VC is for startup programs. Parallax is way beyond that. Crowd funding won't raise enough money.
Chip and Ken
My understanding is that we can have everything that our DE2-115's P2 emulation can handle, plus a little more without worrying about anything.
I like not worrying. Don't go to Mexico. Don't take a day off. Just go to the 4 cog design and immediately design multiple-P2 boards.
With a 4cP2 in production, go public. Pocket a billion or two and move directly to in-house fab of 8,16 or 32 cogs at 40nm or 28nm.
First, we've not released any real spec. What we have done is explore the possible. This is no different than anybody else who is making a new chip. You can bet this same discussion has happened all over the place, and the answer was to position the design for best case adoption, not worry that the peak expectation wasn't the ideal expectation. Remember that. How we position this matters. A. Lot.
This design running at 80Mhz will rock. It's just that simple. That we have 2X the clock in terms of headroom for those willing to manage the power is just a bonus! Marketing 101 indeed! Having this discussion is precisely what the marketing people would do, and they would have the one I mentioned above' namely, who will adopt this and how do we get the message to them?
Finally, 80Mhz has the nice aspect of being FPGA capable! If one wants to test a design, or do some other activities, an FPGA can run the real deal at that clock, given a reasonable FPGA. There is a nice synergy here. I don't know what that means, but it's worth putting out there.
Another aspect of 80Mhz is board design. 80Mhz is a nice, easy number. We know that from P1 and how a 10Mhz or so xtal and related design can be done on a breadboard with some reasonable expectation of success. If people really want that nice, round 100Mhz number, so what? We can do that too, and given the power curve, it changes very little.
Speaking of that power curve, we just put that in the datasheet. People can run it as hard as they want to, just like they do the current P1, and it's curve, that clearly says we can run those at 100Mhz too. And people do, all the time. 100Mhz is my favorite P1 clock.
Re: Cog timing. Yes! I completely agree with this suggestion. Can we add a little temp sensor of sorts? There has gotta be something we can read to see the state of things and adjust accordingly.
I'm against this, and I'm against it for a few reasons:
1. Public companies come with some rather brutal expectations, unless they are big enough to be valued enough for the key players to isolate their shares and remain in complete control of their company. Silicon Valley has seen the bell ring at the school of hard knocks on Wall Street a few times now. Look at what Zuck did with FB and the dual stock share strategy. Zuck holds Class B stock, with 10x the voting of the class A stock, and it's his company, despite public ownership. Awesome, and something the Wall Street guys with their expectations don't like much, but the perception of value in FB was so high, they had to have stock anyway.
Kids, that doesn't happen every day, and the vast majority of companies who go public won't have that Zuck experience, instead finding themselves on a quarterly treadmill that makes exactly none of the things we like to do here possible. Think about it.
2. Culture. Set aside the financial expectations. For those of us who have met Parallax, being a private company means being able to make some decisions that are the right ones, but not optimal financially. This is what makes Parallax what and who it is. I value that very highly. I think most of us do, and I know Ken, Chip, et al. do.
3. IP. A public company would have some additional IP concerns that again break what we are doing here.
That conversation has already started… don't remember exactly where. The answer was yes.
If temperature management is going to be a real issue, then it deserves the usual forum treatment:)
@rjo_ You can have whatever timestamps you want! Go into settings, then general settings, scroll down and find the time settings. I've got mine at GMT -8, which is Pacific Time. Sometimes, when I'm doing stuff with others, I'll set to their time zone, just so I'm grokking what it's like for them. You can do the same.
Many of the issues you raise are flags… not show stoppers. The key is the charter (which grants the rights) and the board structure(which manages the responsibilities).
It is their offer, if people don't like it, they don't invest.
If Chip and Ken don't like the eventual direction… they can be freed by the charter to do just about whatever they want. Anticipating the problems and handling them
properly in the charter is the key. There will be lawyers involved… you have to feed them.
