P2 Feature Creep Thoughts
Kye
Posts: 2,200
I see feature creep has derailed the Prop 2, and looking at the new P2 thread by Chip Gracey, I see feature creep happening again.
I can't enumerate all the features a new chip will need. But, I think at least, for the compute part of the chip, the P1 architecture would be sufficient with hub execute and more memory added.
Most people will want to just run large programs from memory at native speed while having other cogs provide smart peripheral support... which is how people use SPIN/C on the P1 now.
Anyway, flame away!
I can't enumerate all the features a new chip will need. But, I think at least, for the compute part of the chip, the P1 architecture would be sufficient with hub execute and more memory added.
Most people will want to just run large programs from memory at native speed while having other cogs provide smart peripheral support... which is how people use SPIN/C on the P1 now.
Anyway, flame away!
Comments
From this point, I believe Chip knows -- from past comments and his own good judgement -- what should go into the P1+. My personal suggestion would be if Chip still needs to bounce off some ideas throughout the remainder of the development process, how about creating a private forum where he can get input from a very small hand-picked selection of users (the P1+ focus group). This would still provide feedback, but without the distraction of so many chefs stirring the pot.
I am with you there. Ken said that this is not the first time Chips projects went overcomplicated but then reduced them self to simplicity again. This is part of his art.
I am hoping the best...
Enjoy!
Mike
I can:
* 16 or more cores
* 512k or more RAM
* 64 or more I/O pins
Any other features "as permitted", and nothing is permitted unless its free, or more costly to omit.
Get it done and out the door so we can buy them, and work on the design for any next version until this version earns enough to pay for such.
That was easy. If I were clever instead of smart I'd be rich.
I keep seeing the unequal hub access showing up in every thread in as well, or configurable, or whatever. Just KISS.
Or extra instructions to save two instructions. How about a rule that you can only suggest a new instruction only if it can act like THREE OR MORE instructions, then it should be added in, maybe. If it's useful enough.
Or the thought keeps popping up that somehow 32 cogs are going to fit in this space. Yes, probably 32 P1 cogs with zero improvements. but it's all the improvements that keep getting added in... I can just see the "guys are you okay with 8 cogs?" question popping up. I am cringing right now, about the thought of seeing that question in 2 months.
And then I'm personally getting frustrated that an externally clocked Serial shift-in register can't be put in. all this talk about this and that, but no simple, universal hardware like this. Just keep the chip in RCSLOW, let something megahertz fast clock something in, the Propeller chip wakes up and goes fast, does its thing, then powers back down.
I got really mad about the 50 mips not good enough, we need 100 mips. Why? What can 100 do that 50 cannot? why go back to the pipeline craziness that was the P2 just for "double"?
I was going to be alright with a downclocked 8 core P2, would still be happy with 4 core P2, but now it's complete redesign with no end in sight. no timeline again.
/rant... just trying to let off some steam...
Potentially crippling to Parallax.
Assume they ship out the new Prop in 4 months.
Assume $5 margin per chip.
How many units need to be sold to recover past 5-6 years 1-2M investment $?
How many units need to be sold.to cover another 2-3 years 1M R&D $?
Do the math.
Now what if the slam it out into production product also doesn't get picked up by industry because it is seen as too slow?
Parallax will never recover past costs with sales to forum dwellers or academic sales.
Gonna then be hard to fund any future development.
Parallax has a lot more riding on this product than they ever did with the P1.
What I think you are missing or misquoting in this equation is the fact that the money and time spent has already been spent on P2 and P3 but for the sake of having silicon for 2014 in the form of P2 (P16X32B) it is only a matter of a few months extra and the cost of another run from which immediate sales could be generated if it were a small volume run.
As much as P3 is so smart that it will program itself and solve world hunger I still need and find immediate commercial uses for a P2 with 16 cores (vs threads) and 512kB RAM. I will already have software and hardware for this chip as soon as it is available.
What worries me is the P1 replacement still seems mired in debate about what it could be ( and excessively deep discussion about how that could be achieved ) but it's missing the most important thing; what that replacement needs to be to meet market expectations and make it a success for Parallax. It's like 'we know we need something but we don't know exactly what it is we need'. That feels rather worrying at this stage of the game, as if Parallax don't have a grip on where they are heading or need to be heading, after all these years don't have a baseline they are aiming for.
Why is HUBEXEC and multi-tasking and all the other things still being debated ? Are Parallax not sure whether they are needed or not ? If they don't know then why not ? Shouldn't those fundamental issues have been decided by now ?
