Want to read Slashdot from your mobile device? Point it at m.slashdot.org and keep reading!

 



Forgot your password?
typodupeerror
×

Other Uses for an AGP Slot? 160

SleepyHappyDoc asks: "AGP seems to be going the way of the dinosaur, but there's still a lot of slots on legacy motherboards out there. If you don't have need for the graphical advantages of AGP (say, on a headless server), what else could you use the AGP slot for? Could the advantages of AGP over PCI be leveraged in a use other than graphics cards?"
This discussion has been archived. No new comments can be posted.

Other Uses for an AGP Slot?

Comments Filter:
  • by FatSean ( 18753 ) on Sunday February 19, 2006 @02:22PM (#14755338) Homepage Journal
    I would think that perhaps you could use the bus bandwidth and an old/slow card to do additional computation. Leverage the GPUs in the more recent AGP 3D offerings and use it for something...uh....usefull :)

    Perhaps we can user in a new age of game design where you can load your machine up with older cards to assist with the heavy 3D math for a game, or maybe expose those cards as a virtual machine of some sort.
    • by Anonymous Brave Guy ( 457657 ) on Sunday February 19, 2006 @03:08PM (#14755601)

      Only in IT could something that was state-of-the-art five years ago and a clear industry standard even a couple of years ago possibly be described as "vintage" today. :-)

      • Only in IT could something that was state-of-the-art five years ago and a clear industry standard even a couple of years ago possibly be described as "vintage" today.

        Indeed. But as long as enough users buy the new "standards", the industry has zero interest in defining something that lasts.
      • Only in IT could something that was state-of-the-art five years ago and a clear industry standard even a couple of years ago possibly be described as "vintage" today. :-)

        Ummm..fashion industry and wines come to mind. And that car you drive? that was so last century. Medical prosthetics, pharmacology, particle accelerators...may be a few more out there. Granted, they wouldn't have advanced without the rather astounding advances in IT over the last few, but IT isn't the only porpoise in the bow wake.

    • You can sort-of do that, but there's a problem. AGP was designed for fast burst transfers in one direction. You can shove data to the card at a high speed, but you have to switch modes to bring the data back. For video, that's not a huge problem, since most of them time, you're writing to video RAM, not reading from it. However, for outboard processing using the GPU, it impacts performance pretty significantly. That's why companies like Apple are dumping AGP pretty swiftly in favor of PCI Express (PCIe

    • Leverage the GPUs in the more recent AGP 3D offerings and use it for something...uh....usefull :)

      ATi has 24-bit floating point calculations [i.e. 8 bits less than "single precision"]; nVidia & even the new IBM/Sony Playstation Cell processors have only 32-bit floating point calculations [i.e. "single precision"].

      Single precision floats are utterly worthless for real-world ["usefull"] calculations; they even lose their integer granularity at 2 ^ 24:

      16777216 + 0 = 16777216
      16777216 + 1 = 16777216

  • by Idimmu Xul ( 204345 ) on Sunday February 19, 2006 @02:22PM (#14755342) Homepage Journal
    I'm going to have to go with none and move along.
  • No. (Score:2, Interesting)

    by Anonymous Coward
    AGP's architecture makes it unsuited for bi-directional communication. For what it would cost to fabricate an AGP card you could buy a PCI-Express mobo+card.
    • Re:No. (Score:4, Interesting)

      by SharpFang ( 651121 ) on Sunday February 19, 2006 @02:47PM (#14755456) Homepage Journal
      bidi is not everything. If you have a 33k modem connection to a 256-node beowulf cluster, do you claim it's useless? AGP cards have pretty beefy serial processing chips, that can be programmed with any, generic tasks just like CPUs, and for some of these tasks they will suck a big time (but still work) and for some they will rule (stuff like lots of similar rather simple calculations on lots and lots of data - they are unbeatable.) Statistics, rendering, filtering, encoding/decoding, all such stuff is really fast. Now the downstream is pretty slow so it hurts that -very- simple calculations can't be done en masse (the GPU can do them great but they get stuck at sending them back to the PC), and hard calculations with lots of decision-making are better handled in the CPU but there is a class of tasks where the GPU is unbeatable.
      • something to process boinc packets, and only send back the result?
      • bidi is not everything. If you have a 33k modem connection to a 256-node beowulf cluster, do you claim it's useless?

