I think this thread is quite interesting also. Since I have been learning forth, I think I would like a GA144 chip as a coprocessor. Unfortunately it is a 1.8V archetecture, 18Bit(kindof odd), starved for memory (64 words RAM/"computer"), and not in production yet(due Q2 2011), and lacks the support I am used to(this forum is great!) but they are really fast ~700mips/core peak, and low power(~100uW at idle).
Maybe when the Prop II comes out, since it will be able to do 1.8V io, I can strap on one of these little 144 core chips for some added performance.
XMOS would be cool too because of the ability to use C, and the larger memory architecture.
Both of these companies can learn quite a bit from Parallax when it comes to support. I just have not seen anything that compares to these forums, and parallax's support and openness.
Well, yeah. If anyone are building a supercomputer, XS1 chips and prop chips together would make an excellent addition. Basically, using XS1 with Prop's like "Match made in Heaven", much like x86 computers being interlinked to those famous transputer, T414 chips made by InMOS (which I suspect XMOS was InMOS back then...) - for revving the main processors up.
And, bad news: XS1 also lacked floating-points... Oh well, we can hack away at it, though. (Just like we do with Propeller chips...
Another bad news: They get quite hot... (So, Propeller do get hot too, anyways.)
And, I am considering using them both (XS1-G4 and Propeller II chips) for my own supercomputer - I may need to do rather fancy programming tricks on XS1 to make them real useful for my supercomputer, Dendou Oni, otherwise they may be overwhelmed.
Dr. Mario wrote: And, bad news: XS1 also lacked floating-points... Oh well, we can hack away at it, though. (Just like we do with Propeller chips...
Another bad news: They get quite hot... (So, Propeller do get hot too, anyways.)
Just curious, what are you doing to make a hot prop? As long as you use recommended specs, the prop chip does not go hot. I have tried numerous configs and platforms and never noticed any heat increase. If you go beyond specs and push it, perhaps straining it with not-recommended overvoltages, excessive current draw and excessive overclocking, it may need cooling.
After looking at the XMOS datasheets I think the ideal solution is to use the XMOS as the processor due to its larger memory and ease of programming development (C and XC compliers) and then use the prop to implement the various peripherals i.e. UARTS, SPI, I2C etc as well as a FP co-processor perhaps.
I'd use also the XMOS as co-processor. It has a 1 cycle MUL, something the current prop lacks. Those peripherials are low speed and can be easily implemented. SPI is particularly covenient due to the cloked ports, something I hope the next prop has. The advantage of the prop is the easier start-up: 3.3V regulator, no power sequencing, built-in reset, DIP package.
Yea. Yea. Humanoido, Props wouldn't get hot during normal usage (USUALLY!), but overclocking can also affect the die temperature like it do to our desktop CPUs (although on my desktop computer, the AMD Athlon 64 X2 wouldn't get hot at all - 20 - 30 degrees all years long, ever since I modified the AMD Phenom retail fan, topped it off with a powerful Sunon 80mm fan, shimmed with 80mm to 70mm fan funnel).
XMOS XS1 as a co-processor is a neat idea, of course, since you can have Propeller spit out the microthreads for the XS1 chip to munch on, thus doubling both chips' usefuliness.
Also, in my opinion (software stand-point) that XMOS XS1 can also be forced to run Out-of-Order by overriding its thread scheduler (something I want to do for stuck or stalled threads,
to keep the ALUs busy without really wasting electricity, or even the CPU cycles - by intentionally introducing software-based dynamic rescheduler or to introduce delayed issues on
every single programmatically similar [same instructions but different data] threads), something that Propeller 1's COGs can't do (Prop II's COGs can be made to run Out-of-Order,
only if you or I put in much efforts to learn what really make a COG ticks, and to counter its inner workings with software, pretty much similar to what you would do with TI VLIW DSP chip).
Definitely right on, Leon, it's not an easy task to be the first to connect together lots of props or lots of anything. But once you get there with the completed machine, it's like discovering the New World, and if a person is like me, it's a very enjoyable part of life.
I have a prop connected to a stamp so it can be a coprocessor, and after getting the hardware and software just right, it's remarkable the capabilities it adds. So adding in terms of X could be be additional too. I'll need to go back and review the schematic and software you posted for your new xprop.
Yea, I agree! Leon really have went so far to get neat features within reach of Propeller - it also could add speed boosts to the propeller chip.
Still, the idea of using XMOS XS1 as coprocessor is neat. I will go on with that! (I am still picking parts for Dendou Oni - it gotta be right parts to get together: It's the parts that can make or break your project...)
This is some of your best work with the Propeller chip.
Really? I've read about many far more difficult things Leon has accomplished. As for the XProp, it's just getting started. Seems like most of the work is ahead, not behind. Anyone on this forum can draft a board containing one Prop and a couple of headers.
Comments
Maybe when the Prop II comes out, since it will be able to do 1.8V io, I can strap on one of these little 144 core chips for some added performance.
XMOS would be cool too because of the ability to use C, and the larger memory architecture.
Both of these companies can learn quite a bit from Parallax when it comes to support. I just have not seen anything that compares to these forums, and parallax's support and openness.
Doug
And, bad news: XS1 also lacked floating-points... Oh well, we can hack away at it, though. (Just like we do with Propeller chips...
Another bad news: They get quite hot... (So, Propeller do get hot too, anyways.)
And, I am considering using them both (XS1-G4 and Propeller II chips) for my own supercomputer - I may need to do rather fancy programming tricks on XS1 to make them real useful for my supercomputer, Dendou Oni, otherwise they may be overwhelmed.
Another bad news: They get quite hot... (So, Propeller do get hot too, anyways.)
Just curious, what are you doing to make a hot prop? As long as you use recommended specs, the prop chip does not go hot. I have tried numerous configs and platforms and never noticed any heat increase. If you go beyond specs and push it, perhaps straining it with not-recommended overvoltages, excessive current draw and excessive overclocking, it may need cooling.
This is some of your best work with the Propeller chip.
XMOS XS1 as a co-processor is a neat idea, of course, since you can have Propeller spit out the microthreads for the XS1 chip to munch on, thus doubling both chips' usefuliness.
Also, in my opinion (software stand-point) that XMOS XS1 can also be forced to run Out-of-Order by overriding its thread scheduler (something I want to do for stuck or stalled threads,
to keep the ALUs busy without really wasting electricity, or even the CPU cycles - by intentionally introducing software-based dynamic rescheduler or to introduce delayed issues on
every single programmatically similar [same instructions but different data] threads), something that Propeller 1's COGs can't do (Prop II's COGs can be made to run Out-of-Order,
only if you or I put in much efforts to learn what really make a COG ticks, and to counter its inner workings with software, pretty much similar to what you would do with TI VLIW DSP chip).
I have a prop connected to a stamp so it can be a coprocessor, and after getting the hardware and software just right, it's remarkable the capabilities it adds. So adding in terms of X could be be additional too. I'll need to go back and review the schematic and software you posted for your new xprop.
Still, the idea of using XMOS XS1 as coprocessor is neat. I will go on with that! (I am still picking parts for Dendou Oni - it gotta be right parts to get together: It's the parts that can make or break your project...)
Really? I've read about many far more difficult things Leon has accomplished. As for the XProp, it's just getting started. Seems like most of the work is ahead, not behind. Anyone on this forum can draft a board containing one Prop and a couple of headers.