[A pre-apology to Timothy who got this solo - I forgot to CC to ogml.
St00pid me ... anyway, this was meant to arrive much earlier, apologies
if changes have been made in the last 6 hours that void some of this!]
Hey all,
Firstly: Apologies for the huge amount of background info I'm about to
laden you with, I think it's relevant tho', when discussing markets in
research.
Secondly: This is a long message, you might want to get a drink and
possibly make sure you're well fed before you start reading ;)
Thirdly: I'm very excited to see the project going into this
manufacturing stage! From a very personal point of view, I cannot
reasonably afford more than £150 for a piece of hardware ( ~ US$275)
because I don't have and projects that require this addition. The
purchase also seems like it brings the burden of helping the project
beyond just the money, too, which although I agree with, makes me weary
of over-reaching my life. I say this not to change what you have
already done, but knowing that some of your potential market may feel
the same as me and knowing that this information might help you avoid
falling on your face!
With that all said, let me begin ...
I'm a research assistant (means I do project research) at UKC, England,
and will be moving around to another University shortly. Anyway, my
research area, I suppose, is Embedded Systems. My department has just
upped and moved from UKC to Colchester, actually, and they've moved from
a EE department to a CompSci department. Interesting, I feel, because a
lot of university EE/CompSci depts are trying to catch the same students.
Having actually purchased 3 boards from avnet and spoken to other people
in our group (we work with FPGAs *a lot*) I wanted to share with you the
reasons why people choose FPGAs and why you really shouldn't compare the
cost of the with the bits on the board, I believe it's not that simple!
Alright, so, I'm doing a communications project. You heard from me
somewhere way back when about the Xilinx Spartan3, which we'd planned on
building into our board from scratch. We wanted to run at high clock
rates, the Spartan3 would only do 350MHz with the setup we'd established
- we'd borrowed heavily from people at Digilent and their fabulous
US$100 starter board
(http://www.digilentinc.com/Products/Detail.cfm?Nav1=Products&Nav2=Programmable&Prod=S3BOARD)
but in the end there was too much about SI that we didn't know to feel
comfortable making the wretched board and getting it to work.
This board, incidentally, is used in teaching at our university - we
have dozens of them in the lab and assignments and projects are set
based on them. I'd like to convert some of the verilog you use into
VHDL at some point (we only get taught vhdl), just so the option is
there for people to use either.
Anyway, this is where you need to pay special attention, because this is
when we actually needed to find an FPGA dev board to use in our project.
The process went something like:
* What FPGA do we want on the board?
* What tools are available for that FPGA?
* What's the general purposeness of the board (ie: how easy is it to
use for another project after this one is finished)
* How future proofed is the board (kinda linked to the above)
* What kind of support can we get? (for the board AND the FPGA itself)
And all of that lot balances against:
* How much does it cost?
We settled on Virtex-4 boards from Insight Memec, in the end, because
they provided differentially signalling into a daughterboard (we've made
daughterboards to mount our comms components onto) and a nice range of
other stuff for not a lot of money.
Memec have actually sold their Xilinx linecard to avnet, as you've said
provide a lot of FPGA-based boards.
Under EM Home > Products > Design Resource Center
Our boards were the "Memec™ Xilinx® Virtex™-4 LX/SX MB Kit" - we got two
VLX60MBs and one VLX25MB.
US$795 each and US$595 each, respectively. This does not, btw, include
the programming cable. We've got Parallel Cable IVs, which are about
US$70-odd each.
For our purposes they provided (in order of the above):
* A nice big Virtex-4 platform to program into, including all the
bells and whistles a V4 brings with it (the DCMs, DCI, lots of
signalling standards-made-easy, good documentation & plenty of it)
* Xilinx ISE is a nice program to use considering how much of it there
is. It's a nice click 'n' go that a novice can get some instant
gratification, and enough control for experts to really get messy. Of
course, if you don't know what a UCF is, then there nothing's gonna work ;)
* Lots of connectors, extra LVDS connectors, ethernet ... loads of
room for expansion and other "general uses". Our group does a lot of
chip-on-board stuff, so having the ability to pseudo-fabricate ASICs
within FPGAs as a primary test bed is good, then being able to link
those pseudo-ASICs with other stuff, either onboard or off, is very good.
* The Virtex4 is pretty top of the range, we can use it for a goodly
long time yet before we need anything else, and we can always build more
daughter boards.
* By buying a Xilinx FPGA you get a lot of access to their site -
there are a lot of AppNotes, code and white papers that aren't available
without registering your products and logging in. Not to mention the
freely distributed (whopping great) user manuals.
* The board itself, unfortunatly, being a wierd split between memec &
avnet isn't quite as well supported - but again, there is a lot of stuff
available once you register with legacy.memec. Likewise, I also have
the phone number to memec's (now avnet's, I suppose) Xilinx field
engineer - if I get stuck, I'm entitled to help from him.
All of that from one $800 board.
Alright, let me assess your board for some general-use purpose (an
unfair test, I realise, but I want to illustrate my thought process as a
researcher to help you gain market insight):
* Uses a lattice chip: wtf is lattice? I'm very new to the whole
industry, but I was taught how to use Xilinx stuff. I'm not adverse to
trying new stuff, but, to not have crossed it's path at all ... hmm ...
so, off I go to research it. Ok, so, I like what I see. Actually, I
think I'd rather use that than a Xilinx but that's another story.
* Some software. I don't have time to do a proper evaluation at this
stage (heh, rather like I didn't have time to do a major evaluation of
the V4 at the time, either!) but it looks like i can probably work out
how to get it going with minimum of fuss.