        No, but it will be far less useful than the same cluster with several gigabit or faster connections, which would be far more appropriate for such a cluster.

        In any event, your motherboard that has an unused (?) AGP port probably also has PCI ports. Since the only AGP cards that I've ever heard of have been graphics cards, and you need a fast connection to something, I'd suggest just us

        • If you want to use the AGP slot, you'll have to 1) design and build your card yourself from scratch, and 2) apparantly it'll be severely limited in the data rate back from the card to the computer.
          Alternatively take a good 3D-accelerated gfx card off the shelf, never plug anything in the card's outputs and write your own custom software to use the card's GPU. It will be severely limited it data rate back, but that's not a show-stopper. There are tasks where this is pretty useful - hashing algorithms return
    • There are test and measurement applications that require only a uni-directional bus. For example, high-speed digital or analog waveform generation, where the pattern is either does not repeat or depends on other inputs (so that the pattern must stream from the PC, rather than be stored on the generator itself).

      With a faster, dedicated bus like AGP, products like this would allow for faster data rates, with the speed not throttled by other devices or instruments in the same PC.

      However, these exact same bene
  • Well (Score:3, Informative)

    by Eightyford ( 893696 ) on Sunday February 19, 2006 @02:24PM (#14755350) Homepage
    You could always make try to hack your own peripheral. [hardwaresecrets.com]
    • Hacking together a compliant PCI card is a challenge, although if you ignore the PnP stuff and BIOS registration, it's doable for the home hacker.

      As I understand it, the AGP spec would be much harder to do at home. If anyone knows of anybody with a homebrew AGP design, I would love a link.

  • More PCI-E cards (Score:1, Insightful)

    by Ramble ( 940291 )
    Nevemind AGP, give me a maths co-pocessor that goes right into into my spare PCI-E x16 slot.

    I want those floatig point numbers faster, damnit.

  • by BobPaul ( 710574 ) * on Sunday February 19, 2006 @02:26PM (#14755358) Journal
    AGP has more downstream bandwidth to the slot than upstream bandwidth from the slot, whereas PCI and PCIe have the same to and from the slot.

    You could use it for something like a beefy sound board.. or, something...

    No, not much other than graphics output really needs that kind of bandwidth differential.
    • by name773 ( 696972 ) on Sunday February 19, 2006 @03:16PM (#14755662)
      maybe for running hashes on something... a hash is usually smaller than the data you used to get it, and it does take some processing
    • XML Parsing?

      Bulky strings in, nice, lean SAX events out.

      • You are kidding, right? The SAX events are the same data as the XML stuff going in, except they have all the event overhead. A SAX stream is almost certainly going to be bigger then the original XML string in terms of bytes consumed.
    • Hmm. Wonder if it could be used for some kind of custom backup solution, to an external real-freaking-fast drive of some kind, for automatic backups of your server. That would push out a lot of data with little need for backtalk besides "More data please".
      • I considered that, but I have trouble believing such a solution would be any better than existing PCI-X or PCIe solutions. If someone has that kind of bandwidth on their disk array to saturate the downstream on an AGP port, they are most definately using a controller in a PCIe or PCI-X 64bit slot.

        But I suppose if all of your PCIe or PCI-X slots are already in use by RAID controllers, you could make use of the AGP. Although, if that's the case you definately have some $1000+ in RAID controllers alone, plus a
  • That would let you use the GPU of a video card to do other kind of computational tasks.

    I can't remember the name, it was posted in /. about a year ago, maybe someone isn't as lazy as me in a sunday afternoon and will care to looj for it.

    Anyway, AGP is really too 3d graphics specific as to use it for something else. It's designed to let the machine pass enourmous ammounts of information in only one direction.

