Can an FPGA P1 ever be cost-effective?

Dave Hein

I'm just curious whether a P1 implemented in an FPGA could ever be cost competitive with a P1 chip solution. The FPGA that is used for the BE Micro CV board goes for around $45, and it's my understanding that it can support only one P1V, plus some extra logic. It's been shown that the FPGA used in the DE2-115 board can support 5 P1s, but that chip cost $470 at digikey. So I'm just wondering if any implementation of the P1 in an FPGA would ever be practical, except for cases where a custom feature is needed, and cost is not important.

mark

I admit that I'm certainly no expert, but I don't see how they could ever be price competitive considering that at the very least, an FPGA will require significantly more silicon to implement the chip, or be manufactured on a more expensive process. Except in the event where you have an extremely small customer base to spread your costs across (which wouldn't happen unless said customer is the US Gov).

BTW, does anyone know approximate tooling costs for chips manufactured on the 180, 250, and 350nm nodes?

rjo__

I think the question is whether a P1v can be cost competitive... and then the question is "cost competitive with what?"

I get Heater and Jazzed comments all mixed up, but one of them made the point that many times, the issue isn't cost, but availability... which begs the question,
"what can you do with an P1v that you can't do with a P1?"

I am myopic on the issue. I see only my own interests, but even in the narrow margins of my vision, I can see a lot that the P1v offers that the P1 doesn't.
That will change when the P2 comes out of the pipeline, but it is certainly true today.

I would strongly encourage everyone in medical device r&d to put an intern on the issue of the P1v, just to see what it might offer to reduce costs, simplify designs, and improve performance.
I wish Parallax would get their new board(s) out the door, so that the net gain to the companies would create a net gain for Parallax.


Rich

Cluso99

It's quite obvious that the prices all the companies are charging for the fpgas are not the prices paid in large quantities.

Look at all the FPGA boards and you will mostly find that the FPGA cost approaches the total price of the pcb. The configuration Flash chip is way overpriced too. From what I have found the FLASH chip goes for ~$16 whereas an equivalent SPI FLASH chip goes for < $1.

However, I would never expect that an FPGA with a P1 could be obtained for < a real P1. But, the added features you can place in an FPGA may make it cost effective for some products. Also remember, there is a licensing issue to be resolved.

Bob Lawrence (VE1RLL)

re:DE2-115 board can support 5 P1s, but that chip cost $470

You have to compare the price to an individual FPGA chip , a PC board and some support logic and not a full experimental development system such as :DE2-115 board. Also, you would normally purchase them in bulk at a discount unless it was a 1 time project. (example pricing )


EP4CE115F29C8N
IC FPGA 528 I/O 780FBGA
Embedded - FPGAs (Field Programmable Gate Array)
Cyclone® IV E

1
58
341.36








Prices:
http://www.buyaltera.com/scripts/partsearch.dll/multisearch?site=ALTERA&lang=EN&keywords=EP4CE115

If someone experimented with the P1V and make a great design that had market value they would most likely release it as a regular chip and not on a FPGA.chip.


ALthough FPGA's are pricey now they are droping in price fairly rapidly.

Dave Hein

The $470 price at digikey was for the FPGA chip, and not a DE2-115 board. The $341 price is better, but is still many times higher than $40 for 5 Propellers.

Bob Lawrence (VE1RLL) wrote: »

If someone experimented with the P1V and make a great design that had market value they would most likely release it as a regular chip and not on a FPGA.chip.

Based on the comments on the thread about producing a chip from P1V, it doesn't seem likely unless Parallax drives the development. Most of us would rather see the P2 ASAP, so it doesn't seem like a P1V+ chip would happen in the foreseeable future, if at all. If anything, P1V has delayed development of P2.

jmg

Dave Hein wrote: »

... If anything, P1V has delayed development of P2.

Overall, I think not.

Sure, it cost Chip a few days to get P1V release ready, but it has already seeded a LOT of FPGA development boards, and that will translate into a LOT of P2 testing , so I think the overall P2 release time line is actually better off as a result.
Chip also sounded keen on testing elements of the P2 Smart pins, on P1V boards.

Tubular

Dave, its not as much of a stretch as you might think. You can get suitable cyclone iv's (as used De0 Nano) from Aliexpress around the $15 mark. To me that indicates the qty price (probably for 10k+ qtys)

Fully expect these will drop over time.

Dave Hein

jmg wrote: »

Sure, it cost Chip a few days to get P1V release ready, ...

