(also posted to s.e.d.) We're considering a new product, a spectroscopy controller, that would need a fair hunk of data-moving power and gigabit Ethernet capability to talk to a host system. We would center it on a largish FPGA that would run the actual physics, but we still need supervision, self-test, local maintanance capability (maybe just RS-232) and the Ethernet stuff. Possibilities include... 1. Do everything in the FPGA. Possible but nasty. 2. Use a cpu chip on the main board, next to the FPGA. Both powerQuicc and Blackfin chips have the horsepower and include the gbit PHY. Both run uC Linux. 3. Use an embedded board-level PC, 3a PC/104+. Expensive. 3b Some small-form-factor "regular" pc motherboard. Cheap. both 3's would probably use the PCI bus to talk to the FPGA. Economically, and for speedy development, 3b sounds best. Has anybody done things like this with one of the tiny PC motherboards? Can you recommend a form-factor and a vendor? We'd need gbit Ethernet and a free PCI slot to interface to the process. We could maybe live with 100M Ethernet if we were confident that an upgrade would be available in a year or two. The thing that scares me about using an embedded PC is that the product lifetimes tend to be short. John
embedded PC
Started by ●April 8, 2008
Reply by ●April 8, 20082008-04-08
On Tue, 08 Apr 2008 08:41:53 -0700, John Larkin wrote:> (also posted to s.e.d.) > > > We're considering a new product, a spectroscopy controller, that would > need a fair hunk of data-moving power and gigabit Ethernet capability to > talk to a host system. We would center it on a largish FPGA that would > run the actual physics, but we still need supervision, self-test, local > maintanance capability (maybe just RS-232) and the Ethernet stuff. > Possibilities include... > > 1. Do everything in the FPGA. Possible but nasty. > > 2. Use a cpu chip on the main board, next to the FPGA. Both powerQuicc > and Blackfin chips have the horsepower and include the gbit PHY. Both > run uC Linux. > > 3. Use an embedded board-level PC, > > 3a PC/104+. Expensive. > > 3b Some small-form-factor "regular" pc motherboard. Cheap. > > both 3's would probably use the PCI bus to talk to the FPGA. > > > Economically, and for speedy development, 3b sounds best. Has anybody > done things like this with one of the tiny PC motherboards? Can you > recommend a form-factor and a vendor? We'd need gbit Ethernet and a free > PCI slot to interface to the process. We could maybe live with 100M > Ethernet if we were confident that an upgrade would be available in a > year or two. > > The thing that scares me about using an embedded PC is that the product > lifetimes tend to be short. > > JohnI've said it here other times: if you're building 100/year to build something and you can save $50000 engineering time buying instead of engineering, then $500/ea is a good price. If you have the luxury of having application requirements that are not terrifically real time, if you have a big, square space for your boards, and if engineering time is a big part of the system end cost, then a PC- derived part is the way to go. If you have a long product life, then why not PC-104? If your volumes are low then an inexpensive motherboard could get very expensive the first time you have to swap it out (in a hurry because Purchasing didn't see it coming). -- Tim Wescott Control systems and communications consulting http://www.wescottdesign.com Need to learn how to apply control theory in your embedded system? "Applied Control Theory for Embedded Systems" by Tim Wescott Elsevier/Newnes, http://www.wescottdesign.com/actfes/actfes.html
Reply by ●April 8, 20082008-04-08