    Maybe back in 98 one would try to reuse old hardware to it's last breath, now, the prices of hardware
  • Co-CPU. (Score:5, Interesting)

    by SharpFang ( 651121 ) on Sunday February 19, 2006 @02:35PM (#14755389) Homepage Journal
    I don't know of any non-gfx cards that would use the CPU but there was a C compiler released that would use the GPU instead of CPU for your generic computations (instead of 3d gfx) and for certain kinds of calculations/programs it would be equivalent of 10GHZ P4 class CPU in the means of speed. Look up archives of Slashdot for it.
    • i thoughed this couldn't be done properly because conditional statements are not supported by gfx cards
      because they lack branch prediction skillz

      i read it somewhere dont remember where so i dont know if this info is valid
      • They do, but the support sucks. (I'm not quite sure how, likely in pre/postprocessing, so you're allowed for one if() in, say, 400 instructions, at fixed places, deciding to pass the result back to the same pipeline again, forward it to other pipeline (out of 8?) or pass it to the output. Want more than 8 branches of program? Pause, reprogram GPU with new 8 pipelines, unpause. Repeat if you want the old 8 back.)
  • VRAM Storage Device (Score:5, Interesting)

    by dastrike ( 458983 ) on Sunday February 19, 2006 @02:36PM (#14755393) Homepage

    Well, this still involves to use a graphics card, but in a bit different way.

    1. Acquire a cheapo graphics card with lots of memory, e.g some low-end NVIDIA or ATI with 256 megs
    2. Read and apply VRAM Storage Device - How to use the memory on GFX board in a different way... [linuxnews.pl]
    3. You have a bunch of memory that can be used for a ramdisk type of device or swap space

    YMMV with the performance though.

    • Or you could just put another gig of ram in your PC. Odds are low that a slow, awkward 256meg of RAM will be useful for anything. Maybe a ramdisk, but even then it's tiny.
      • by SmallFurryCreature ( 593017 ) on Sunday February 19, 2006 @03:26PM (#14755718) Journal
        I got an old dual p3 wich is limited to 768mb. Anymore and it won't even boot.

        So how would propose I add another gig when it cannot even accept a single gig?

        It does however have a 32mb graphics card that is not used. Oh sure it is a tiny amount of memory but when the kernel is forced to start swapping it makes a difference. Not a huge amount to be sure and it doesn't help at all when it really needs to swap a lot but it gives me just a little bit more room to play with.

        Haven't thought about upgrading the card but I guess if I ever see a really cheap 256mb card it might be worth it.

        A dual P3 is still plenty fast for desktop use especially since the linux kernel keeps on improving. Windows users may wish to close their ears to save themselve from terminal shock but linux installs get better with age.

        Sure sure someday I am going to have to buy a new system and now that dual core chips are here the hurdle is not as big as having to buy a dual single core machine was but still, the longer I can keep this system running the happier I am

        Hardware/software hacking is about making stuff go that extra mile. Just plonking a wad of cash on the counter is totally missing the point.

        • So how would propose I add another gig when it cannot even accept a single gig?

          well, you could break out the solder, and a bread board, and pick up a memeory controller that can work across a pci bridge, preferably as an ide/scsi controller, and of course a memory socket or two (depepnding on the memory controller you picked out) and then , on boot up initialize that ram as a swap drive, using something like norton ghost, or dd ;) and viola a gig of swap, with all the performance of 'real' ram. sure findin
    • I've always loved that hack.
    • Dude, look at the prices. You can buy a gig of PC3200 for like $50 these days. It's incredibly cheap.
      • True, it wouldn't be much point purchasing a graphics card to use for this kind of purpose, but if one has already a graphics card left over, it could be put into use in this way.

        And a gig for $50 USD? I'd love to see such low prices over here in the People's Republic of Sweden... Can't be had for lower than $101 USD currently, after searching for the lowest prices. Sigh.

    • Looks to me like they've just reinvented the RAMcards we used to expand available memory space in XTs, 286s, and the occasional low-end 386.

      In fact my 286 has one that was used as a RAMdisk, effectively doubling performance.

      Of course, in that era we were talking adding between 2mb and 8mb, and being happy to have it!