Whatever the charter does to protect Ken and Chip's capacity to direct the company, investors will be there. Parallax has an international reputation. They could probably get enough to do whatever they want on the Hong Kong exchange. The Chinese love business ethics but rarely see them. They and are always happy to find an honest American to invest in:)
Yes they are, but the overall value proposition and it's strength and appeal impact those things. How it works is this:
If you really need the money, you are going to give up the control. If you don't, then you might have a shot at doing what Zuck did.
And that, of course, also depends on the broad appeal inherent in the proposal. We don't have that broad of appeal here, so then we would yield a lot of control. Like you say, they don't have to invest. And I'm saying they won't without the control being handed over, and so we need to figure this out private or give up who we are.
Seriously. I've been around that enough to know better.
Becoming a public class C corporation is an entirely different beastie. Parallax is a closely held company. As I recall from casual chats, Ken and Chip want it to stay that way.
As far as funding goes, one or more LLP loans would be best because the only thing lenders get are interest payments until the principal is repaid (I.E. not influence whatsoever in company decisions as defined by the LLP "Limited Liability Partnership" law). There is no stake in the company. Kickstarter is an interesting idea, but it is not time-tested like an LLP and I'm personally very worried about such things as cross-border contributions violating our investing (SEC) laws. I suspect most kickstarter contributors are not qualified investors. LLP loans can be made with anyone using a simple standardized legal form.
Haven't had time to read the whole thread, But I'd vote for option pare the features back. The feature I'm looking forward too the most is the analog IO pins. Tied for a distant second, the next most important features are the 8x higher instruction throughput and added math functions. After that, I'd rank SERDES, external SRAM, and multi-tasking as the third most useful features I'm looking forward to. (though switching to a smaller process would be fine too.)
Have you looked at "power pad" packages? The ADS1278 from TI has a 64-pin TQFP package with a power pad. That should let you dump around 2 watts into a ground plane in a easy to hand-solder and inspect package. (i.e. for the power pad use vias larger than your soldering iron, and fill them from the back of the board)
Marty
P.S. have you looked at using a smaller DRAM process? The few articles that have mentioned this say DRAM processes are a good bit slower, but like 1/5th the cost per transistor.
This problem is like trying to fit a large block V8 in a mini. A lot of you guys are essentially saying things like "lets change the spark plug wires, that'll make it fit" or "if we use a different fuel injection system, it'll work" or "let's paint the block blue instead of red, then it'll surely fit". The real solutions are more dramatic like switch to a 4 banger, reduce the engine to a small block with much smaller bore pistons, or pick a different car to put the large block V8 into.
This problem is like trying to fit a large block V8 in a mini. A lot of you guys are essentially saying things like "lets change the spark plug wires, that'll make it fit" or "if we use a different fuel injection system, it'll work" or "let's paint the block blue instead of red, then it'll surely fit". The real solutions are more dramatic like switch to a 4 banger, reduce the engine to a small block with much smaller bore pistons, or pick a different car to put the large block V8 into.
This problem is like trying to fit a large block V8 in a mini. A lot of you guys are essentially saying things like "lets change the spark plug wires, that'll make it fit" or "if we use a different fuel injection system, it'll work" or "let's paint the block blue instead of red, then it'll surely fit". The real solutions are more dramatic like switch to a 4 banger, reduce the engine to a small block with much smaller bore pistons, or pick a different car to put the large block V8 into.
Removing those items would not change the power envelope noticeably as they take too small a fraction of the total gates to matter.
Lowering clock frequency, removing cogs, lowering vCore - those make the difference.
Of the suggestions so far, the one I really like is setting the multiplier (of xin) for each cog individually; that would definitely help.
Correct, and Chip has indicated ~90% is already clock gated, and adding this small detail, is not too hard.
Power is set by nodes toggling, and the device only every does one opcode at a time..
Clock gating is already used to lower power, so all this diversion over what details to remove miss the point.
That discussion makes sense, when P2 fails to FIT, not when the Power envelope is higher.