On Tue, 08 Apr 2008 10:50:51 -0500, Tim Wescott <tim@seemywebsite.com> wrote:>On Tue, 08 Apr 2008 08:41:53 -0700, John Larkin wrote: > >> (also posted to s.e.d.) >> >> >> We're considering a new product, a spectroscopy controller, that would >> need a fair hunk of data-moving power and gigabit Ethernet capability to >> talk to a host system. We would center it on a largish FPGA that would >> run the actual physics, but we still need supervision, self-test, local >> maintanance capability (maybe just RS-232) and the Ethernet stuff. >> Possibilities include... >> >> 1. Do everything in the FPGA. Possible but nasty. >> >> 2. Use a cpu chip on the main board, next to the FPGA. Both powerQuicc >> and Blackfin chips have the horsepower and include the gbit PHY. Both >> run uC Linux. >> >> 3. Use an embedded board-level PC, >> >> 3a PC/104+. Expensive. >> >> 3b Some small-form-factor "regular" pc motherboard. Cheap. >> >> both 3's would probably use the PCI bus to talk to the FPGA. >> >> >> Economically, and for speedy development, 3b sounds best. Has anybody >> done things like this with one of the tiny PC motherboards? Can you >> recommend a form-factor and a vendor? We'd need gbit Ethernet and a free >> PCI slot to interface to the process. We could maybe live with 100M >> Ethernet if we were confident that an upgrade would be available in a >> year or two. >> >> The thing that scares me about using an embedded PC is that the product >> lifetimes tend to be short. >> >> John > >I've said it here other times: if you're building 100/year to build >something and you can save $50000 engineering time buying instead of >engineering, then $500/ea is a good price.And $79 is a better price!> >If you have the luxury of having application requirements that are not >terrifically real time, if you have a big, square space for your boards, >and if engineering time is a big part of the system end cost, then a PC- >derived part is the way to go. If you have a long product life, then why >not PC-104? If your volumes are low then an inexpensive motherboard >could get very expensive the first time you have to swap it out (in a >hurry because Purchasing didn't see it coming).Do PC104 boards have long lifetimes? The CPU and support chips sure come and go fast. At roughly $100 a pop, we could keep 100 or so motherboards in stock, so at least a discontinuance or change wouldn't sneak up on us unexpected. PC104 has mechanical advantages; it would just plug onto our main application board. But they are expensive. John
Reply by ●April 8, 20082008-04-08
On Apr 8, 12:06 pm, John Larkin <jjlar...@highNOTlandTHIStechnologyPART.com> wrote:> Do PC104 boards have long lifetimes? The CPU and support chips sure > come and go fast.You should talk to the suppliers about EOL plans for the boards in question. My experience has been that these SBCs have very long lifetimes vis-a-vis COTS mobos; typically six to ten years vs 12-18 months max for a COTS board. One of the reasons SBCs are much more expensive than comparable COTS motherboards is because the mfrs will buy and hold stock of the critical components. (Or at least so the Advantech rep told me). Certainly boards I was buying five and six years ago are still available today.> PC104 has mechanical advantages; it would just plug onto our main > application board. But they are expensive.I am 350% in agreement with Tim that this application sounds like a candidate for a PC architecture. It's not absolutely clear to me how the cost-benefit of a COTS mobo vs PC104 will break down for your specific application, though. However I would say that you'll be enduring mobo changes 3~5x more frequently with the COTS path. Whether these ongoing eng costs are worth it to you depends on how difficult it is to qualify a new mobo. My instinct would be to say "buy as much time between profitless sustaining-engineering efforts as possible". I'm sure your product is not something I'm going to see for $39.95 at Wal-Mart.
