should look at the 500mhz+ advances and do a few GHz version. When the Thz, if it ever does, gets worked out,, they can move onto that. No science fiction involved, it's what scientists are actually working towards. It's business, you plan for the future and take a bet on which direction to start taking steps into. It's often not clear when using external innovations.
> > > You talk about planning as if it were inevitable these things will get designed and built. Where does the money come from? With no track record to speak of the hard part is finding someone who wants to start spending millions and millions of dollars on totally unproven design ideas.
> > Wherever it:s conventional silicon or not, it requires money. It's about where they will be in time, even 5. It's about commerce survival, and having the edge. They have to have something in offer for people to buy.
> Sorry, I don't understand what you are saying, literally.

It's pretty simple. The spell auto corrector trashed the sentence without me realising. Maybe where they will be in 5 years. About commercial survival. That's all.

> > > > Stack machines aren't the problem. Shboom, and rtx went on to have success. It's implementation we are concerned about.
> > > What success? They may have found a few design wins. I think the RTX gets used in space apps because it is rad hard (very hard to come by in general). There's nothing about this pedigree that would attract the sort of investor who will pay for such grandiose chips.
> > Those where significant designs of their day. Again, there is only normal level design being talked about at this stage. If they earn money, then they can advance to better designs. At the moment, its not optimal, so something has to change to continue.
> Significant to whom? They were tiny blips on the RADAR screen. I'm still waiting for someone to show any real world advantages to stack processor chips in the real world of today.

Well, objectively they were out there and successful
That's it.

> The GA144 attempted to be the universal peripheral laden MCU, but failed in being an MCU at all. The proponents talked about how low power the individual processors were and the low power when not processing, but the programming was so complex they came up with a virtual machine implementation that negated the power savings. In fact, everything they provided was a dollar short of being useful to a user.

Now, that is a historical blip. What you say is true.

> Partly, the problem of the GA144 was overcoming the entrenchment of conventional processors, but that is the world at this point. Talking about some power savings or speed advantage or even the flexibility of peripherals is not of much use if it solves a problem the users don't have while creating problems users don't have with conventional solutions.

They have been operating on presenting solutions to businesses. So, for them. In that way, it has been an employment opportunity. The businesses hi neatly have seen a potential there, and we wouldn't even know if it was in the bionic eat, or their hearing aids here.

> So how would money be made from such designs?

What they have been doing, yes, but it is scary design for people, and the regular arm is more comfortable. In the 1980's, it would have been a great design, even in the 1990's, but it really needed the 18 bit 640kB address space, even back then. At least one processor with access, if not most or all of them. Now. I wonder why for nearly two decades. I remember, you talked about doing software radio with it, but it was just st too our there and restrictive for a modern high datarate format. My recent designs proposals are suitable for that, but this needs to work at lower data rates. Where a custom asic can dominate it. I am concentrating on how changing tac might produce a better marketable product.

On Sunday, December 5, 2021 at 1:45:57 AM UTC+10, gnuarm.del...@gmail.com wrote:
> On Saturday, December 4, 2021 at 10:40:58 AM UTC-5, Wayne morellini wrote:
> > On Sunday, December 5, 2021 at 1:30:00 AM UTC+10, gnuarm.del...@gmail.com wrote:
> > > On Saturday, December 4, 2021 at 4:42:46 AM UTC-5, Marcel Hendrix wrote:
> > > > On Saturday, December 4, 2021 at 12:53:02 AM UTC+1, Paul Rubin wrote:
> > > > [..]
> > > > > Yeah the main attraction of stack machines (if there is one at all) is
> > > > > very tiny implementation for a useful amount of processing. That can be
> > > > > helpful for deeply embedded micropower things, or on resource limited
> > > > > FPGA's. For example there's a new FPGA with 1000 LUT4's that will
> > > > > supposedly cost around 50 cents per unit. I don't think a RISC-V soft
> > > > > cpu will fit in there, but a single GA or b16-like cpu node might fit.
> > > > > That would let you compute some stuff and also have enough left over for
> > > > > some logic functions.
> > > > By the time you have finished that, that FPGA has a successor with
> > > > 16,000 LUT4's and as a company you have no follow-up to fend off
> > > > the RISC-V guys. There might be a possibility when a time comes
> > > > that it is impossible to increase instructions/Watt or the number
> > > > of LUTs/mm^2.
> > > I think you missed something. He is not talking about arbitrary devices. He specifically mentioned a $0.50 FPGA. I don't know who's chip this is, but I vaguely recall someone announcing that. You won't find a 16,000 LUT FPGA at that price for more than a decade, even if then. Even when you put a RISC-V into a low priced FPGA you will find it runs much, much slower than a small stack machine and uses a lot more power. I believe there have been some impressive implementations of RISC-V in FPGAs that use very little space. However, they run very slowly.
> > Ok, basically simple. Just overclocking potential to get to a higher speed, despite the power. Means, you have an interim alternative step to try a stack design. It's business.
> I'm sorry, but I only understand about 10% of what you write. I often can't identify the verbs in your sentences. I don't get any of this.

