Anyone use XMOS multicore chips?

[Corporate666]

I have a new project I will be jumping into in the next month or two that will require some slightly tricky timing and I have some other processes that will be a bit CPU intensive.  I've been worried how to manage both of those conflicting needs.... and last night while doing some intarwebbing I came across XMOS multi-core microcontrollers.  I figured they would be $20/ea but I was almost shocked to see they start in the $3 range from Digikey.

Of course, XMOS makes them sound like the greatest thing since sliced bread on their website, but before I buy a dev kit and download all their tools, I figured I'd ask here.  I did a search and only got a few results and nobody really seemed to be using these beasts.

So... anyone care to comment?  Are they good?  Bad?  What are the tools like?  Is the learning curve steep?  I haven't quite gotten the "no interrupts" thing... it sounds wonderful, but it also sounds like you could very quickly need a lot of cores if you need to use one for each separate process that would otherwise use an interrupt.

[marshallh]

Not horrible for something a bit more timing sensitive than a regular MCU.. It is nowhere close to approaching the capability of an FPGA with respect to timing and throughput, but it will be more suited to CPU type sequential processing.

Basically they occupy a very thin niche between "needs timing a wee bit tighter than 1 cpu" and "needs 5ns reaction or huge throughput"

[Paul Price]

If you were in Sicily, it would be better to make your deal the local Sicilian Mafia, and ignore the single Russian Mafia rep. asking you to do business.

Google around and you will see a wide difference between the number of sites and fans that could offer you tech help and code support. How long would it take you to Google around and find how many links exist to ARM or PICC or Arduino or 8051 chips derivatives, etc, and who has heard about XMOS, anyway? And yeah, there is great difference between the quality of the software and hardware tools that are available from the point of price or utility or ease of use or trying to find someone else who might have his head screwed on right to help you with a question. And beware, what is promised is not often so often delivered by software and compiler or debugging tools or the IDE can have bugs that can stop you dead in your tracks.

Make sure your processor group of choice does not strictly prefer to talk Chinese to communicate with others.

[Corporate666]

If you were in Sicily, it would be better to make your deal the local Sicilian Mafia, and ignore the single Russian Mafia rep. asking you to do business.

Google around and you will see a wide difference between the number of sites and fans that could offer you tech help and code support. How long would it take you to Google around and find how many links exist to ARM or PICC or Arduino or 8051 chips derivatives, etc, and who has heard about XMOS, anyway? And yeah, there is great difference between the quality of the software and hardware tools that are available from the point of price or utility or ease of use or trying to find someone else who might have his head screwed on right to help you with a question. And beware, what is promised is not often so often delivered by software and compiler or debugging tools or the IDE can have bugs that can stop you dead in your tracks.

Make sure your processor group of choice does not strictly prefer to talk Chinese to communicate with others.

I get your general point but I am confused about whether you have specific experience with XMOS or are talking in generalities about going with bigger names?

I'm not too worried about IDE bugs or going with a chip other than the market leaders.  I've been using Cypress PSoC chips for some years and I find them to be wonderful, despite having a pretty narrow user base that pales in comparison to Microchip/Atmel. 

Is there some specific experience you have with XMOS?

[Corporate666]

Not horrible for something a bit more timing sensitive than a regular MCU.. It is nowhere close to approaching the capability of an FPGA with respect to timing and throughput, but it will be more suited to CPU type sequential processing.

Basically they occupy a very thin niche between "needs timing a wee bit tighter than 1 cpu" and "needs 5ns reaction or huge throughput"

One application they seemed interesting for was driving arrays of LED's using the World Semi 2811 chip.  That chip has some relatively tight timing requirements, which can be met without too much trouble with a normal MCU.  But some guys are driving them with 16Mhz PIC's and AVR's but they are turning off interrupts to get there.  The multi-core approach seemed like it might work very well for that... and not having to sacrifice interrupts and other processing just to get the timing right for the 2811 chips.

I have no experience with FPGA work other than the limited similarities between them and the PSoC chips we use, so I tend to think trying to get into FPGA's would be a huge learning curve - much more than with the XMOS.  Maybe I am wrong?

[Paul Price]

No experience with XMOS, just the other chips mentioned. Can you find enough others who have enough experience with the XMOS experience to tell their story, relative to other chips...that is the question.

If you say you have done Cypress and you think this is the gal (pun intended) to marry, go for it, if she best fits your special needs.

[nctnico]

Did you look at the LPC4300 from NXP? It has two cores so you could dedicate one to ethernet handling and the other to time critical stuff. But maybe you can combine the two time cricital processes into one (something I often do) so they can get in eachother's way.

[millerb]

It's been a few years since I did anything with the XCore line, but I had a high opinion of it. XC is easy to learn and the devices are easy to work with. They put a ton of work into their tool chain and made it available for free from day 1 and it really shows. If you only had one thing to say about them, it's that they did the tools right.

As fun as they were to play with, I ultimately didn't use them in any products as they came in annoying packages, ran too hot and didn't really solve any problems I needed to slay. They may have fixed the first two issues but I doubt they've solved the third. As much as I liked it, it seemed to be a solution in search of a problem.