That was probably state of art when it was designed, probably 5 years before launch, so 2006.
I couldn't help but go look up what is current top of the line offerings ... BAE Systems still tops out at that very 400MIPS PowerPC that's in Curiosity. Atmel has a 90MIPS Sparc offering with some PCI+FPGA on board. And there is many slower parts, MIPS R3000 and Pentium1's would head that lot up, but seems to be about it.
It does look like 150nm is as fine as they're going. As we've seen with the Propeller, that limits how much can be packed in. BAE's RAD750 was announced in 2002 I think and, again from Wikipedia, has been operationally deployed from 2005. Not bad, less than three years from the starting blocks.
I'm wondering what the advantage of having one multithreaded CORDIC in the Hub is, verses having 16 ordinary CORDICs in the 16 Cogs.
Ha, I might have the info after all ... 2500 per P2Hot-Cog without mult/div, and 20000 for the whole P2Cool. Of the 20000, 8000 are mult/div. That leaves 12000 for the equivalent of P2Hot variant, so the mult/div part adds a further 67% ... 2500 * 1.67 * 16 = 66800.
20000 when Hub based vs an estimated 66800 if duplicated in the Cogs. I'll count that a solid space saving.
Flip-flop count is probably equivalent - pipelined vs more duplication.
That seems more in line with the reasoning used to come up with making the video system lean, smart pins, and this shared CORDIC.
Primary heat issue in the hot chip was way too many things toggling too much of the time. Big busses and a ton of muxes were needed to make it all work out. Chip identified several sources of trouble, and "minimizing the toggling per cycle" was the key to an efficient and capable P2. No reason to believe he's not maximizing that right now.
For the video system, the color model was a killer. All that correction, which we can do in software very reasonably, drove a lot of the heat. VGA support is the minimum needed to get us all the good stuff, and having a lot of COGS and seriously improved HUB throughput gets us the rest when wanted / needed, without the hardware issues. (composite, sprites, component, textures, etc...) Non hardware on those makes sense anyway. People who want 'em can build them easily, and then we've got them.
It's more in line with the Propeller design philosophy anyway. It's gonna be really flexible. Some of those things were kind of baked in, and nobody needs that, and it gets hot. Given what we made P1 do, this is going to be excellent and running much cooler.
Smart pins and the matching up of DACs to COGS got rid of another bunch of units switching, as did changing the HUB memory to the 16x throughput address interleaved model we have now. We are all going to have to give that a go, of course, but I think it will play out just fine. We all will see soon enough! Moving some complex things out to the pins means running cooler, and the HUB issues all discussed seemed to require hot types of constructs too. Plus, we seem to be getting some great measurement capabilities, and who knows what else? The high HUB throughput and Smart Pins are the most exciting parts to me.
I'm confident we won't get a hot chip this time. Chip has the data from the last analysis and is working under much improved design rules and with a team for this iteration. But, we did get a lot of great ideas from the last one, which was a lot of fun to work with. We are going to get the best of the possible, given the power constraints this time around.
Regarding the math, I personally think that is a killer feature, and it's going to be an attractive differentiator. The tasks it's useful for have already been mentioned, and they can be done at good performance, and doing that will be pretty easy. It was kind of stunning to just type in the math needed for fractals, for example, and do it right in PASM. It took me an afternoon, and it was like, "that's it?" The CORDIC bugs program is also small and efficient given what it did.
This kind of thing will make the tech accessible and it will perform on specialized things people often have to do a lot of work on. Of course, PASM is nice, but those nice features will roll right up to C, SPIN, whatever. I think this will get used a whole lot more than we think. Of course, it can all be done in software, as we see on P1, but that has sharp limits, and or really needs a lot of COGS.
One of the design goals Chip has is for the device to be interactive. Gather data via sensor, the DACS, whatever, perform operations on it, transform it, then display it in ways people can find useful. IMHO, that too is going to be fairly attractive given what we've seen done with P1.
Regarding the math, I personally think that is a killer feature, and it's going to be an attractive differentiator. The tasks it's useful for have already been mentioned, and they can be done at good performance, and doing that will be pretty easy. It was kind of stunning to just type in the math needed for fractals, for example, and do it right in PASM. It took me an afternoon, and it was like, "that's it?" The CORDIC bugs program is also small and efficient given what it did.
This kind of thing will make the tech accessible and it will perform on specialized things people often have to do a lot of work on. Of course, PASM is nice, but those nice features will roll right up to C, SPIN, whatever. I think this will get used a whole lot more than we think. Of course, it can all be done in software, as we see on P1, but that has sharp limits, and or really needs a lot of COGS.