Reply by ●April 8, 20082008-04-08
John Larkin wrote:> We're considering a new product, a spectroscopy controller, that would > need a fair hunk of data-moving power and gigabit Ethernet capability > to talk to a host system. We would center it on a largish FPGA that > would run the actual physics, but we still need supervision, > self-test, local maintanance capability (maybe just RS-232) and the > Ethernet stuff. Possibilities include... > > 1. Do everything in the FPGA. Possible but nasty. > > 2. Use a cpu chip on the main board, next to the FPGA. Both powerQuicc > and Blackfin chips have the horsepower and include the gbit PHY. Both > run uC Linux.BlackFin has MAC only (needs external PHY), and can run only 100M ethernet. As for uClinux, I stay away from it.> > 3. Use an embedded board-level PC, > > 3a PC/104+. Expensive. > > 3b Some small-form-factor "regular" pc motherboard. Cheap. > > both 3's would probably use the PCI bus to talk to the FPGA.4. Make a PC of your own with all of the necessary stuff on the board. I know some people who make the embedded PCs customised for the particular applications.> > > Economically, and for speedy development, 3b sounds best. Has anybody > done things like this with one of the tiny PC motherboards? Can you > recommend a form-factor and a vendor? We'd need gbit Ethernet and a > free PCI slot to interface to the process. We could maybe live with > 100M Ethernet if we were confident that an upgrade would be available > in a year or two. > > The thing that scares me about using an embedded PC is that the > product lifetimes tend to be short.In the addition to the problems you mentioned, there could be the very unexpected hardware and driver issues when they change the revision of the mother board. Vladimir Vassilevsky DSP and Mixed Signal Design Consultant http://www.abvolt.com
Reply by ●April 8, 20082008-04-08
On Apr 8, 12:46 pm, Vladimir Vassilevsky <antispam_bo...@hotmail.com> wrote:> As for uClinux, I stay away from it.There are of course many factors that go into any choice. One argument for this application that might be in favor of uClinux however is the relative ease of porting the application to a new computing platform should that become necessary. Certainly when you use on-chip peripherals you invest in an architecture, but there's no external vendor-lock in that prevents you from writing new drivers to adapt to the peripherals of a different processor. Leaving the low level drivers aside, you can of course achieve some of this flexibility be writing your application program in a way that is not platform or operating system dependent.
Reply by ●April 8, 20082008-04-08
On Apr 8, 11:41 am, John Larkin <jjlar...@highNOTlandTHIStechnologyPART.com> wrote:> (also posted to s.e.d.) > > We're considering a new product, a spectroscopy controller, that would > need a fair hunk of data-moving power and gigabit Ethernet capability > to talk to a host system. We would center it on a largish FPGA that > would run the actual physics, but we still need supervision, > self-test, local maintanance capability (maybe just RS-232) and the > Ethernet stuff. Possibilities include... > > 1. Do everything in the FPGA. Possible but nasty. > > 2. Use a cpu chip on the main board, next to the FPGA. Both powerQuicc > and Blackfin chips have the horsepower and include the gbit PHY. Both > run uC Linux. > > 3. Use an embedded board-level PC, > > 3a PC/104+. Expensive. > > 3b Some small-form-factor "regular" pc motherboard. Cheap. > > both 3's would probably use the PCI bus to talk to the FPGA. > > Economically, and for speedy development, 3b sounds best. Has anybody > done things like this with one of the tiny PC motherboards? Can you > recommend a form-factor and a vendor? We'd need gbit Ethernet and a > free PCI slot to interface to the process. We could maybe live with > 100M Ethernet if we were confident that an upgrade would be available > in a year or two. > > The thing that scares me about using an embedded PC is that the > product lifetimes tend to be short. > > JohnI have no detailed suggestions, but am bidding on a similar project. Drop me a line and we can compare notes off-line. As things progress maybe we can present our final choices and why. (non-disclosure prevents me from saying more right now.) Ed -- Magic Interface, Ltd. www.magicinterface.com 440-498-3700 Hardware/Software Alchemy
Reply by ●April 9, 20082008-04-09
