Parallax Propeller 2

[Sal Ammoniac]

The Parallax Propeller (P1) has been around for over a decade, and development has been ongoing for its successor, the Propeller 2 (P2) for almost as long. P2 development has taken so long that I haven't really been following its progress, so I was surprised recently when I went to the Parallax website looking for a sensor and found that the Propeller 2 is now an actual product.

In many ways it's similar to the original Propeller in that it has 8 32-bit processor cores and general purpose I/O ports that are used to implement peripherals such as I2C, UARTS, and SPI. The P1 has 32 I/O pins, and the P2 ups that to 64 I/O pins. Each I/O pin can be used as either digital input/output (serial/async, PWM, quadrature decoder, and a few other functions), an 8-bit DAC, or a delta-sigma ADC.

Hub RAM has been expanded from 32KB to 512KB and there's now a 32-bit CORDIC coprocessor. They've implemented a hidden debug interrupt for breakpoints and single-stepping. The max CPU clock has been raised from 80 MHz to 180 MHz.

Overall, it's a good mix of new features and capabilities, but there are some disappointments as well. I was hoping they'd increase the number of CPU cores (called Cogs) from 8 to 16 (or more), but that didn't happen. It's also unclear to me whether each cog can execute code from hub RAM directly--the P1 could only execute code from the 2KB of RAM local to each cog. The P2 ups the cog-local ram to 4KB, but that's still a big restriction and it would be nice if cogs could directly execute code in external (to the cog) memory.

Breadboard fans will be disappointed that the P2 is only available in a 100 pin TQFP package (the P1 was available in a 40 pin DIP package).

https://www.parallax.com/propeller-2/

[tggzzz]

Hardware parallelism is easy; it has been done many times over the decades.

Software that can make use of the parallelism is the weakness.

An ecosystem that merges system architecture/patterns with hardware capabilities with software capabilities is a rare thing. We need more and better.

[SiliconWizard]

I was intrigued by, but never really interested in the previous version, but the v2 looks like a nice improvement.

Reading the documentation, yes, the cogs can run code from hub RAM (there is 512KB of it):

When the PC is in the range of $00400 and $FFFFF, the cog is fetching instructions from hub RAM.  This is commonly referred
to as "hub execution mode."  When executing from hub RAM, the cog employs the FIFO hardware to spool up instructions so
that a stream of instructions will be available for continuous execution. Branching to a hub address takes a minimum of 13
clock cycles.  If the instruction being branched to is not long-aligned, one additional clock cycle is required.

There is a restriction though, that could be annoying:

While in hub execution mode, the FIFO cannot be used for anything else.

And all the cogs can also access hub RAM simultaneously for data access.

I took a look at the Spin2 language, and it doesn't look particularly appealing to me. I'd be curious to give this chip a try though.

[Sal Ammoniac]

Reading the documentation, yes, the cogs can run code from hub RAM (there is 512KB of it)

Thanks. I haven't gotten that far in the documentation yet, but it's nice that the previous restriction is (mostly) lifted.

And all the cogs can also access hub RAM simultaneously for data access.

That's interesting. So it's not really a "hub" anymore, at least not in the same way it was on P1 where a cog could only access hub RAM during its slot.

I took a look at the Spin2 language, and it doesn't look particularly appealing to me. I'd be curious to give this chip a try though.

Spin is a curious language--useful for quick one-off tasks and proof-of-concepts that would take longer on a more traditional MCU. I've implemented a few projects with the P1 using a combination of Spin and PASM (the P1's assembly language). The IDE is fast (it's written in x86 assembly language), but has no debugging facilities at all. The ecosystem is small and mainly restricted to the forum on the Parallax website. There also used to be a bi-monthly column devoted to the Propeller in Nuts and Volts magazine, but I don't know if it still exists.

[ali_asadzadeh]

Do you have any price info on the chip itself?

[woofy]

https://www.parallax.com/product/propeller-2-p2x8c4m64p-multicore-microcontroller-chip/

[ali_asadzadeh]

The price is so high! it does not make any sense

[BradC]

I took a look at the Spin2 language, and it doesn't look particularly appealing to me. I'd be curious to give this chip a try though.

As a pascal user Spin made a bit of sense to me, then I wrote a spin compiler and it made a lot more sense. There’s a strong history there going back to the basic tokenisers Parallax have had for years.