It is perhaps not my place to tell Parallax how things should be or how they should do things but I don't think it's unreasonable to note how it may look to others who are mostly observers. In many ways it looks like 'the lunatics are running the asylum' and the whole process is out of control but that may just be a result of 'open collaboration done in public'. Whatever; it doesn't look good from the outside, doesn't inspire confidence.
The sooner Parallax can decide and categorically state what it is they are going to do and head for that the better I think things will be. It seemed things were going that way a week ago but what clarity there was then was quickly lost in the ensuing debate and discussions. The earlier P2 stalled, the 'even better' P2 hit a wall, so out goes the new, and back to the P1, and now all or some of the P2 is coming back in, with new and more as well. But this time it will succeed, if it's ever decided what it is. It seems to be back to 'blue sky' dreaming of what it could be rather on the home run to delivering a product to meet needs which are well defined.
I hope this doesn't offended Parallax, Chip or Ken, or any of the contributors here as harsh as it may sound on first reading - It doesn't help anyone if the P1 replacement does not meet the needs it needs to meet or feature creep derails its progress.
Amen to that, brother hippy.
Chip should just take Easter off and decide whether what he wants to build is a successor to the P1, or something else entirely.
Becasue we sure as **** ain't going to get either one, the way we are currently going.
Ross.
Merde!
I see what you mean, however I'm not sure I agree.
The money spent was originally for P2, then the P2+Kitchen_Sink, shuttle run, and then switch to OnSemi and whatever those costs are.
Not sure I've read of anything being spent on P3, unless you are saying all of this recent past R&D can be said to have been potential P3 use.
OK, won't argue that. However the main thing is, all this capitol investment has been sunk, and more will be required in the coming months, and it is $50K/shot shuttle runs.
I just fail to see any real, reasonable profit that Parallax is goiung to make.
Again, assume $5/chip pure profit. Parallax has to sell 10,000 to make up just the most recent failed shuttle run.
I can see the non-commercial Forum folks and other enthusiasts buying maybe 1000, 2000.
Maybe a few commercial folks like yourself take another 2000,4000, heck make it the remaining 8000.
Without knowing Parallax's sales numbers, neither of us can really say what happens then.
If it garners decent interest, then thats n sales. The question then is, what is the repeat sales, and ultimately is Parallax doing production runs every 1,2 months, or every 6,8,10?
If the new Prop is only received as well as the last one, then Parallax is 'potentially' in worse shape than with P1, as they have all that sunk $$, and a ROI that is measured in multiple years/decades.
That then leads one to ask, how does Parallax dig out of this hole, and get the funds to even think about a P3.
If it does in fact meet this sort of reception, then at some point, don't people think that Ken's smart enough to realize that the Prop just isn't going to be financially successful enough to continue this line of R&D?
I hope I am wrong.
However, every time I see a post on the forum from people just pushing for the P2+Kitchen_Sink to be pursued, or to rush to market with the current design, I can't help but think that they probably don't realize they just may be cutting off their nose to spite their face.
Hopefully Parallax sells enough gear on the website to continue to support the Prop.
. I hope I
Marty
Whatever minimum is necessary and sufficent to meet the minimum requirements.
If the minimum requirements are more cogs, more ram, more pins; then what is in place to support these? There's already circuitry requied to implement the pins, if it include ADC on every pin, great; if it doesn't don't invent it.
Mutlitasking? The prop is already multitasking.
Execute from hub? It already executes from hub (with one operation of "copy from hub to cog")
Its already complicated enough, fer cryin' out load, it a micro controller! Now we need the simplest thing that will get 'er done!
(This is my standard anti-feature creep presentation, with microcontroller in place of "development-project-name-here")
The fewer the new inventions (unknown variables) the greater the change of completing the project, at any given time.
Of course, I own fewer companies than Chip, so I defer to his desisions.
But the HubExec found an elegant solution in the shadow registers up at the top of cog memory.
And the bus widths were resized to both grab quads or four instructions at a time.
So this one isn't actually feature creep. It just flowed together.
An idea might trade one thing for another,
but a solution solves two or more problems:
To read quads, set the cog ram to 128x128 bits...
1) Becomes more efficient by reducing receive mux complexity compared to before (reduced logic, reduced power)
2) If you can read quads, you can get 4 instructions at a time to be executed. Faster than LMM could be.
3) Do something with otherwise wasted cog shadow registers at the top. Why not make this the hubexec cache?