I checked some Thermal specs on Power PAD TQFP128, and they are
0.12 °C/W jc .from die to case, and
17.17 °C/W. ja PCB mounted, thermal vias
( source tvp5158 TQFP 128-Pin PowerPADTM Package )
So that package is capable of some watts.
To normalize the figures, 5W Max total is 0.625W/160MHz/COG, and if we target say +50°C, that's 2.912W.
Using that, all COGS can run at 93.184MHz , which makes a ~100MHz target possible,
A 100MHz target makes more marketing sense, than 80MHz.
If 2 COGS tun at 160Mhz, that allows the power budget to look something like this for a sub 3W thermal profile
0.625*40/160+0.625*5*80/160+0.625*2 = 2.96875W
ie
1 COG at 40MHz (160Mhz time precision)
5 COGS at 80MHz (160Mhz time precision)
2 COGS at 160MHz
I think that type of COG combination covers a lot of use cases
There are also cheap SMD (solderable) heatsinks, that can add maybe another watt, perhaps 2, to this, or just buy cooler operation.
Next, would be the Aluminum top-hats, and really serious users, can always use this !!
This problem is like trying to fit a large block V8 in a mini. A lot of you guys are essentially saying things like "lets change the spark plug wires, that'll make it fit" or "if we use a different fuel injection system, it'll work" or "let's paint the block blue instead of red, then it'll surely fit". The real solutions are more dramatic like switch to a 4 banger, reduce the engine to a small block with much smaller bore pistons, or pick a different car to put the large block V8 into.
Nice analogy Roy. Very helpful to those of us following along in the shadows.
This problem is like trying to fit a large block V8 in a mini. A lot of you guys are essentially saying things like "lets change the spark plug wires, that'll make it fit" or "if we use a different fuel injection system, it'll work" or "let's paint the block blue instead of red, then it'll surely fit". The real solutions are more dramatic like switch to a 4 banger, reduce the engine to a small block with much smaller bore pistons, or pick a different car to put the large block V8 into.
Err no, wrong analogy.
The issue here is not physically fitting, it is Power/Thernal Profile, which is a rather different problem.
The term is used loosely here, it means multiple MHz specs, not outside spec.
Like all Logic devices, the P2 has a Thermal Profile, if you want more MHz, you put more effort into cooling.
Some companies spec Commercial and Industrial operation on the same part number now (used to be 2 part codes)
The Commercial spec has a higher MHz value, purely from this thermal Profile.
So it makes sense to have a Spec that covers
a) Minimal thermal effort --> Give the users a MHz spec for this point.
b) More designed thermal spread, more layers, more copper --> Give the users a MHz spec for this point.
c) Added heatsinks, eg simple, solder-able ones --> Give the users a MHz spec for this point.
It's better to skip the car analogies. They never hold up. Cars, and peoples expectations of them, are two varied.
I was just recalling the time I first ever heard of CMOS logic. Back in 1970 something. As opposed to the TTL us kids all knew and loved. The article was all about low power and even had a circuit running off a lemon.
Where did it all go wrong?
The idea of a micro-controller with a heat sink on it is more than I can bear.
OK, V8 going into a mini which only has so much cooling capacity with its radiator and no room to put in a larger radiator.
Change plug wires? paint it red? Add fans? Etc?
Does that work better?
Hehe yes, that's closer, and I've actually seen exactly that problem tackled.
They solved the "no room to put in a larger radiator." with a little lateral thinking, and fitted a second radiator underneath the car, below the passenger seat.
The ultimate issue is that the 'Propeller Philosophy' is not well suited to scaling.
What we see here is a concrete example of a 'cost' for having every COG be identical.
Allowing individual COGs to run at a reduced speed is a step that will be helpful and is a way forward from where we are today and may make the current design viable.
IMHO in the long term the philosophy needs some tweaking.
Allowing individual COGs to run at a reduced speed is a step that will be helpful and is a way forward from where we are today and may make the current design viable.
Agreed.
I think this also means that some means to measure COG temperature is going to be needed.
Chip was talking of mapping RC Oscs divided to buried pins - that may be enough for Temp sense, or a proper temp cell may be needed ?