    • Yup, YM definitely varies. I did this with a video card in a server, and did a (IIRC) hdparm -tT to test the speed of the device. Turns out it was *way* slower than the harddrive.
  • Not a lot (Score:4, Interesting)

    by Kijori ( 897770 ) <ward,jake&gmail,com> on Sunday February 19, 2006 @02:42PM (#14755426)

    AGP is a one-way architecture - the motherboard sends data to the graphics card, the graphics card processes it and sends it to the monitor. The limitations of this way of working are why dual graphics card solutions were never practical on AGP once you started increasing the complexity of the data - the bus wasn't capable enough.

    That said, it's not impossible to get it working. You just need to get around the one-way bus problem. There are two obvious solutions for this, to my mind: (ignoring the fact that no cards exist to do it for you)

    Use it for one way data
    You create a card that acts only to process and send away data. At its simplest, this might be an audio card (without line-in, obviously). Getting slightly more creative, the card could take the 'load' of preparing documents and printing them off the CPU, although I can't see this being useful. Using a rather crossfire-like setup, you could send the output of a suitable graphics card into an input on another, and use it as a pre-processor; at its most basic this could be used to divide a signal in half to be processed by two (or more) cards, or getting more complex it could render something simple - perhaps hidden windows, for use in transparency effects, or perhaps acting as a 2D processor and leaving 3D work to the 'bigger' card - tag this as 'rendered' and send the output to its big brother.To be honest though, this seems a little ridiculous.

    Creating a feedback path for 2-way data
    This, in my opinion, is where it could be useful. The moment you add a way to send data back - at its simplest, I suppose this would be a SATA or IDE cable and suitable software that continuously reads the contents of the 'hard disk' - you have an opportunity for a specialised processor. The hack would be incredible difficult, granted, but the processor on a graphics card would seem to be well suited to encode video. You send your stream to the AGP card, it converts it to mpeg4 (for example) and sends it back via SATA, taking 99% of the load off the processor. (These cards have recently started to appear for PCIe, so the is definitely a market). With some sort of feedback path, the card could do anything a PCI card can do, but substantially faster thanks to AGP's higher bandwidth - the trick is getting a decent feedback loop.

    After all that, though, I think the practical answer is no, there is no use for an AGP slot other than graphics; there is no demand for other cards, so they just don't exist.

    • I suppose you could have some sort of device which acts like a RIP and drives a big printer. Something involving lots of page generation. Something like a graphics card ;-)
      • That's the problem - there are lots of applications, but only as long as what you want is effectively a graphics card. AGP is just too specialised to allow anything massively different without changing the fundamentals of how the interface works.
    • Re:Not a lot (Score:5, Interesting)

      by DRJlaw ( 946416 ) on Sunday February 19, 2006 @03:39PM (#14755782)

      AGP is a one-way architecture - the motherboard sends data to the graphics card, the graphics card processes it and sends it to the monitor. The limitations of this way of working are why dual graphics card solutions were never practical on AGP once you started increasing the complexity of the data - the bus wasn't capable enough.

      No.

      AGP is a two-way, point to point architecture that has a single master and a single target. Data can be written to and read from the graphics card memory, but you can't exercise the full range of PCI I/O operations. The data transfer rates are asymmetric, with sending data to the card greatly favored over reading data from the card, but they are most certainly two-way.

      The SLI argument is a lesser error, if you would even call it that. You could have, but never as far as I know actually did have two AGP busses in a system. Thus I suspect that it would have been possible to do SLI with AGP, especially when you consider that existing implementations of SLI require an additional card-to-card link, which means (likely, this last part is speculation) that there is very little return data being transmitted from the cards back to the PCI express switch beyond that which you would see in a single card system, whether it is PCI express based or AGP based.

  • It was designed to only ever be used for graphics, and it cannot be used for anything else. True, VLB may have been only for graphics, and yet I've seen the occassionaly scsi card or ethernet nic for it, but it was still more general-purpose. Read up on the AGP docs, and they make it clear... there can be no other uses.
    • What about that ATI All-In-Wonder Radeon 8500 DV I had? That not only managed to send uncompressed Video over the AGP bus to the CPU, but also a Firewire port!
    • actually, VLB was designed as a higher-speed replacement for ISA, and had a definite speed advantage over the PCI that was just becoming available at the time. PCI ended up surviving a heck of a lot longer, though.