Well Chip said he would post a P2 image a week or so after August 10, so if "a few days" is 5 days then we're on target if we get a P2 image tonight. I guess whenever the P2 image is available we'll know how much P1V impacted P2 development.

jmg

Dave Hein wrote: »

I'm just curious whether a P1 implemented in an FPGA could ever be cost competitive with a P1 chip solution.

For a only 80MHz part, 3.3V single rail and low power, I would expect the P1 chip will always win.(for at least another 5 years)
- but there are some places a P1 chip cannot go.
Already P1V has higher MHz, and spare logic, so it becomes what those extras are worth to the designer.
Or, maybe the designer needs a package smaller than P1 Chip, but only need a few COGs & RAM ?

Dr_Acula

Prices are falling all the time - I think it will be cost effective in a few years. Take a Z80 plus all the peripheral chips - that is maybe $60. Or emulate it on a cyclone II for $18. I mention a Z80 because it is possibly on a par with a single cog. Getting technical, there still is not quite a cost effective Z80 solution in a single fpga that does not require external ram . You get maybe 30k to play with and need 64k for a Z80 so it is cheaper to add a physical ram chip. Looking through the various cyclone fpga families, the low end ones are the cheaper ones and they tend to need external ram. You can work out the extra cost of internal ram on a fpga vs the cost of a real ram chip (about $3) and it is still cheaper to go external ram. Internal vs external ram is an important point because with external ram, only one thing (cpu/process) can be accessing the ram at any one time, but with internal ram, you can split the ram up and have, say, 8 cogs working with their own ram.
I think the sub $20 fgpa boards are the interesting ones to watch, because these boards have other things already added like a xtal and regulators. So if you are comparing like with like, the comparison is a real propeller plus pcb plus regulator plus xtal plus eeprom and that is about $20.

So - like for like, how much can you get into a $20 low end cyclone II, III or maybe IV? I'm guessing a few cogs, and then maybe the hub as external ram. Only one process at a time looks at external ram (ie the propeller part of the propeller) so that could be done in external ram so you can add another $3. So I'm going to guess that it might be possible to get some sort of stripped down propeller working with a couple of cogs for about the same price.
And then the prediction is - if you can do one or two cogs now, and the prices keep falling, how long before it is 8 cogs? I'll guess maybe two years?

And of course, there are other smart solutions. Say you had one cog doing SD card, and one for keyboard, and one of VGA, and one for serial port(s) and you only actually were using one cog for code, you could port those other peripherals into softcores in fgpa for the same functionality, and then I think it might actually be possible to get a one cog propeller working now, for the same price as a real propeller, and with the same display/keyboard etc.
Very exciting times!

Cluso99

Drac,
If you compare a Cyclone II board that maybe built in 10K quantities by Chinese with very small margins, then you need to compare a P1 board built in the same quantities and margins.
The P1 board would be sub $10.

LoopyByteloose

I suppose the real question is whether FPGA devices are going to at some point decline in cost to the point that Parallax will chose to quit developing a chip of its own and market FPGA devices instead.

At this point, I am still trying to ignore FPGAs. My interest in Propellers has been purely hobby. And at 67, my memory skills are not so great while my construction skills are less precise.

I have been optimistically waiting for the P2. A shift to FPGAs may never happen for me.

Tor

My very subjective impression is that FPGAs aren't dropping substantially in price, except when compared to the earlier years. It looks to me that the vendors instead replace older versions with newer, more powerful (or at least different) types, and thus the price drop doesn't happen. But that's just my impression from the relatively small sample I have looked at over the last, say, 3-4 years, and I may of course be wrong. Which wouldn't be bad actually.

Heater.

I would imagine that a P1 in an FPGA may be cost effective in designs that need some external logic as well. It could save on total device count for the system, shrink the board size, simplify production and increase yeild.

Also the flexibilty of the finished design may result in cost savings.

An FPGA as only a P1 I don't thing will ever beat the real chip on price.

VBB

What would be the minimum FPGA device that could run the P1V?

Seairth

P1 happens to ideally fit FPGAs because it was all purely digital I/O. P2 and on will have analog I/O, which will not be a good match for FPGAs (as a target platform).

jmg

Tor wrote: »

My very subjective impression is that FPGAs aren't dropping substantially in price, except when compared to the earlier years. It looks to me that the vendors instead replace older versions with newer, more powerful (or at least different) types, and thus the price drop doesn't happen. But that's just my impression from the relatively small sample I have looked at over the last, say, 3-4 years....