The ultimate issue is that the 'Propeller Philosophy' is not well suited to scaling.
I don't believe this is true.
A COG, despite it's many transistors consumes almost no power if it is not doing any work. Or at least I hope that is the case.
As a UART or PWM driver or whatever it has a lot of "free" time. When it actually has to work that takes energy of course. It should not be sitting in "idle loops" wasting power.
Any way, regardless of that, Moore's law has come to an end and all the worlds chip makers are looking at parallel solutions to make performance increases.
The "Propeller Philosophy" was perhaps ahead of the game. If it does not scale they all have a problem.
The P2 with a heatsink isn't a marketable item it's the equivalent of Mr. Creosote in a restaurant - it would be the butt of jokes. Yeah I can just imagine some Prop guru doing a sales pitch where he pulls out a P2 with cooling fins on it. Eyes start bugging out, giggles are heard in the audience and the boss figures it's a prank and calls security on the guru.
Really can anyone make valid sales pitch for the P2 with a heatsink over a equivalent ARM or Pic32?
Comments
VC is for startup programs. Parallax is way beyond that. Crowd funding won't raise enough money.
Chip and Ken
My understanding is that we can have everything that our DE2-115's P2 emulation can handle, plus a little more without worrying about anything.
I like not worrying. Don't go to Mexico. Don't take a day off. Just go to the 4 cog design and immediately design multiple-P2 boards.
With a 4cP2 in production, go public. Pocket a billion or two and move directly to in-house fab of 8,16 or 32 cogs at 40nm or 28nm.
Rich
First, we've not released any real spec. What we have done is explore the possible. This is no different than anybody else who is making a new chip. You can bet this same discussion has happened all over the place, and the answer was to position the design for best case adoption, not worry that the peak expectation wasn't the ideal expectation. Remember that. How we position this matters. A. Lot.
This design running at 80Mhz will rock. It's just that simple. That we have 2X the clock in terms of headroom for those willing to manage the power is just a bonus! Marketing 101 indeed! Having this discussion is precisely what the marketing people would do, and they would have the one I mentioned above' namely, who will adopt this and how do we get the message to them?
Finally, 80Mhz has the nice aspect of being FPGA capable! If one wants to test a design, or do some other activities, an FPGA can run the real deal at that clock, given a reasonable FPGA. There is a nice synergy here. I don't know what that means, but it's worth putting out there.
Another aspect of 80Mhz is board design. 80Mhz is a nice, easy number. We know that from P1 and how a 10Mhz or so xtal and related design can be done on a breadboard with some reasonable expectation of success. If people really want that nice, round 100Mhz number, so what? We can do that too, and given the power curve, it changes very little.
Speaking of that power curve, we just put that in the datasheet. People can run it as hard as they want to, just like they do the current P1, and it's curve, that clearly says we can run those at 100Mhz too. And people do, all the time. 100Mhz is my favorite P1 clock.
Re: Cog timing. Yes! I completely agree with this suggestion. Can we add a little temp sensor of sorts? There has gotta be something we can read to see the state of things and adjust accordingly.
I'm against this, and I'm against it for a few reasons:
1. Public companies come with some rather brutal expectations, unless they are big enough to be valued enough for the key players to isolate their shares and remain in complete control of their company. Silicon Valley has seen the bell ring at the school of hard knocks on Wall Street a few times now. Look at what Zuck did with FB and the dual stock share strategy. Zuck holds Class B stock, with 10x the voting of the class A stock, and it's his company, despite public ownership. Awesome, and something the Wall Street guys with their expectations don't like much, but the perception of value in FB was so high, they had to have stock anyway.
Kids, that doesn't happen every day, and the vast majority of companies who go public won't have that Zuck experience, instead finding themselves on a quarterly treadmill that makes exactly none of the things we like to do here possible. Think about it.
2. Culture. Set aside the financial expectations. For those of us who have met Parallax, being a private company means being able to make some decisions that are the right ones, but not optimal financially. This is what makes Parallax what and who it is. I value that very highly. I think most of us do, and I know Ken, Chip, et al. do.