* General purposeness: Alright, so it's on a PCI bus. This is good. Is
it reprogrammable via the PCI bus? I remember talk on this subject, but
I don't remember the answer. I've searched the archive I hold locally,
but "pci" and "programmable" doesn't bring up what I wanted - apologies.
It's got a couple of FPGAs on it, so I could potentially invent a new
bus scheme and program the controller, or I could use it in a different
topology if I plug lots of them in together. I wonder what the maximum
speed rating on those PCI pins is ... I've got a colleague working on
parallel/distributed algorithms, using about 4 FPGAs and throwing data
around them. His problem involves moving 32-bits of data in parallel,
and very fast. He's complaining that LVDS might not be fast enough at
644MHz ... I suppose a PCI bus might be good if you could turn bus into
point-to-point, possibly by an intermediate hardware "router" ...
anyway, it's got *some* general use, although with the lack of other
sockets, it's use is limited.
* Future proof: can I get more if we need more? Currently, yes, in the
long run? HMm, maybe, maybe not. That's a concern. If we buy two,
find they're really good but need 16 ... we're 14 short and you've moved
on to OGC. My boss would not be happy purchasing this stuff from anyone
who is not absolutely reputable, which means at least a decent corporate
site, some phone numbers, an address, and shipping details. If you're
planning on shipping world wide - make sure you stamp: "As an importer,
you are responsible for any duty/tax incurred on delivery to your country."
* Support: for the chip itself, the docs look quite nice. For the
board, questionable support, but with the limited applications (unlike
purpose-built general purpose boards) the board is understandably less
complex, so likely requires less support. However, if you *have* got
any user-tweakable components (open sockets for extra oscillators seem
very popular) or interfaced components that need special instructions -
details of how to use them (both from the FPGAs and manually via dip
switches or headers) is essential. Ok, so maybe I'm preaching to the
choir, pushing on an open door, or even just liking the sound of my own
voice ... but the new board ships with a 2x16 LCD display, there are no
instructions on how to use this display ... and since it's only got 8
data in lines I can safely assume it's not as simple as setting each bit
to what I want =)
Xilinx stuff itself can be incredibly overpriced, but the market they're
targetting is big enough to support that kind of money. Unless your
board does a lot of stuff, you can't afford to ramp up the cost to
compare to them. Your board has, understandably, a small feature set
and is not useful in different ways to different people.
In my opinion you cannot compete with these people and even thinking of
comparing your product with specially designed general-purpose boards is
folly of the highest order. You need to aim yourself at the home
hobbiest / specialist research groups, because that's how you've started
this endevour - with that in mind, US$1000 is a lot of cash for a
hobbiest. If you try and aim for general research or education, you
will fall flat on your face - your product does not provide what they
need. The home hobbiest, however, would love nothing more than an FPGA
in a PCI slot ... if for no other reason that *that is cool!*.
If you can make testing and tinkering risk free (having an easy way to
flash the chips back to their factory shipped state ... ie: a working
PCI controller!) then suddenly these things are more accessable. I have
a GP32 ... I'm scared to flash the bios too often, 'cause I'm gonna make
a mistake. So I spend a lot of time with an emulator. I don't need the
GP32 hardware to use an emulator. The same with this. If people do
nothing but simulate, they don't need your hardware to synthesise into.
Almost there, folks, another paragraph or two and you can get some rest ;)
None of what I've said is intended as a criticism on the path you've
taken or the device you've designed, I would've done exactly the same
thing, I merely wish to offer up my experiences so you can freely take
advantage of them. Everything that's set up on the hardware is that way
for a damn good reason, I ask you to just think about what you are
trying to accomplish and how you are intending to get there. This OGD
platform is not a sideline to produce capital (otherwise you may have
taken a different path - perhaps a more generic PCI-pluggable FPGA dev
board), it is a test bed to provide you with the means to take the board
to the next level: ASIC.
Finally, let me say that much of this is rhetorical, I neither expect
nor desire responses - you have better things to be doing!
Good luck! I for one am still watching with interest in what you're doing!
- Piete.
Timothy Miller wrote:
Howard found an FPGA-based graphics kit board on Xilinx's web site.
It's based on a Spartan 2, not much memory, etc. It goes for $1500.
I think it's on the avnet site below.
Avnet (avnet.com) makes a bunch of FPGA-based boards. If this URL
doesn't work for you, this is under EM Home > Products > Design
Resource Center.
http://www.em.avnet.com/evk/home/0,1719,RID%253D0%2526CID%253D0%2526CCD%253DUSA%2526SID%253D4742%2526DID%253DDF2%2526LID%253D18806%2526PVW%253D%2526BID%253DDF2%2526CTP%253DEVK,00.html
From this page, I've found some interesting FPGA-based products. One
has a small Xilinx and an ACD for over $5000. Another is a
Xilinx-based PCI-X board for $1200, but the PCI-X IP core sells for
$16000. Isn't that nice. There may be other sections of their site
that have other similar products.
Anyhow, If anyone's interested, I suggest looking through this site to
make comparisons with what's out there. Keep in mind that the FPGA
we're using is many times the size of most of the ones on there, not
to mention way more memory, etc.
I was worried that charging $1000 might be too much. Now I'm starting
to worry that it's not enough. If another board with 1/8 the capacity
of everything sells for $1500, what should we charge for ours?
_______________________________________________
Open-graphics mailing list
[email protected]
http://lists.duskglow.com/mailman/listinfo/open-graphics
List service provided by Duskglow Consulting, LLC (www.duskglow.com)
.
_______________________________________________
Open-graphics mailing list
[email protected]
http://lists.duskglow.com/mailman/listinfo/open-graphics
List service provided by Duskglow Consulting, LLC (www.duskglow.com)