One of the design goals Chip has is for the device to be interactive. Gather data via sensor, the DACS, whatever, perform operations on it, transform it, then display it in ways people can find useful. IMHO, that too is going to be fairly attractive given what we've seen done with P1.
I'd agree, and the broad market trend helps the Prop2 here.
There are more and more Linux Modules of the RaspPi / BeagleBone / CHIP being released, and then there are those WiFi portable display screens some call cell-phones.
Often those higher end systems are programmed in scripts, which are easy to craft, but not great for speed.
That leaves room for MCUs that can process sensor information, and do the maths, to have numbers 'script ready' at the other 'user' end.
The P2 will do all that, and allow sensor-add-in and control-add-in, in a way the single-core ARMs can only envy.
I think the CORDIC stuff plus the presence of the smart pins with DAC/ADC on each pin could make the Prop2 really useful for audio hobbyists -- in particular modular synthesizer enthusiasts... Eurorack modular components are a really huge popular market right now. Being able to do some DSP stuff (which the Prop1 isn't great at) and have the I/O there to handle a bunch of inputs/outputs and the like is really cool.
That leaves room for MCUs that can process sensor information, and do the maths, to have numbers 'script ready' at the other 'user' end.
The P2 will do all that, and allow sensor-add-in and control-add-in, in a way the single-core ARMs can only envy.
Agreed.
Re: Audio
The thing is going to be a playground for audio and many video applications. Fun times ahead!
I think a Eurorack compatible module with a bank of 1/8" inputs and 1/8" outputs, a wack of knobs and switches, maybe even a MIDI port or two, an LCD screen, and a USB port for programming could be a salable product. A generic sound module that DIY types could program to be whatever they like, using the onboard DAC/ADC. Supply it with some example applications in C++ ... yum.
I think the CORDIC stuff plus the presence of the smart pins with DAC/ADC on each pin could make the Prop2 really *snip* Being able to do some DSP stuff (which the Prop1 isn't great at) and have the I/O there to handle a bunch of inputs/outputs and the like is really cool.
Now if it'd just ship :-)
Yes. I suspect a significant portion of Prop 2 demand could come from people using it as an ADC/DAC for another chip. 64-channel ADC's are rare, and the only other one I've seen that came close was a >$50 chip. (anyone want to sell pre-programmed EPROMS to do this?) And having useful DSP capabilities on chip with your ADC is a wonderful bonus.
I think a Eurorack compatible module with a bank of 1/8" inputs and 1/8" outputs, a wack of knobs and switches, maybe even a MIDI port or two, an LCD screen, and a USB port for programming could be a salable product. A generic sound module that DIY types could program to be whatever they like, using the onboard DAC/ADC. Supply it with some example applications in C++ ... yum.
Program it in Spin and it would be even more delicious!
Seriously, though, that sounds awesome. Something like it would be perfect for an audio device I want to make. I would think 1/4" or XLR jacks would be way more useful than 1/8", though, especially for professional audio engineers. I've never heard of 1/8" jacks used for anything other than headphones and small speakers. MIDI would also be extremely useful. So would an SD card and maybe even VGA out and keyboard in.
Program it in Spin and it would be even more delicious!
Seriously, though, that sounds awesome. Something like it would be perfect for an audio device I want to make. I would think 1/4" or XLR jacks would be way more useful than 1/8", though, especially for professional audio engineers. I've never heard of 1/8" jacks used for anything other than headphones and small speakers. MIDI would also be extremely useful. So would an SD card and maybe even VGA out and keyboard in.
1/8" is the standard for modular synths in the Eurorack format, for CV/Trigger input/outputs.
I think a Eurorack compatible module with a bank of 1/8" inputs and 1/8" outputs, a wack of knobs and switches, maybe even a MIDI port or two, an LCD screen, and a USB port for programming could be a salable product. A generic sound module that DIY types could program to be whatever they like, using the onboard DAC/ADC. Supply it with some example applications in C++ ... yum.
Imagine what eric could have done with a Prop 2! :
Parallax should release the new P2 FPGA image on the first day of summer just to spite your impatience.
I suppose I am impatient, and it would serve me right if Parallax has already completed the P2 FPGA image and they are withholding it just to spite me. Of course, that means they would be withholding it from all the other people on the forum and their customers. Somehow that doesn't make a lot of sense to me. Personally, I think Chip and company are working toward meeting the goals that Ken and Chip have expressed on three different occasions, and I would expect that a P2 FPGA image will be available within a few weeks, and possibly within a few days.