3. IP. A public company would have some additional IP concerns that again break what we are doing here.
To me, on a basic level, this is a no-go.
That conversation has already started… don't remember exactly where. The answer was yes.
If temperature management is going to be a real issue, then it deserves the usual forum treatment:)
By the way, could we do something about the timestamps here?
How about moving the clock to Rocklin time or possibly Australian time:)
Thanks. I now have my time set for Brisbane.
Many of the issues you raise are flags… not show stoppers. The key is the charter (which grants the rights) and the board structure(which manages the responsibilities).
It is their offer, if people don't like it, they don't invest.
If Chip and Ken don't like the eventual direction… they can be freed by the charter to do just about whatever they want. Anticipating the problems and handling them
properly in the charter is the key. There will be lawyers involved… you have to feed them.
Whatever the charter does to protect Ken and Chip's capacity to direct the company, investors will be there. Parallax has an international reputation. They could probably get enough to do whatever they want on the Hong Kong exchange. The Chinese love business ethics but rarely see them. They and are always happy to find an honest American to invest in:)
Rich
If you really need the money, you are going to give up the control. If you don't, then you might have a shot at doing what Zuck did.
And that, of course, also depends on the broad appeal inherent in the proposal. We don't have that broad of appeal here, so then we would yield a lot of control. Like you say, they don't have to invest. And I'm saying they won't without the control being handed over, and so we need to figure this out private or give up who we are.
Seriously. I've been around that enough to know better.
As far as funding goes, one or more LLP loans would be best because the only thing lenders get are interest payments until the principal is repaid (I.E. not influence whatsoever in company decisions as defined by the LLP "Limited Liability Partnership" law). There is no stake in the company. Kickstarter is an interesting idea, but it is not time-tested like an LLP and I'm personally very worried about such things as cross-border contributions violating our investing (SEC) laws. I suspect most kickstarter contributors are not qualified investors. LLP loans can be made with anyone using a simple standardized legal form.
Have you looked at "power pad" packages? The ADS1278 from TI has a 64-pin TQFP package with a power pad. That should let you dump around 2 watts into a ground plane in a easy to hand-solder and inspect package. (i.e. for the power pad use vias larger than your soldering iron, and fill them from the back of the board)
Marty
P.S. have you looked at using a smaller DRAM process? The few articles that have mentioned this say DRAM processes are a good bit slower, but like 1/5th the cost per transistor.
This problem is like trying to fit a large block V8 in a mini. A lot of you guys are essentially saying things like "lets change the spark plug wires, that'll make it fit" or "if we use a different fuel injection system, it'll work" or "let's paint the block blue instead of red, then it'll surely fit". The real solutions are more dramatic like switch to a 4 banger, reduce the engine to a small block with much smaller bore pistons, or pick a different car to put the large block V8 into.
You mean we have a choice of colors???
V8 in a mini you say? No problem:
http://www.youtube.com/watch?v=45PDbowUh8c
http://http://www.youtube.com/watch?v=6iK4dy74ibY
And my favourite
http://www.youtube.com/watch?v=lh1DoGzHFl4
Never underestimate how nuts the Brits can be:)
I do take your point though.
Correct, and Chip has indicated ~90% is already clock gated, and adding this small detail, is not too hard.
Power is set by nodes toggling, and the device only every does one opcode at a time..
Clock gating is already used to lower power, so all this diversion over what details to remove miss the point.
That discussion makes sense, when P2 fails to FIT, not when the Power envelope is higher.
I checked some Thermal specs on Power PAD TQFP128, and they are
0.12 °C/W jc .from die to case, and
17.17 °C/W. ja PCB mounted, thermal vias
( source tvp5158 TQFP 128-Pin PowerPADTM Package )
So that package is capable of some watts.
To normalize the figures, 5W Max total is 0.625W/160MHz/COG, and if we target say +50°C, that's 2.912W.