John Larkin wrote:> (also posted to s.e.d.) > > > We're considering a new product, a spectroscopy controller, that would > need a fair hunk of data-moving power and gigabit Ethernet capability > to talk to a host system. We would center it on a largish FPGA that > would run the actual physics, but we still need supervision, > self-test, local maintanance capability (maybe just RS-232) and the > Ethernet stuff. Possibilities include... > > 1. Do everything in the FPGA. Possible but nasty.Not necessarily bad. Have you looked into using one of the embedded processors? The Altera Nios or Xilinx MicroBlaze "soft" cores can be programmed into general logic. Xilinx also has FPGA's with embedded "hard" PowerPC processors. I've only used the Nios for a small project, but I found it easy and convenient to use (programmed using C code). The soft processors are like micrcontrollers. Great for easy stuff like self-test, etc. They also have a lot of optional peripherals that can be added on. I've also heard good things about the Xilinx PowerPC option from my coworkers if you need to do more serious processing. Derek
Reply by ●April 9, 20082008-04-09
Tim Wescott wrote:> >John Larkin wrote: > >> We're considering a new product, a spectroscopy controller, that would >> need a fair hunk of data-moving power and gigabit Ethernet capability to >> talk to a host system. We would center it on a largish FPGA that would >> run the actual physics, but we still need supervision, self-test, local >> maintanance capability (maybe just RS-232) and the Ethernet stuff. >> Possibilities include... >> >> 1. Do everything in the FPGA. Possible but nasty. >> >> 2. Use a cpu chip on the main board, next to the FPGA. Both powerQuicc >> and Blackfin chips have the horsepower and include the gbit PHY. Both >> run uC Linux. >> >> 3. Use an embedded board-level PC, >> >> 3a PC/104+. Expensive. >> >> 3b Some small-form-factor "regular" pc motherboard. Cheap. >> >> both 3's would probably use the PCI bus to talk to the FPGA. >> >> >> Economically, and for speedy development, 3b sounds best. Has anybody >> done things like this with one of the tiny PC motherboards? Can you >> recommend a form-factor and a vendor? We'd need gbit Ethernet and a free >> PCI slot to interface to the process. We could maybe live with 100M >> Ethernet if we were confident that an upgrade would be available in a >> year or two. >> >> The thing that scares me about using an embedded PC is that the product >> lifetimes tend to be short. > >I've said it here other times: if you're building 100/year to build >something and you can save $50000 engineering time buying instead of >engineering, then $500/ea is a good price. > >If you have the luxury of having application requirements that are not >terrifically real time, if you have a big, square space for your boards, >and if engineering time is a big part of the system end cost, then a PC- >derived part is the way to go. If you have a long product life, then why >not PC-104? If your volumes are low then an inexpensive motherboard >could get very expensive the first time you have to swap it out (in a >hurry because Purchasing didn't see it coming).Short product lifetimes can be compensated for if you know how many units per year will be needed. That $50,000 in saved NRE could buy 500 low-end motherboards and CPUs that you could store until needed. http://www.newegg.com/Product/Product.aspx?Item=N82E16813185113 http://www.newegg.com/Product/Product.aspx?Item=N82E16819103197 For what John Larkin is describing, if there is room a server motherboard might be a good choice -- they tend to stay in production a lot longer and usually have gibabit Ethernet. $99.99 Asus MB, Intel Xeon CPU, Two 1000Mbps Ethernet ports: http://www.newegg.com/Product/Product.aspx?Item=N82E16813131193 $169.99 Asus MB, Intel Xeon CPU, Four 1000Mbps Ethernet ports: http://www.newegg.com/Product/Product.aspx?Item=N82E16813131192 $234.99 Intel MB, Intel Xeon/Core 2 CPU, Two 1000Mbps Ethernet ports: http://www.newegg.com/Product/Product.aspx?Item=N82E16813121329 $459.99 Intel MB, Dual Intel Xeon 2 CPU, Two 1000Mbps Ethernet ports: http://www.newegg.com/Product/Product.aspx?Item=N82E16813121038 If space is tight, the server boards are a bit large, so he might want to go with a Micro-ATX and store enough boards to meet his needs: $49.99 Foxconn MB, Intel Core 2 Duo CPU, 1000Mbps Ethernet port, 9.6" x 8.0": http://www.newegg.com/Product/Product.aspx?Item=N82E16813186139 $69.99 Intel MB, Intel Core 2 Duo CPU, 1000Mbps Ethernet port, 9.6" x 9.6": http://www.newegg.com/Product/Product.aspx?Item=N82E16813121319 -- Guy Macon <http://www.guymacon.com/>
Reply by ●April 9, 20082008-04-09
John Larkin wrote:>Do PC104 boards have long lifetimes? The CPU and support chips sure >come and go fast.I have found that they end to buy large quantities of the chips around the time they are discontinued, and thus are able to keep the same design in production for a long time. -- misc.business.product-dev: a Usenet newsgroup about the Business of Product Development. -- Guy Macon <http://www.guymacon.com/>
