No you cannot change x86 instructions or anything in the Atom core. An FPGA allows you to add soft cores to work along with the Atom. So for example if you wanted a more powerful video decoder, you could design this at the chip level and write it to the FPGA. If the FPGA is big enough, you might be able to add more than one soft core like an encryption engine or a DSP engine. The FPGA can be reprogrammed, so the possibilities are endless.
FPGAs can be used for all sorts of things. Think of it as a blank peice of silicon that you can program whatever hardware you want onto. Need additional USB controllers? Put that in there. Need a second video card output? put that on there? Need a 10MHz reference oscillator to sync a bus with? put that on there. Need to add a CAN bus, I2C bus, or Dallas 1-wire? put that on there. Raid controller? ditto.
Obviously there are trade offs. FPGAs can't hold a huge amount of hardware because they can be programmed with whatever hardware you need at the time. Another problem with them is that they eat power for lunch.
The plus side though is that you're able to bypass the limitation that you have with standard hardware which is what you buy is what you get. With an FPGA, if you have a memory controller for example, you might start out with one that works well, but if the manufacturer finds a bug in the controller that causes it break if it sees a certain pattern in the data, they could issue a new version of the controller that you would apply as a patch. Bug gone.
Things like the intel floating point bug that happened in the Pentium days could be fixed with an update rather than having to replace the chip. (Assuming that the CPU was on an FPGA.)
Personally, I'm really excited about this. I think this is the direction we should be going rather than things like CUDA. Imagine you develop a cool new application that does something amazing, but it takes tons of CPU power and there's no way to optomize it with general purpose instructions found in todays CPUs. Well, you could just write a hardware accelerator for your application and it could interface the CPU and you offload the work to the accelerator. Then when you're done running the program, the OS could unload the hardware and use the FPGA for something else. very cool stuff.
It is very cool indeed. I'm not too sure how much FPGAs have advanced recently. Last I knew they where also very clock speed limited. Although you can build some very useful DSPs that run at only 100mhz or so. Obviously you wouldn't want to build a cpu or anything. I guess some of the newer top end stuff can even hit 600mhz. Interesting I could see a very powerful RAID controller built using this type of approach. That'd be cool, having a full x86 CPU on your RAID controller.
A 100MHz DSP on an FPGA would CRUSH the best CPU money can buy in terms of processing power. CPUscores are not parallelized, so they can only do 1 operation per clock cycle. The amount of work an FPGA does in a single clock cycle is limited by only the size of the FPGA.
Unless Intel puts more useful information out on the subject, and supplies good design software for the FPGA design; I really do not see the appeal for the average Joe tinkerer. But perhaps Intel is targeting established businesses. In which case the average Joe tinkerer is already SoL. Even then, what would a well established company in the embedded market need with a FPGA coupled with an Atom processor ?
Also, if Intel were to match the power/performance ratio of say an ARM SoC. Perhaps there would be more appeal. As it stands, being an x86 based SoC with the ability to run windows is not necessarily a plus. Besides the fact that an ARM platform can run on Windows mobile, when/if needed. Which would be more in tune with this hardware.
So yeah, I do not know. It *is* or would be kind of a cool toy if Intel provided the software for programming the FGPA. For free, or at a reasonable price. We'll see how that works out though.
I would suspect that Intel is trying to capture some marketshare from TI. It seems like these processors would be good for any basic level calculating, from internal car systems, to restaurant tracking and deployment systems.
Yeah I know, and it might even be ok with Windows embedded as a Web KIOSK etc.
FPGA is a design tool, not an end user solution. That coupled with the fact that most embedded systems businesses are experienced enough to not even have to bother with rabid prototyping through FPGA. They already know what they want/need, and how to implement it. *THAT* was my point.
And how you read it, I pretty much read it too ( Intel playing kids games *again*- that is ). Too bad for Intel though that anyone who is really serious about embedded systems already knows the ARM Architecture. Knows how to setup software around the same, and can very likely do a good portion of it for free. Hell, do not get me wrong, I use Windows on my desktops every day, all day, but in the embedded arena my personal belief is that Microsoft has little, or no place. Not unless they change several things. One of the first, and most important in my own personal book would be support for ARM architecture on more than windows mobile. At the very least, every variant of Windows embedded *should* have support. Last I looked though there was nothing.
Anyways, I am not a MS or Intel hater, But sometimes I think both need to pull their heads out and pay attention to what is going on around them. Intel needs to realize that for Atom to be taken serious in the embedded systems arena, they need to at least offer a variant that can compete with ARM where power efficiency is concerned. I would love to have a tiny, fast enough processor that could run on Windows embedded using the .NET framework. One that used a lot less power than the current Atom CPUs . . .