Using that, all COGS can run at 93.184MHz , which makes a ~100MHz target possible,
A 100MHz target makes more marketing sense, than 80MHz.
If 2 COGS tun at 160Mhz, that allows the power budget to look something like this for a sub 3W thermal profile
0.625*40/160+0.625*5*80/160+0.625*2 = 2.96875W
ie
1 COG at 40MHz (160Mhz time precision)
5 COGS at 80MHz (160Mhz time precision)
2 COGS at 160MHz
I think that type of COG combination covers a lot of use cases
There are also cheap SMD (solderable) heatsinks, that can add maybe another watt, perhaps 2, to this, or just buy cooler operation.
Next, would be the Aluminum top-hats, and really serious users, can always use this !!
Now that we have a nearly complete car, it simply time to get it on the road. Big cars can be driven hard, or not. That is up to the buyer of the car.
We made a pretty nice, powerful car. Seems to me, we figure out how to package that car up and get people driving it.
I think I want the liquid cooler.
Err no, wrong analogy.
The issue here is not physically fitting, it is Power/Thernal Profile, which is a rather different problem.
OK, V8 going into a mini which only has so much cooling capacity with its radiator and no room to put in a larger radiator.
Change plug wires? paint it red? Add fans? Etc?
Does that work better?
The term is used loosely here, it means multiple MHz specs, not outside spec.
Like all Logic devices, the P2 has a Thermal Profile, if you want more MHz, you put more effort into cooling.
Some companies spec Commercial and Industrial operation on the same part number now (used to be 2 part codes)
The Commercial spec has a higher MHz value, purely from this thermal Profile.
So it makes sense to have a Spec that covers
a) Minimal thermal effort --> Give the users a MHz spec for this point.
b) More designed thermal spread, more layers, more copper --> Give the users a MHz spec for this point.
c) Added heatsinks, eg simple, solder-able ones --> Give the users a MHz spec for this point.
I was just recalling the time I first ever heard of CMOS logic. Back in 1970 something. As opposed to the TTL us kids all knew and loved. The article was all about low power and even had a circuit running off a lemon.
Where did it all go wrong?
The idea of a micro-controller with a heat sink on it is more than I can bear.
If it was a standard CPU, then we could use liquid nitrogen or whatever.
Hehe yes, that's closer, and I've actually seen exactly that problem tackled.
They solved the "no room to put in a larger radiator." with a little lateral thinking, and fitted a second radiator underneath the car, below the passenger seat.
What we see here is a concrete example of a 'cost' for having every COG be identical.
Allowing individual COGs to run at a reduced speed is a step that will be helpful and is a way forward from where we are today and may make the current design viable.
IMHO in the long term the philosophy needs some tweaking.
C.W.
Then are you ok with this ?
0.625*40/160+0.625*5*80/160+0.625*2 = 2.96875W
ie
1 COG at 40MHz (160Mhz time precision)
5 COGS at 80MHz (160Mhz time precision)
2 COGS at 160MHz
Agreed.
I think this also means that some means to measure COG temperature is going to be needed.
Chip was talking of mapping RC Oscs divided to buried pins - that may be enough for Temp sense, or a proper temp cell may be needed ?
A COG, despite it's many transistors consumes almost no power if it is not doing any work. Or at least I hope that is the case.
As a UART or PWM driver or whatever it has a lot of "free" time. When it actually has to work that takes energy of course. It should not be sitting in "idle loops" wasting power.
Any way, regardless of that, Moore's law has come to an end and all the worlds chip makers are looking at parallel solutions to make performance increases.
The "Propeller Philosophy" was perhaps ahead of the game. If it does not scale they all have a problem.
Really can anyone make valid sales pitch for the P2 with a heatsink over a equivalent ARM or Pic32?
Only in northern climates if you can remove the heatsink when the office gets a little chilly!!
I agree. The vision of needing a heat sink on a Propeller is as unthinkable as Mr. Creosote eating "just one more wafer thin mint"
However: There is no such equivalent device. (As far as I know)