If the first P2 image is available on the first day of Summer I will be overjoyed. I would feel honored if my reminders caused Parallax to focus on delivering the P2 FPGA image by a certain date.
Like Dave, I'm keen on giving this new design a go. Fun times are in our future, and it makes perfect sense to remind Parallax that we are here, and ready.
... it makes perfect sense to remind Parallax that we are here, and ready.
Seriously ... I really don't think they need reminding. The four apocalyptic horsemen: Destiny, Profitability, Reputation, and Survival have way more influence than any prodding we could ever hope to provide.
In any case, I don't understand how 'end of spring' could be as late as four weeks from now.. not even here (I'm currently in the north of Norway). Spring finished quite a while ago. In four weeks we celebrate mid summer.
I'm happy to wait patiently no matter how long it takes, be it summer or Xmas, I'll happily wait patiently knowing it'll be awesome when it's here! Don't rush Chip, crunch time and pressure to finish is where errors can slip in, so the more comfortable and well slept he is the better it will be!
You guys got your pennies saved for an FPGA board? I'm close. If Chip gets done early, or in a way that is close to Dave's watch, I'll likely be on the DE2. So, later might be better for me. Honestly, I'm really eager for the jam session to come and will hitchhike to Rocklin, if I have to.
In any case, I don't understand how 'end of spring' could be as late as four weeks from now.. not even here (I'm currently in the north of Norway). Spring finished quite a while ago. In four weeks we celebrate mid summer.
-Tor
IIRC March 21 or there about the first day of spring, which would make June 20 or so the last day of spring. At least that's how I understand our calendar and seasons work.
Comments
I couldn't help but go look up what is current top of the line offerings ... BAE Systems still tops out at that very 400MIPS PowerPC that's in Curiosity. Atmel has a 90MIPS Sparc offering with some PCI+FPGA on board. And there is many slower parts, MIPS R3000 and Pentium1's would head that lot up, but seems to be about it.
It does look like 150nm is as fine as they're going. As we've seen with the Propeller, that limits how much can be packed in. BAE's RAD750 was announced in 2002 I think and, again from Wikipedia, has been operationally deployed from 2005. Not bad, less than three years from the starting blocks.
Ha, I might have the info after all ... 2500 per P2Hot-Cog without mult/div, and 20000 for the whole P2Cool. Of the 20000, 8000 are mult/div. That leaves 12000 for the equivalent of P2Hot variant, so the mult/div part adds a further 67% ... 2500 * 1.67 * 16 = 66800.
20000 when Hub based vs an estimated 66800 if duplicated in the Cogs. I'll count that a solid space saving.
Flip-flop count is probably equivalent - pipelined vs more duplication.
Primary heat issue in the hot chip was way too many things toggling too much of the time. Big busses and a ton of muxes were needed to make it all work out. Chip identified several sources of trouble, and "minimizing the toggling per cycle" was the key to an efficient and capable P2. No reason to believe he's not maximizing that right now.
For the video system, the color model was a killer. All that correction, which we can do in software very reasonably, drove a lot of the heat. VGA support is the minimum needed to get us all the good stuff, and having a lot of COGS and seriously improved HUB throughput gets us the rest when wanted / needed, without the hardware issues. (composite, sprites, component, textures, etc...) Non hardware on those makes sense anyway. People who want 'em can build them easily, and then we've got them.
It's more in line with the Propeller design philosophy anyway. It's gonna be really flexible. Some of those things were kind of baked in, and nobody needs that, and it gets hot. Given what we made P1 do, this is going to be excellent and running much cooler.
Smart pins and the matching up of DACs to COGS got rid of another bunch of units switching, as did changing the HUB memory to the 16x throughput address interleaved model we have now. We are all going to have to give that a go, of course, but I think it will play out just fine. We all will see soon enough! Moving some complex things out to the pins means running cooler, and the HUB issues all discussed seemed to require hot types of constructs too. Plus, we seem to be getting some great measurement capabilities, and who knows what else? The high HUB throughput and Smart Pins are the most exciting parts to me.
I'm confident we won't get a hot chip this time. Chip has the data from the last analysis and is working under much improved design rules and with a team for this iteration. But, we did get a lot of great ideas from the last one, which was a lot of fun to work with. We are going to get the best of the possible, given the power constraints this time around.