Or maybe Intel would rather that I used a TI ARM CPU, on Debian, using Mono ? Hey Intel here is a clue. How about stop wasting your time on half assed attempts of wagging your weenie at TI, when you can just put your efforts into something that would actually be a good thing. If TI captures some of "your* netbook, or server market. That would be your own fault for not paying attention to whats going on around you in the world.
Embedded developers would buy a solutions together with software libraries/drivers to use whatever function is implemented in the FPGA for whatever shit you need implemented in hardware. Enthusiasts wouldn't be interested in embedded products any way as they are not the target directly. Just as embedded software developers won't start designing hardware HDL/FPGA designs. Plenty of stuff need FPGAs though, everybody can't make their own ASICs. Intel isn't a FPGA company though. And if you want to use it with something else (not Atom) thats fine, it's not for you. It's not an developer platform for prototyping.
Atom was fun when it was introduced. But it's speed hasn't improved, ever. They were to slow when they came to life, they are way to slow now. I wonder if that 2Ghz dual A9 ARM can beat an atom? M.
Atom's problem is that the software that uses it demands too much. Additionally, didn't someone put together a server with like 1000 Atoms that had some decent results, especially given the power consumption.
I'm curious how well Atom performs on Linux machines.
This chip is going to cut my development and production costs to next to nothing. Currently, we use Altera FPGAs to connect high speed broadcast video cables to TI DSPs. We don't use x86 because there is just too much glue logic and the power consumption was too high. This chip will allow us to produce boards 1/15th the size of our current design with 1/10th the parts at 1/20th the price.
Think of it this way... if you wanted to make a monitoring solution for a broadcast video signal which is transmitted over 3G-SDI (a 2.9[0-9]Ghz coaxial medium). Currently, you'd have to design fancy FPGAs, make your own fancy overlay stuff, deal with audio decoding etc... with this chip, you can make a 5 chip solution which could even perform video analysis.
If you wanted to use a broadcast H.264 signal transmitted over ASI (basically a 270Mbit/sec coaxial link) in the past to do the design you needed either a big FPGA for the decoder or an FPGA and an ASIC to do all that. Now, you can do it with a CPU, RAM and a chip from National Semiconductor or Cirrus Logic.
If they were to add 256Megs of on chip RAM and 32/64megs of on chip flash, the entire design could be done in two chips. Cutting the cost of these devices from $1000 to $100.
The trick is, that now we can add all our glue logic for the SDI and ASI chips into the CPU itself. No need to use an all custom solution. The OS isn't that important, can use any real-time OS that has driver support or hardware documentation. This chip is going to revolutionize the entire broadcast video market.
Don't I recall some noise with TSMC over whether they would do funky customized Atoms? Conceptually, are FPGA's used to recapture some of the benefit originally targeted by the TSMC stuff that did not work out? My sense is that Intel knows where they want to go and has a bunch of different initiatives all targeted in the same direction. It is premature to say whether the direction is any good, but they appear to be making a sincere, thoughtful attempt to get there.
We’ve updated our terms. By continuing to use the site and/or by logging into your account, you agree to the Site’s updated Terms of Use and Privacy Policy.
Facebook
Twitter
RSS
26 Comments
Back to Article
npoc - Tuesday, September 14, 2010 - link
This would be a fun processor to have on a beagle board or other embedded dev system *(arduino).A lot more power than those, but the fpga still makes it interesting.
yyrkoon - Wednesday, September 15, 2010 - link
Perhaps I missed something here, but enlighten me please. What would you want an Atom processor on an Arduino system for ?Spazweasel - Thursday, December 2, 2010 - link
Embedded data acquisition and industrial control computers.lexspop - Tuesday, September 14, 2010 - link
SoC = System on a chip?What does PCH stand for?
Antelopite - Tuesday, September 14, 2010 - link
Platform Controller Hub.http://en.wikipedia.org/wiki/Platform_Controller_H...
There's a reference made to this (EG20T) in the text of the article.
Cheers!
lexspop - Tuesday, September 14, 2010 - link
Much appreciated!vol7ron - Tuesday, September 14, 2010 - link
I can honestly say that I don't know too much about the FPGA.I'm curious how one would use it. Is this used to change the instructions that we commonly see with x86 procs?