Largely that is correct, in that rather like flash drives, FPGAs have packed more in, but the price-point has changed little.
- worse, the more capable FPGAs are now in BGA only, which lifts the PCB/Assembly costs.
Some CPLDs have seen better price points, and package choices, but those cannot fit 8 COGS.

I can see an opening for a P1 Chip plus a CPLD with 1(+) P1V COG + Special Logic.


The MAX 10 from Altera will be interesting - they claim it can run NIOS and is 55nm Flash.
Has anyone seen hard data on that, or done a build on the beta tools ?

rod1963

I think so.

I know that hobbyists have made a go FPGA based retro computer boards like the MIST (200 Euro)
https://code.google.com/p/mist-board/
http://lotharek.pl/product.php?pid=96

I see another guy offering another FPGA/computer board based on the Lattice XO2 with 7000 LUT's (Fleasystems.com) for $65.00. It can handle a complete 68K SOC

But these are based on 144pin tqfp FPGA's not 400+ pin BGA monsters.IMO we would need to look at implementations that can use non-BGA fpga's.

KeithE

I wonder if it would fit into one of the Atmel rad-hard FPGAs?

Kerry S

It all depends on what you are trying to do.

A commercial application where the P1 needs a custom PC board and a lot of external stuff to get the job done? Yes the P1V could be very competitive for advanced application solutions. My P1 proto board cost me $66.00 each (with a 4 PC board order, no components) and I can do it better with the $60.00 BeMicro CV.

That is why I am really hoping that the Parallax FPGA board is designed to be very flexible with commercial applications in mind.

VBB

Kerry S wrote: »

My P1 proto board cost me $66.00 each (with a 4 PC board order, no components) and I can do it better with the $60.00 BeMicro

It certainly makes sense to buy an FPGA board to learn BOTH propeller AND open up the world of FPGA also. Actually I think this speaks to one of the risks of releasing the P1V. Due to the high volume of 'off the shelf' FPGA boards it will be a) hard for parallax to offer their own FPGA board with enough margin to make it highly profitable and b) potentially eat into the sales of any P1 developer boards from parallax or 3rd parties. Given the nice margins on dev boards historically enjoyed by parallax hopefully the PROS of P1V outway these CONS.

prof_braino

Dave Hein wrote: »

Well Chip said he would post a P2 image a week or so after August 10, so if "a few days" is 5 days then we're on target if we get a P2 image tonight. I guess whenever the P2 image is available we'll know how much P1V impacted P2 development.

I don't think there is any hard schedule dates for P2. So what if the P1 V cost a a couple weeks, or a couple months? The P2 is already YEARS over the original guestimate. Any impact will likely be "negligable" at most. It just goes to show that the only reliable estimate can be based on how long something took LAST time. In the case of brand new development, there hasn't been any last time yet, so any estimate is no better than any other wild guess.

Not recognizing this is the primary injection vector for errors attributatble to management in most profession projects I've worked on. This is how management sabotages their own organization's efforts, and its proven extrememly effective.

Dave Hein

Doug, you are rehashing a discussion from 9 days ago, and this was also discussed on another thread. Yes, I know that schedules from Parallax are meaningless. As far as P1V, I would much rather see P2 earlier than to have Chip spend even a second on P1V. Most companies would set project goals and then strive to hit them. Without goals and schedules projects will progress at a slow rate because developers have no sense of urgency and focus.

Parallax is a different type of company. It seems to be more of a hobby for Chip, from which we have all benefited by the products that he and Ken, and the other employees of Parallax have produced. However, Parallax hasn't kept up with the rest of the market as far as chip development and tools. I do applaud them for taking the giant step to provide GCC for the Prop.

Most companies strive to grow and develop products that outperform their competitors. Parallax seems to be content to remain the same size, or grow at a very slow rate. This helps to maintain an enjoyable work environment with little pressure. However, I find it a bit frustrating from a user perspective. We have been promised the P2 for as long as I've been using the Prop, and it's still not here. I would just like to see a firm commitment from Parallax to hit at least some of their milestones within a reasonable timeframe.

rod1963

I've given up on Parallax ever putting out a time frame for the P2. If I see working silicon before 2016 I'll be elated, but if they take longer, it's fine too. It's their company and they can do as they please. In the meantime there are plenty of interesting alternative micros and retro systems to play with.