      I had a motherboard with 3 VLB slots and 3 standard ISA slots.
  • Leverage (Score:3, Insightful)

    by _Splat ( 22170 ) on Sunday February 19, 2006 @03:12PM (#14755632)
    Leverage is not a verb. Please stop using it as such. See the article posted today about loss of literacy.
    • Re:Leverage (Score:2, Informative)

      by Anonymous Coward
      Hey genius, apparently you have neither encountered "leverage" as a verb nor taken the trouble to make sure you're right, because leverage can indeed be used as a verb.
      http://www.thefreedictionary.com/dict.asp?Word=lev erage [thefreedictionary.com]
      • First of all, I actually checked my Webster's Third New International Dictionary (Unabridged). I admit it does date from 1981.

        Second, if you use the definition you linked, the posting doesn't make sense. Try it out.
    • Any word can be a verb if a group of people use it as a word. The fact that a huge number of people have a good idea of what "levereage (v.)" means, I think it is fairly safe to say that it is a verb now.

      This is how languages evolve--take a social linguistics class. Or, at least leverage the knowledge contained in this Wikipedia article:
      http://en.wikipedia.org/wiki/Prescriptive_linguist ics [wikipedia.org]
    • At least it's been around since the 1930's. What gets me are the Olympic announcers using 'podium' as a verb. No really. "The fact that he podiumed is an amazing indication of how far he's come in the last 4 years."
    • Language changes to suit a society's needs and values. Fixed definitions and spellings of words is a recent (150 years) development. Verbing nouns is popular and accepted tactic by western society. All your protestations to the contrary will not modify this behaviour.
    • Trolling is a verb, and it means "To fish for by trailing a baited line behind a slowly moving boat".

      Since I see nothing resembling a boat in this forum, please do not claim this poster is "trolling". _Splat is clearly concerned about the daft dialect epitomized by the Slashdot front page. If we are to be taken seriously as the technocratic elite, we must not expose our banality in such a manner.

      Following in this spirit of progress, I implore the Slashdot editors to take down that representation of Mr. Bi
  • AGP might be a one-way bus, but if you're planning to run a headless server like GP would suggest – assuming, of course, that most of what it does is sends rather than receives packets, what about having an AGP-based network card that could handle sending more bits at once? And since it's mostly a one-way, maybe a cheaper PCI card exclusively for receiving?

    The only problems with that as far as I can see is that no such card exists, and that unless you have a really wicked high-speed connection (OC-6
    • Wouldn't work. AGP can send a lot faster than PCI can receive, so your PCI reception speed is the bottleneck for that network. The PCI reception speed is equal to the PCI transmission speed. So it makes no sense to use anything that can send faster than PCI if the receiving end is also PCI.

      However, if you need to stream a lot of different data to a couple of separate machines, you might be able to implement a multihead network card in AGP, and split the superior bandwidth over multiple physically separate

  • /dev/null (Score:5, Funny)

    by Anne Thwacks ( 531696 ) on Sunday February 19, 2006 @03:13PM (#14755640)
    As it is write only, it is ideal for implementing a hardware /dev/null on Unix systems.
    • On further investigation, this is not true ... the GPU can do graphics operations on host memory, which necessarily requires writing to it (and AND, OR and XOR into it). Although access is limited to 32 bit quantities, it would be quite feasible to define a command address into which you write the address of a parameter block containing instructions. These could operate like scsi commands, which would be appropriate for block (vector) operations. One location in the block contains the command status, whihc
  • Video compressor (Score:2, Interesting)

    by Anonymous Coward
    Many seem to be saying that the AGP has a lot of bandwidth going TO the card but not coming back from the card. This seems well suited to feed uncompressed video to a card in the AGP slot, have the AGP card compress the video into whatever format you wish, and then send back the compressed data.
  • You could make a lovely planter out of it. The two tiered pins would make for an exceptionally easy 'layered' approach, maybe some lucious greens in back, and shorter flowering bits in front.