Regarding the math, I personally think that is a killer feature, and it's going to be an attractive differentiator. The tasks it's useful for have already been mentioned, and they can be done at good performance, and doing that will be pretty easy. It was kind of stunning to just type in the math needed for fractals, for example, and do it right in PASM. It took me an afternoon, and it was like, "that's it?" The CORDIC bugs program is also small and efficient given what it did.
This kind of thing will make the tech accessible and it will perform on specialized things people often have to do a lot of work on. Of course, PASM is nice, but those nice features will roll right up to C, SPIN, whatever. I think this will get used a whole lot more than we think. Of course, it can all be done in software, as we see on P1, but that has sharp limits, and or really needs a lot of COGS.
One of the design goals Chip has is for the device to be interactive. Gather data via sensor, the DACS, whatever, perform operations on it, transform it, then display it in ways people can find useful. IMHO, that too is going to be fairly attractive given what we've seen done with P1.
May June Su Mo Tu We Th Fr Sa Su Mo Tu We Th Fr Sa -- -- 1 2 3 4 5 6 -- -- -- -- -- -- -- 7 8 9 10 11 12 13 -- -- -- -- -- -- -- 14 15 16 17 18 19 20 17 18 19 20 21 22 23 -- -- -- -- -- -- -- 24 25 26 27 28 29 30 -- -- -- 31I'd agree, and the broad market trend helps the Prop2 here.
There are more and more Linux Modules of the RaspPi / BeagleBone / CHIP being released, and then there are those WiFi portable display screens some call cell-phones.
Often those higher end systems are programmed in scripts, which are easy to craft, but not great for speed.
That leaves room for MCUs that can process sensor information, and do the maths, to have numbers 'script ready' at the other 'user' end.
The P2 will do all that, and allow sensor-add-in and control-add-in, in a way the single-core ARMs can only envy.
Now if it'd just ship :-)
Agreed.
Re: Audio
The thing is going to be a playground for audio and many video applications.
Yes. I suspect a significant portion of Prop 2 demand could come from people using it as an ADC/DAC for another chip. 64-channel ADC's are rare, and the only other one I've seen that came close was a >$50 chip. (anyone want to sell pre-programmed EPROMS to do this?) And having useful DSP capabilities on chip with your ADC is a wonderful bonus.
Marty
Program it in Spin and it would be even more delicious!
Seriously, though, that sounds awesome. Something like it would be perfect for an audio device I want to make. I would think 1/4" or XLR jacks would be way more useful than 1/8", though, especially for professional audio engineers. I've never heard of 1/8" jacks used for anything other than headphones and small speakers. MIDI would also be extremely useful. So would an SD card and maybe even VGA out and keyboard in.
1/8" is the standard for modular synths in the Eurorack format, for CV/Trigger input/outputs.
Imagine what eric could have done with a Prop 2! :
http://www.learn.parallax.com/inspiration/openstomp
http://www.openstomp.com/frontpage/
May June Su Mo Tu We Th Fr Sa Su Mo Tu We Th Fr Sa -- -- 1 2 3 4 5 6 -- -- -- -- -- -- -- 7 8 9 10 11 12 13 -- -- -- -- -- -- -- 14 15 16 17 18 19 20 -- -- -- -- -- -- -- -- -- -- -- -- -- -- 24 25 26 27 28 29 30 -- -- -- 31Parallax should release the new P2 FPGA image on the first day of summer just to spite your impatience.
If the first P2 image is available on the first day of Summer I will be overjoyed. I would feel honored if my reminders caused Parallax to focus on delivering the P2 FPGA image by a certain date.
PUB exciting_world p2_fpga_available := cnt + when_its_ready waitcnt(p2_fpga_available) repeat fun_times++-Phil
Like Dave, I'm keen on giving this new design a go. Fun times are in our future, and it makes perfect sense to remind Parallax that we are here, and ready.
It just wastes my time checking to find nothing
-Phil
Everyone, of course, is completely free to evaluate as they will.
-Tor
You guys got your pennies saved for an FPGA board? I'm close. If Chip gets done early, or in a way that is close to Dave's watch, I'll likely be on the DE2. So, later might be better for me. Honestly, I'm really eager for the jam session to come and will hitchhike to Rocklin, if I have to.
IIRC March 21 or there about the first day of spring, which would make June 20 or so the last day of spring. At least that's how I understand our calendar and seasons work.
They also don't consider 3am as being in the morning, but still middle of the night.
Sounds similar, except one thing. Night is when we are sleeping. Even if you are going to sleep at 6AM after all-night-long party.