Don't worry about it.

On Sunday, December 5, 2021 at 1:30:00 AM UTC+10, gnuarm.del...@gmail.com wrote:
> On Saturday, December 4, 2021 at 4:42:46 AM UTC-5, Marcel Hendrix wrote:
> > On Saturday, December 4, 2021 at 12:53:02 AM UTC+1, Paul Rubin wrote:
> > [..]
> > > Yeah the main attraction of stack machines (if there is one at all) is
> > > very tiny implementation for a useful amount of processing. That can be
> > > helpful for deeply embedded micropower things, or on resource limited
> > > FPGA's. For example there's a new FPGA with 1000 LUT4's that will
> > > supposedly cost around 50 cents per unit. I don't think a RISC-V soft
> > > cpu will fit in there, but a single GA or b16-like cpu node might fit..
> > > That would let you compute some stuff and also have enough left over for
> > > some logic functions.
> > By the time you have finished that, that FPGA has a successor with
> > 16,000 LUT4's and as a company you have no follow-up to fend off
> > the RISC-V guys. There might be a possibility when a time comes
> > that it is impossible to increase instructions/Watt or the number
> > of LUTs/mm^2.
> I think you missed something. He is not talking about arbitrary devices. He specifically mentioned a $0.50 FPGA. I don't know who's chip this is, but I vaguely recall someone announcing that. You won't find a 16,000 LUT FPGA at that price for more than a decade, even if then. Even when you put a RISC-V into a low priced FPGA you will find it runs much, much slower than a small stack machine and uses a lot more power. I believe there have been some impressive implementations of RISC-V in FPGAs that use very little space. However, they run very slowly.

I thought I replied to this. It's about overclocking to get a similar performance, giving you an intermediate product solution to make money off of, and hopefully pay towards custom designs latter.

On Saturday, December 4, 2021 at 11:00:13 AM UTC-5, Wayne morellini wrote:
> On Sunday, December 5, 2021 at 1:24:59 AM UTC+10, gnuarm.del...@gmail.com wrote:
> > On Friday, December 3, 2021 at 11:26:41 PM UTC-5, Wayne morellini wrote:
> > > On Saturday, December 4, 2021 at 11:31:55 AM UTC+10, gnuarm.del...@gmail.com wrote:
> > > > On Friday, December 3, 2021 at 6:35:47 PM UTC-5, Wayne morellini wrote:
> > > > > On Saturday, December 4, 2021 at 7:30:47 AM UTC+10, gnuarm.del...@gmail.com wrote:
> > > > > > On Friday, December 3, 2021 at 1:40:16 PM UTC-5, Paul Rubin wrote:
> > > > > > But, I see no
> > > > > > > > reason that it doesn't suit a stack machine.
> > > > > > > What I mean is that this is science fiction technology at the moment.
> > > > > > > If it becomes viable for microprocessors, then it would presumably be
> > > > > > > fine for stack machines, but also for non-stack machines.
> > > > > > I agree, but I think the real issue is why chase pie in the sky implementations at 10s of THz when just a few GHz would be a significant improvement? It is normal for technology to proceed in steps rather than great leaps forward.
> > > > > >
> > > > > > In general the world has rejected stack machines for many, many years. What stack machines really need is a stack machine application, not a stack machine implementation.
> > > > > >
> > > > > > --
> > > > > I'm not saying that they should do a Thz processor. I'm saying that QDCA is a descent bet. That they should look at the 500mhz+ advances and do a few GHz version. When the Thz, if it ever does, gets worked out,, they can move onto that. No science fiction involved, it's what scientists are actually working towards. It's business, you plan for the future and take a bet on which direction to start taking steps into. It's often not clear when using external innovations.
> > > > You talk about planning as if it were inevitable these things will get designed and built. Where does the money come from? With no track record to speak of the hard part is finding someone who wants to start spending millions and millions of dollars on totally unproven design ideas.
> > > Wherever it:s conventional silicon or not, it requires money. It's about where they will be in time, even 5. It's about commerce survival, and having the edge. They have to have something in offer for people to buy.
> > Sorry, I don't understand what you are saying, literally.
> It's pretty simple. The spell auto corrector trashed the sentence without me realising. Maybe where they will be in 5 years. About commercial survival. That's all.

Still not following. "Maybe where they will be in 5 years" is not a sentence, no verb. Who's commercial survival???