Trisagion - Tuesday, September 14, 2010 - link
No you cannot change x86 instructions or anything in the Atom core. An FPGA allows you to add soft cores to work along with the Atom. So for example if you wanted a more powerful video decoder, you could design this at the chip level and write it to the FPGA. If the FPGA is big enough, you might be able to add more than one soft core like an encryption engine or a DSP engine. The FPGA can be reprogrammed, so the possibilities are endless.CorvetteZ0606 - Tuesday, September 14, 2010 - link
FPGAs can be used for all sorts of things. Think of it as a blank peice of silicon that you can program whatever hardware you want onto. Need additional USB controllers? Put that in there. Need a second video card output? put that on there? Need a 10MHz reference oscillator to sync a bus with? put that on there. Need to add a CAN bus, I2C bus, or Dallas 1-wire? put that on there. Raid controller? ditto.Obviously there are trade offs. FPGAs can't hold a huge amount of hardware because they can be programmed with whatever hardware you need at the time. Another problem with them is that they eat power for lunch.
The plus side though is that you're able to bypass the limitation that you have with standard hardware which is what you buy is what you get. With an FPGA, if you have a memory controller for example, you might start out with one that works well, but if the manufacturer finds a bug in the controller that causes it break if it sees a certain pattern in the data, they could issue a new version of the controller that you would apply as a patch. Bug gone.
Things like the intel floating point bug that happened in the Pentium days could be fixed with an update rather than having to replace the chip. (Assuming that the CPU was on an FPGA.)
Personally, I'm really excited about this. I think this is the direction we should be going rather than things like CUDA. Imagine you develop a cool new application that does something amazing, but it takes tons of CPU power and there's no way to optomize it with general purpose instructions found in todays CPUs. Well, you could just write a hardware accelerator for your application and it could interface the CPU and you offload the work to the accelerator. Then when you're done running the program, the OS could unload the hardware and use the FPGA for something else. very cool stuff.
Einy0 - Tuesday, September 14, 2010 - link
It is very cool indeed. I'm not too sure how much FPGAs have advanced recently. Last I knew they where also very clock speed limited. Although you can build some very useful DSPs that run at only 100mhz or so. Obviously you wouldn't want to build a cpu or anything. I guess some of the newer top end stuff can even hit 600mhz. Interesting I could see a very powerful RAID controller built using this type of approach. That'd be cool, having a full x86 CPU on your RAID controller.2uantuM - Thursday, September 16, 2010 - link
A 100MHz DSP on an FPGA would CRUSH the best CPU money can buy in terms of processing power. CPUscores are not parallelized, so they can only do 1 operation per clock cycle. The amount of work an FPGA does in a single clock cycle is limited by only the size of the FPGA.It's not comparing apples and oranges at all.
2uantuM - Thursday, September 16, 2010 - link
It's like comparing apples and oranges*Zoomer - Wednesday, September 22, 2010 - link
It all depends on the workload. There are many papers out that that tries to compare them. Try Chimera, Piperench etc.Problem with this: If you think multicore programming was hard, this is a whole new ballgame.
For example, is:
if (a and b)
equal to
if (a) if (b)
(No, it may not)
vol7ron - Tuesday, September 14, 2010 - link
My guess is that there's a little latency hit too, in addition to the power woes.I'm curious how someone might take advantage of something like this in a Virtualization environment eg mutli-core servers, etc.
yyrkoon - Wednesday, September 15, 2010 - link
Unless Intel puts more useful information out on the subject, and supplies good design software for the FPGA design; I really do not see the appeal for the average Joe tinkerer. But perhaps Intel is targeting established businesses. In which case the average Joe tinkerer is already SoL. Even then, what would a well established company in the embedded market need with a FPGA coupled with an Atom processor ?Also, if Intel were to match the power/performance ratio of say an ARM SoC. Perhaps there would be more appeal. As it stands, being an x86 based SoC with the ability to run windows is not necessarily a plus. Besides the fact that an ARM platform can run on Windows mobile, when/if needed. Which would be more in tune with this hardware.
So yeah, I do not know. It *is* or would be kind of a cool toy if Intel provided the software for programming the FGPA. For free, or at a reasonable price. We'll see how that works out though.
vol7ron - Wednesday, September 15, 2010 - link
My guess is that this isn't for the average-joe.I would suspect that Intel is trying to capture some marketshare from TI. It seems like these processors would be good for any basic level calculating, from internal car systems, to restaurant tracking and deployment systems.
yyrkoon - Wednesday, September 15, 2010 - link
Yeah I know, and it might even be ok with Windows embedded as a Web KIOSK etc.FPGA is a design tool, not an end user solution. That coupled with the fact that most embedded systems businesses are experienced enough to not even have to bother with rabid prototyping through FPGA. They already know what they want/need, and how to implement it. *THAT* was my point.