    You would simply be the envy of all the other sysadmins in the data center.

          -Charlie

    P.S. Paint your boxes in pastels to compliment the florals. Black and silver is so 2004.
    • > The two tiered pins would make for an exceptionally easy 'layered' approach

      This definitely calls for some shrubbery.
  • Headless, then... (Score:2, Interesting)

    by chivo243 ( 808298 )
    Gotta have'm, 90% of our servers are now running headless(yikes) where have all the monitors gone? As for all the other slots, I guess it was poor planning from the beginning. But if you look at the market as being constantly in the state of BETA! then it all makes fucking sense.... just my two euro cents. Wait a damn, minute, as long as I have been drinking, and can type..Does the fact that MS has stopped support for some OS's now and others soon, that they have finally found all the bugs, and do not need
  • by xeeazgk ( 850506 ) on Sunday February 19, 2006 @10:06PM (#14758098)
    I don't think you guys understand the kind of massive speed differential we're talking about. I don't remember the numbers, but it's like G/s to the card and K/s back. It's just enough to tell the processor that the card is ready for the next rendering task, nothing more.

    Someone mentioned doing video compression... because you could send the compressed file back. Well ok, except, A. video cards only have 256mb of ram... so your uncompressed video would only be like what 30 seconds? B. getting the data back to the hard drive would be like transfering files over a serial cable... like old PS/2 serial, not USB2 serial.

    Now... a card with a SATA out would work. That's the kind of bandwidth that would help, although for most applications just an IDE out would do the trick.

    But these cards don't exist. So no... nothing to be done with agp slots.
    • Sorry, not quite true. AGP is a 100% spec PCI 2.1 interface, just pumped up in speed, then with a couple of nice add-ons added. (Direct Memory Execute, for one.) On some server motherboards, you will see the AGP slot replaced with one or more higher-speed PCI slots. (I haven't seen it too much recently, but Intel used to sell the L440GX board that had two 66 MHz PCI slots that were run off the AGP controller, in addition to the 33 MHz PCI slots off the Southbridge. But this was back in the Pentium II d
    • AFAIK, MPEG2 (DVD) is constructed of 16-frame blocks, so 30 seconds of source material would be far more than you'd need.

      All you need to do is keep the source material streaming into the graphics card's memory.

      256MB would probably be more than sufficient for MPEG4 encoding too.

      I think the main problem would be how to fit the the encoder into the old shader v1 GPU's we're talking about here.
    • ...for video compression, e.g. the PCI bus would not be the bottleneck, the encoding would be.
    • Someone mentioned doing video compression... because you could send the compressed file back. Well ok, except, A. video cards only have 256mb of ram... so your uncompressed video would only be like what 30 seconds? B. getting the data back to the hard drive would be like transfering files over a serial cable... like old PS/2 serial, not USB2 serial.

      A: so do it in chunks. My computer's only got half a gig of memory, but I've compressed whole DVDs using it... hey, how did I manage that?

      B: There's the trou

  • Combined video - TV Capture cards exist for AGP, such as the Matrox Marvel.
  • by Anonymous Freak ( 16973 ) <anonymousfreak@nOspam.icloud.com> on Sunday February 19, 2006 @11:05PM (#14758380) Journal
    Where do all these other top-level posters get their information?

    AGP is a subset of PCI. The original AGP spec (1.0) defined a dedicated slot with a 32-bit, 66 MHz PCI connection directly to the Northbridge, plus the ability to directly access main memory more quickly than conventional DMA allowed. AGP 2x then increased speed by using a double data rate system, similar to DDR memory, transferring two data chunks per clock cycle.

    AGP 4x then added a quad data rate connection, Fast Writes (the ability to write to main memory out of normal order,) and Direct Memory Execute (the ability for the AGP card to execute directly out of main memory, rather than having to load into on-board memory first.)

    AGP 8x just oct-data rate'd it. It's still 32-bit, 66 MHz PCI, though.