> > > > > Stack machines aren't the problem. Shboom, and rtx went on to have success. It's implementation we are concerned about.
> > > > What success? They may have found a few design wins. I think the RTX gets used in space apps because it is rad hard (very hard to come by in general). There's nothing about this pedigree that would attract the sort of investor who will pay for such grandiose chips.
> > > Those where significant designs of their day. Again, there is only normal level design being talked about at this stage. If they earn money, then they can advance to better designs. At the moment, its not optimal, so something has to change to continue.
> > Significant to whom? They were tiny blips on the RADAR screen. I'm still waiting for someone to show any real world advantages to stack processor chips in the real world of today.
> Well, objectively they were out there and successful
> That's it.

Were they? I don't know what definition of "successful" you are using. I suppose you could call the RTX successful in that they sold more than a handful, but what happened with the Shboom that would be called "success"???

> > The GA144 attempted to be the universal peripheral laden MCU, but failed in being an MCU at all. The proponents talked about how low power the individual processors were and the low power when not processing, but the programming was so complex they came up with a virtual machine implementation that negated the power savings. In fact, everything they provided was a dollar short of being useful to a user.
> Now, that is a historical blip. What you say is true.
> > Partly, the problem of the GA144 was overcoming the entrenchment of conventional processors, but that is the world at this point. Talking about some power savings or speed advantage or even the flexibility of peripherals is not of much use if it solves a problem the users don't have while creating problems users don't have with conventional solutions.
> They have been operating on presenting solutions to businesses. So, for them. In that way, it has been an employment opportunity. The businesses hi neatly have seen a potential there, and we wouldn't even know if it was in the bionic eat, or their hearing aids here.

I wasn't aware that anyone at GA was actually an employee in the sense of drawing a significant salary. If the company were selling any real quantity of parts, they would report the sales even if not the customer. I think they bought some thousands of chips and are still working on selling those.

> > So how would money be made from such designs?
> What they have been doing, yes, but it is scary design for people, and the regular arm is more comfortable. In the 1980's, it would have been a great design, even in the 1990's, but it really needed the 18 bit 640kB address space, even back then. At least one processor with access, if not most or all of them. Now. I wonder why for nearly two decades. I remember, you talked about doing software radio with it, but it was just st too our there and restrictive for a modern high datarate format. My recent designs proposals are suitable for that, but this needs to work at lower data rates. Where a custom asic can dominate it. I am concentrating on how changing tac might produce a better marketable product.

In the 1980s feature sizes crossed 1 um. The GA144 would have been a much larger chip (around a square inch) and run much more slowly.

The GA144 could implement a software radio easily. It samples at software determined rates up to MHz. You might be able to tune the FM band, but the AM band for sure. I don't recall the frequencies used for hand held unlicenced radios in the US, but they are probably in the UHF, so not as practical.

To create a product you typically start with the requirements and look for technology to implement it. The GA144 was a technology experiment to see what the chip could do with no application in mind. Maybe the device being designed now by another company will have a purpose.

--

Rick C.

---+ Get 1,000 miles of free Supercharging
---+ Tesla referral code - https://ts.la/richard11209

Rick C <gnuarm.deletethisbit@gmail.com> writes:
>No, there will never be 5 cent FPGAs. The low end price of FPGAs is limite=
>d by testing costs.

That might be an opportunity for a minimal-area processor core (e.g.,
something like the b16): Instead of testing the FPGA on an expensive
testing machine, put a low-area core on the FPGA, and it does the
testing. Of course you still need a testing machine that puts power
to the die, gives the testing command and reads the result of testing,
but that could be much cheaper than the more powerful testing machines
used now.

- 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: http://www.forth200x.org/forth200x.html
EuroForth 2021: https://euro.theforth.net/2021

Brian Fox <brian.fox@brianfox.ca> writes:
> Does Java qualify as a stack machine application?

No it really doesn't. The JVM is a stack virtual machine but high
performance applications translate the JVM code into register machine
code. Also I believe (not sure) that the JVM stack can be indexed like
an array, so it is not a true stack with only LIFO access (other than
the top few elements).

Rick C <gnuarm.deletethisbit@gmail.com> writes:
> The GA144 could implement a software radio easily. It samples at
> software determined rates up to MHz.... I don't recall the frequencies
> used for hand held unlicenced radios in the US, but they are probably
> in the UHF, so not as practical.

Software radios often have conventional mixer stages, so the SDR part is
digital demodulation of the mixer ouput. The sample rate doesn't have
to be anywhere near as high as the carrier.

> The GA144 was a technology experiment to see what the chip could do
> with no application in mind.

Or certainly without a clear enough picture of how to implement the
envisioned application. The underpants gnome school of design.