I still use the prop for quick prototypes, but for 99% of what I do can be done in a PIC for a 10th of the cost.

[Sal Ammoniac]

The price is so high! it does not make any sense

Given the amount of work that went into the P2, and the small volume they're likely to ship (it is a very niche product, after all), that's the only way for them to recoup their development costs. Parallax is a tiny company with only 35 employees--hardly an ST or an NXP.

[SiliconWizard]

Fifteen bucks for what can be considered a specialty chip is NOT expensive.

[techman-001]

Fifteen bucks for what can be considered a specialty chip is NOT expensive.

$15 bucks is dirt cheap from my POV. The first micro I ever saw, the National PACE 16 bit cpu in 1976 cost $270 each.

I paid $11.50 USD each for six STM32H7xx  in Jan 2021.

Can you even get a pizza for $11 USD any more ?

[ali_asadzadeh]

Given the amount of work that went into the P2, and the small volume they're likely to ship (it is a very niche product, after all), that's the only way for them to recoup their development costs. Parallax is a tiny company with only 35 employees--hardly an ST or an NXP.

The price is high, they should make it around 1-2$ each, so the sales would sky rocket, with 15Bucks I can have a zynq20 or an STM32H7 with two Gowin FPGA's with 20K LUT and 32MB sdram each! or I can have 3 allwinner GHz CortexA parts! the choice is yours.

[Sal Ammoniac]

Given the amount of work that went into the P2, and the small volume they're likely to ship (it is a very niche product, after all), that's the only way for them to recoup their development costs. Parallax is a tiny company with only 35 employees--hardly an ST or an NXP.

The price is high, they should make it around 1-2$ each, so the sales would sky rocket

Skyrocketing sales are not going to happen, at any price. The P2, like the P1 before it, is a quirky, non-mainstream MCU that's unlikely to derail the ARM juggernaut. If anything has a chance to replace ARM in the MCU marketplace, it's RISC-V, not the Propeller.

[tggzzz]

For any radically different item (processor/language/etc), there has to be a clearly articulated way in which it is radically better than the existing mainstream offerings. Ideally "radically" means ten times; two times isn't sufficient.

I never saw such a USP for the propeller.

[SiliconWizard]

The Propeller chips are niche controllers IMO, but they have been around for 15 years, so I guess they have a market.
If they don't fit your needs, just use something else.

Why would some people assume that all MCUs should be exactly the same at an ultra low price? Such MCUs do exist, just use them and let people with other views and targetting other markets do their job. I find it funny that some would challenge the relevance of products that have been around for that long - just start your own company designing and selling MCUs, and let's see how far you get.

As to what tggzzz said - which is a more technical point - I personally agree, but I'd still be curious to give them a shot and see if my prejudice is justified.

Edit: Also note that the Propeller 1 has been made open source for years now: https://www.parallax.com/propeller-1/open-source/

[Bassman59]

The price is so high! it does not make any sense

I don't know, people happily design boards that use off-the-shelf Arduinos as the compute/control element and the Nano is $20.

[tggzzz]

As to what tggzzz said - which is a more technical point - I personally agree, but I'd still be curious to give them a shot and see if my prejudice is justified.

I quite agree. Kicking the tyres can be fun and informative.

When I looked at the XMOS xCORE+xC the USP was immediately obvious: hard realtime limits guaranteed by design, not by executing/measuring and hoping you had encountered the worst case.

I didn't see anything comparable for the propeller, just that it was different.

[Sal Ammoniac]

The Propeller caters mainly to the hobbyist and education markets, neither of which are high volume, nor are they ever likely to become so. The P1 has been around a long time because it doesn't cost Parallax much to keep producing it as long as there is a small demand for it.

It really doesn't matter what anyone here thinks about the relevance of the Propeller, the XMOS parts, or anything else--the market will decide what lives and dies. Parallax have apparently been selling enough P1 parts over the last 15 years to keep it in production.

[Monkeh]

Given the amount of work that went into the P2, and the small volume they're likely to ship (it is a very niche product, after all), that's the only way for them to recoup their development costs. Parallax is a tiny company with only 35 employees--hardly an ST or an NXP.

The price is high, they should make it around 1-2$ each

This coming from the guy who thinks he can manufacture DRAM sticks cheaper and better in small quantity than the companies which sell millions of them a year.