    But, either way, AGP *IS* a PCI connection. Fully compliant with PCI 2.1, with full bandwidth in each direction.

    There are/were bridge chips that converted the AGP connection into one or more PCI slots, which would become fully-compliant PCI 32-bit, 66 MHz slots. These bridge chips were sometimes used on lower-end server motherboards with onboard PCI video, as a cheaper alternative to adding a separate 64-bit PCI controller. They could be found on products from Intel (L440GX,) and others.

    BUT, since it is only 32-bit, you're limited to a 32-bit, 66 MHz PCI connection. PCI-X requires 64-bit for its faster bus speeds. That means that there are no bridge chips that will give you anything better than a 32-bit, 66 MHz PCI 2.1 connection. You can run multiple cards off this connection (As the Intel board listed above did,) but just as with 'regular' PCI, you are sharing the speed among all the cards.

    But, any 66 MHz PCI card (or any correctly backwards-compatible PCI-X card,) would take advantage of the doubled speed over 33 MHz PCI, though.

    See http://web.archive.org/web/20040205095311/http://w ww.gcsextreme.com/agpfaq.htm [archive.org] for more info. (Sorry, Slashdot's code doesn't want to let me make that into a proper link, it breaks it into 'archive.org' and 'gcsextreme.com' segments, you'll have to copy and paste, then remove the space yourself.)
    • by NekoXP ( 67564 ) on Monday February 20, 2006 @01:51AM (#14759052) Homepage
      I was wondering the same thing myself. Fools.

      It should be noted that the AGP bus in general has snooping turned off; the GART in the northbridge
      handles all of the memory access therefore it can and should always tell when memory is being accessed
      (therefore you can't rely on caching video memory like you would on a PCI card). Without snooping on
      DMA transactions this speeds the bus up somewhat. It also lacks the interrupt routing lines. What this
      basically means though, is that without a bridge chip, it ISN'T exactly the same as a PCI slot - if
      you put more than one device on there, only the first will work, and even if you could, you'd
      effectively trash memory every time you did PCI DMA.

      As PCI ('frame mode') you're right, it's just a 66MHz 32-bit PCI slot. In fact we make two board
      designs at the place I work, one of which puts an AGP slot onto a 66MHz 32-bit PCI bus (and it works
      fine up to the point of having a 3.3V keyed slot, and the industry moving on to 1.5 and 0.8V devices)
      and one which has a 66MHz 32-bit PCI slot which we ship an AGP riser for. Everything Just Works (tm).

      AGP specs *also* has a USB connection routed to it but I dare say it's not been connected on most
      motherboards since the dawn of AGP 2.0 (everyone seems to use I2C on the card and talk via some
      kind of PCI configuration/register space logic instead).

      There is plenty of stuff you can do with AGP but seriously who'd want to these days. You're picking
      up old boards now, trying to do "cool" geeky things with them? What for? You're too cheap to move
      to PCI Express? :)
  • by GrpA ( 691294 ) on Monday February 20, 2006 @02:49AM (#14759266)
    First, install a used graphics card. Then reprogram the graphics card to do other stuff.

    Any time domain project might work.

    eg,
    Audio Card. (Yes, you can produce audio on a graphics card).
    Signal Generator (All kinds of repetative signals you can generate)
    TV Remote (Just connect to a IR led on the output port).
    Digital TV Modulator. http://www.hackaday.com/entry/1234000113073480/ [hackaday.com] This is the Best idea made practical.
    Transmitter (on MANY different frequencies).
    Ultrasonic transducer driver for driving 3 ultrasonic transducers. (Spot sound)

    Just keep in mind you have 3 Digital to Analogue Controllers,
    Programmable clocks
    Memory (and a means of moving it to the DACs)
    and two other digital outputs,

    ALL PACKED NEATLY INTO A VIDEO CARD FORMAT... and it even works with AGP. :)

    GrpA

  • 5 years = obsolete? You all need to get a life and appreciate the age you life in. Snot nose kids. " faster faster' .. .

Don't be irreplaceable, if you can't be replaced, you can't be promoted.

Working...