why are PCI-based FPGA cards so expensive ?

These peope think there is ...

and these people seem to think that's just too much money.

- Brian

This web site is your friend:

When I was researching kits for myself I discovered that either the kits didn't have all the hardware I was looking for, had a lot of hardware I wasn't looking for or were way out of my price range. The conclusion I came to for why a lot of them are so expensive was the size and speed grade of the device the board was centered around and licensing for development software and PCI cores.

One of the kits I gave some serious consideration to but decided they were still outside my price range was Alpha Data

The have a selection of card offerings and a lot of the software is available from their ftp server. You could start with an unpopulated card and add modules later. I even found a company with a 3D graphics accelerator that I think could have fitted one of the module's sites.

If you take a look at the kits that connect through the parallel, serial or USB ports tend to be the bottom of the line as far as speed grades (well maybe not all of them). But what I finally ended up settling on was one of the Altera's NIOS II development kits. For a limited time they are offering it with Stratix II EP2S60ES (engineering sample). After the initial offering I guess they are only going to sell it with the EP2S30 (production device).

I also have an interest in high performance, parallel processing and supercomputer (like) applications. In the corner of one of the rooms in my apartment sits a couple of cabinets with ~160 T800 transputer processors. One of the projects I have in the back of my mind is using NIOS II processor cores design something similar.

While at one of the Altera sponsored events in Ma. (I think it was SOPC World) one of the other attendees asked how many processors could fit on a device. The rep couldn't provide an answer but it was one of the first things I tried with my kit. I found I could add 16 fast NIOS II processors with 9k of CACHE and a SDRAM controller and still have about

25 % of the device left for other peripherals. While I realize it isn't anywhere as fast as PCI, I could have easily added an Ethernet core to communicate to a host.

Most of the kits offered are meant to just get you started. After you get started you really are kind of expected to started developing you own platform or boards.

Derek

Writing a PCI driver for the various Windows PC platforms is not that trivial. Then come the powerusers that wish changes or don't the stuff to run because they understand it differently. Meaning such an undertaking requires a hotline with seasoned staff to answer questions that the users are able to get it going. Who is going to pay for this support ? Beside that the powerful chips, meaning the big and fast ones, are sometimes beyond the 500$ mark for the chip alone, no printed circuit board yet.

Rene

--
Ing.Buero R.Tschaggelar - http://www.ibrtses.com
& commercial newsgroups - http://www.talkto.net

Please include the full URL even if you do provide a shortcut.

The shortcut company may go out of business or decide to blast my screen with ads or ... Or I may recognize the URL or host as a place I do/don't want to visit. (Some require registration or cookies. Some provide good info.)

--
The suespammers.org mail server is located in California.  So are all my
other mailboxes.  Please do not send unsolicited bulk e-mail or unsolicited
commercial e-mail to my suespammers.org address or any of my other addresses.
These are my opinions, not necessarily my employer's.  I hate spam.

That is being generous. Actually, writing a Windows PCI driver is a royal pain if you want any kind of high performance bus mastering et al. I can't in fact imagine being able to get away with giving a customer any kind of Windows driver for an FPGA.

The alternative is to put an extra chip, a PCI to local bridge, that *can* have a common Windows driver, and then write that Windows driver and ship it with the board. I can tell you from experience that once that Windows driver is written, no vendor is going to feel magnanamous about giving it away. It's painful and expensive to develop, and surely comes with plenty of support headaches:-(

--
Steve Williams                "The woods are lovely, dark and deep.
steve at icarus.com           But I have promises to keep,
http://www.icarus.com         and lines to code before I sleep,
http://www.picturel.com       And lines to code before I sleep."

Another option would be to write a PCI driver for Linux - there's lot of source code for drivers that already exist, for a start. It all depends on who you want using it :) I know this was/is the approach taken by WAND

when they were originally doing research on network measurement gear.

Does anybody know how the relative difficulty of the two approaches differs? I would have suspected that linux would be slightly easier due to more examples/code available, better documentation, and more open-source people to discuss the code with.

Jeremy

That is what I assumed. To deliver a prebuilt FPGA functionality for the PCI to be added to the FPGA was another thinkable solution.

In the last couple years I tended to get materials from a company called OSR which appear to be only writing drivers and telling people how to write a driver. The one week courses for a few k$ are considered a bargain when compared to the effort required to get the information together yourself.

Anyway writing drivers appeared to me as specialty job that only bigger companies can afford. Meaning whoever does write the driver has to be writing drivers the whole year round, otherwise the investment into the knwoledge and keeping it up to date is hard to justify. Then comes this chipset with this quirk and, and, and, ...

Yes, and then comes the 96 channel 1GHz logic analyzer to have a look at the 64bit PCI bus. Setting this one up and working with it efficiently is another half employee.

Then divide the effort by the number of boards sold.

I too believe there'd be a market in supercomputing to the desktop. But first some data would have to be released. How many floating point units can be integrated in parallel running at what speed. And then the software manufacturers writing the powerhungry software should have a standardized interface, such that the boards are multi-sourceable. The competition is using clusters of PCs or clusters of other machines. The hardware is fairly cheap and softwarewise not that difficult to distribute. Compare this to a to be written compiler to let the software run partly on an FPGA.

Rene

--
Ing.Buero R.Tschaggelar - http://www.ibrtses.com
& commercial newsgroups - http://www.talkto.net

Would a low-cost USB-based board work? Transfer rates aren't as high as PCI, but a fat lot more than parallel.

Jake

Michel Billaud wrote:

Linux drivers are tons easier. Day Job writes Linux drivers for our boards *first*, so that we can debug the boards if we need to, and only after the Linux driver and hardware is stable do we even *start* writing the Windows driver.

This approach really does save us a great deal of pain and suffering.

--
Steve Williams                "The woods are lovely, dark and deep.
steve at icarus.com           But I have promises to keep,
http://www.icarus.com         and lines to code before I sleep,
http://www.picturel.com       And lines to code before I sleep."