4) With hubexec, meet current expectations to run large C programs at a fast-enough rate to be worthwhile.
5) Instant cog-start possible (instead of reload all memory then execute)
But ugh, glad things have calmed down a little.
Yesterday was fearing we were going to get the CLUTs back in this one, with all the management instructions, rotating pointers, etc.
Just because a 4 level deep call stack "isn't big enough".
Sandy
1+
As long as we hold on to the one thing that always kept me coming back to the prop.... it's simplicity. Multi-processing is easy, counters are easy, GPIO is easy, even the assembly is fairly easy. It's the relatively tiny RAM space that really was the main limitation I ran into after C/C++ was ported to it 512K would be a massive improvement. Most of my C++ programs run short of memory space before I run out of cogs. I hope things don't deteriorate to the point that we are still having this debate in 2015 though.
This is not a diss on photomankc. He is simply the latest to use an oft-repeated and mildly petulant device to express dismay and perhaps to trigger greater activity or resolve on the part of Parallax.
But this idea of waiting and waiting, and finally moving on, in addition to being a bit melodramatic, is also kind of useless. Why is anyone waiting for anything? Do something today!
Like Professor Harold Hill said, "You pile up enough tomorrows, and you'll find you are left with nothing but a lot of empty yesterdays. I don't know about you, but I'd like to make today worth remembering."
There are precious few obstacles that can't be conquered by devices available today. Get busy on something now and before you know it the new Parallax chip will be here.
Excellent advice User Name!
When things got crazy and we turned a corner a couple weeks ago I decide to spend less time on the forum and more time working on things I can control.
I started walking 2 miles a day instead of spending time restating the obvious and either preaching to or arguing with the choir.
Already lost 5 pounds and feeling good enough to bump up to 4 miles a day.
So yeah, we get what we get from Chip when we get it, but do something yourself today.
C.W.
That sort of smacks of "with all due respect". Perhaps the phrase "decided to move on" has a bit of an emotional connotation. End of the day I have no control over it, no say in the business, and no more valid or invalid opinion than any other outsider. Ken/Chip and crew know far better then I do what risks they are in a position to take. The chip gets produced when Parallax produces it, that's completely correct. The thread is about feature creep and my opinion here is that this runs the risk of eroding interest in the product and allows applications that could generate income to pass by as more and more features delay the product.
Petulance or melodrama.... well... none are intended for my part.
+1 to everything you wrote. C.W. too. Sandy makes an excellent point as well - that Chip has surely received all the input he needs(!).
Speaking of input, the sudden interest by Adafruit to consider a P2-based Arduino is stunning (and well-deserved). That is surely the most galvanizing input Chip and Ken have received in a while. What a great opportunity to finish the chip post-haste and get it out there!
I don't think there is any "interest by Adafruit to consider a P2-based Arduino".
All that happened is that the same guy posted similar ideas on the Arduino forum and here. Today we see he has posted same to the Adafruit forum, despite receiving suggestions from this side that it's way too premature to discuss such a board design.
This basically amount to trolling, even it is not intended that way.
And thank you to Roy for showing up on the Adafruit forum and providing the facts around Parallax and the Propeller 2.
Ken Gracey
However, I DO WISH that Parallax would bring the knife down on this chocolate bar and put it in a wrapper for sale.
I'd be very content to purchase a P2, then later purchase P3, P4 with all the "creepy features" that Chip incorporates into following designs..
..... ... Ship it!
Jeff
Jeff, you're not alone and we welcome your important input. In many ways, Propeller Powered and Parallax have much in common, especially the challenges of a eight-year development process and the effects it has on our respective businesses.
You'll notice that Chip's comments are very much oriented around completion. Aside from some errant comments about opium and government, he's focused on completion right now. I've seen this process before and know where he is in the pipeline. He's doing a great job avoiding the nonsense and moving forward rapidly. I'm trying to run Parallax so he can achieve this goal without additional noise from the business, too.
We all know this business is more productive with frequent releases, too. We failed to make frequent releases but the delay and costs were for an education.
Hang in there, stick with us as you are. We will see lots of embedded stuff come out in the meantime, but I've been here before and I know when we're on the cusp of dong exactly what our customers have asked for and it'll seem nearly magical. In the last 25 years these big strides occurred with similar steps: ISEPC, PIC Programmer, BS2, SX28/48 and P8X32A.
For companies like your own who are part of the early adoption phase there is much potential. This time we won't be making a bunch of boards, too. We're supporting and leaving it to the ecosystem to grow with us.
Ken Gracey