And how you read it, I pretty much read it too ( Intel playing kids games *again*- that is ). Too bad for Intel though that anyone who is really serious about embedded systems already knows the ARM Architecture. Knows how to setup software around the same, and can very likely do a good portion of it for free. Hell, do not get me wrong, I use Windows on my desktops every day, all day, but in the embedded arena my personal belief is that Microsoft has little, or no place. Not unless they change several things. One of the first, and most important in my own personal book would be support for ARM architecture on more than windows mobile. At the very least, every variant of Windows embedded *should* have support. Last I looked though there was nothing.
Anyways, I am not a MS or Intel hater, But sometimes I think both need to pull their heads out and pay attention to what is going on around them. Intel needs to realize that for Atom to be taken serious in the embedded systems arena, they need to at least offer a variant that can compete with ARM where power efficiency is concerned. I would love to have a tiny, fast enough processor that could run on Windows embedded using the .NET framework. One that used a lot less power than the current Atom CPUs . . .
Or maybe Intel would rather that I used a TI ARM CPU, on Debian, using Mono ? Hey Intel here is a clue. How about stop wasting your time on half assed attempts of wagging your weenie at TI, when you can just put your efforts into something that would actually be a good thing. If TI captures some of "your* netbook, or server market. That would be your own fault for not paying attention to whats going on around you in the world.
Penti - Wednesday, September 15, 2010 - link
Altera supplies the tools as it's their product.Embedded developers would buy a solutions together with software libraries/drivers to use whatever function is implemented in the FPGA for whatever shit you need implemented in hardware. Enthusiasts wouldn't be interested in embedded products any way as they are not the target directly. Just as embedded software developers won't start designing hardware HDL/FPGA designs. Plenty of stuff need FPGAs though, everybody can't make their own ASICs. Intel isn't a FPGA company though. And if you want to use it with something else (not Atom) thats fine, it's not for you. It's not an developer platform for prototyping.
piroroadkill - Wednesday, September 15, 2010 - link
I'm so bored of Atom now.mschira - Wednesday, September 15, 2010 - link
Atom was fun when it was introduced. But it's speed hasn't improved, ever.They were to slow when they came to life, they are way to slow now.
I wonder if that 2Ghz dual A9 ARM can beat an atom?
M.
vol7ron - Wednesday, September 15, 2010 - link
Atom's problem is that the software that uses it demands too much. Additionally, didn't someone put together a server with like 1000 Atoms that had some decent results, especially given the power consumption.I'm curious how well Atom performs on Linux machines.
ssj4Gogeta - Wednesday, September 15, 2010 - link
AFAIK they haven't updated the core since it came out. We need to wait for a "tock."submux - Friday, September 17, 2010 - link
This chip is going to cut my development and production costs to next to nothing. Currently, we use Altera FPGAs to connect high speed broadcast video cables to TI DSPs. We don't use x86 because there is just too much glue logic and the power consumption was too high. This chip will allow us to produce boards 1/15th the size of our current design with 1/10th the parts at 1/20th the price.Think of it this way... if you wanted to make a monitoring solution for a broadcast video signal which is transmitted over 3G-SDI (a 2.9[0-9]Ghz coaxial medium). Currently, you'd have to design fancy FPGAs, make your own fancy overlay stuff, deal with audio decoding etc... with this chip, you can make a 5 chip solution which could even perform video analysis.
If you wanted to use a broadcast H.264 signal transmitted over ASI (basically a 270Mbit/sec coaxial link) in the past to do the design you needed either a big FPGA for the decoder or an FPGA and an ASIC to do all that. Now, you can do it with a CPU, RAM and a chip from National Semiconductor or Cirrus Logic.
If they were to add 256Megs of on chip RAM and 32/64megs of on chip flash, the entire design could be done in two chips. Cutting the cost of these devices from $1000 to $100.
The trick is, that now we can add all our glue logic for the SDI and ASI chips into the CPU itself. No need to use an all custom solution. The OS isn't that important, can use any real-time OS that has driver support or hardware documentation. This chip is going to revolutionize the entire broadcast video market.
joker441 - Monday, September 20, 2010 - link
besides this is from xilinx, i'd say this is an even more elegant solution:http://www.linuxfordevices.com/c/a/News/Xilinx-Ext...
but since both of them are not yet available it's hard to say...
dealcorn - Sunday, September 19, 2010 - link
Don't I recall some noise with TSMC over whether they would do funky customized Atoms? Conceptually, are FPGA's used to recapture some of the benefit originally targeted by the TSMC stuff that did not work out? My sense is that Intel knows where they want to go and has a bunch of different initiatives all targeted in the same direction. It is premature to say whether the direction is any good, but they appear to be making a sincere, thoughtful attempt to get there.wazzap123 - Friday, October 7, 2011 - link
You can now run an x86 soft core on low end FPGA.http://www.dailycircuitry.com/2011/10/zet-soft-cor...
Very interesting...