What is the max ram a VSA-100 can actually have?

    Hey all :)

    For some time, I still remember seeing Voodoo5 5500's have 128MB SDRAM, meaning 64MB per VSA-100, even plans of doing a V5 6000 with 4x 64MB, and there was also a V4 4500 AGP even with 128MB SDRAM.

    V4 4500 128MB = 1x 8x 16MB
    V5 5500 128MB = 2x 4x 16MB
    V5 6000 256MB = 4x 4x 16MB

    Even though, the VSA-100 White Papers describe that the VSA-100 have a max VRAM Support of 64MB.
    As those heavily VRAM upgraded V4 4500's go, was their Bios altered, so 128MB was readable?

    Would it be possible to have a Voodoo5 5500 AGP with 2x 128MB or even a Voodoo5 6000 with 4x 128MB?

    Even that V5's have 4 Ram sockets per VSA-100 over the 8 the V4 4500 AGP has.
    I then thought if you could find 32MB SDRAM chips, this can get interesting, with 128MB per VSA-100 you could do a lot of crazy stuff, the idea of running out of VRAM would also be dealt with.

    So to get 128MB Per chip we could get this:

    V4 4500 256MB = 1x 8x 32MB
    V5 5500 256MB = 2x 4x 32MB
    V5 6000 512MB = 4x 4x 32MB

    If a 2x 128MB V5 5500 is possible I would stand open as Test subject.

    Obi Wan, welcome back my Friend !!!
    Nice to see ya again with us :spitze:

    Really, there are some moded boards around:


    Komponent mod succesfully a v4 with 128mb
    Also he mod for me and others a v5 5500 card with 128mb. ( 8x16mb )

    More was not possible because of the missing chips with 32mb....he was willing to try this mod too, we even found a supplyer for this chips as i recall corectly..but we never manged to do something :adsh:

    As i talk in the passt a lot with him, he also try to mod a v3 ith 32mb, but no luck. I have this moded card, with soldered memory chip :)

    Osckar planned once also a v5 6000 with 4x64mb, but i never heard more about this....

    Ah thanks and yeah I kind of missed this retro stuff , 3dfx mainly being, the x-3dfx group @ Facebook kind of got me back in to doing more with 3dfx hardware and then I kind of got some old memory of what my password for VA was.,
    Well that did the trick so yes I plan on staying for ever now.

    Anyways, so 2x 128MB or 4x 128MB is just a Hypothesis for now then, as in it could work but it has never been proved that it would work...
    Since this was never physically tested to be a proven Theory, hence me calling it a thinkable Hypothesis.

    Even though if a Voodoo4 4500 AGP can handle 128MB this does answer us that the VSA-100 like it's newer spouse the VSA-101 both can support 128MB VRAM.
    Only for the Voodoo4 4500 AGP it's provable.

    Oke, I am quite sure there are more people than that one guy that could deliver the 32MB Sized SDRAM Chips out there, well currently hoping.
    Unless I am not getting something here.

    It would be nice to rebirth some old projects and see if we can get more out of them.

    English:

    Second guess: is it one try of Hanks inspired Voodoo5 5500 PCI - Voodoo5 5500 PCI four-way SLI - modification? And in case: does it work (I guess not)?

    Gold Leader / Topic: I personally have never seen or used a VSA100 - Card with more than 32 MB SD per chip. The original V4 4800 AGP Proto* has 64 Megs as we all know, so does that Voodoo5 5500 (AGP?) 128 MB - Mod (with 'stacked' RAMs) and of course the Quantum 3D Aalchemy 8164.
    *We all know the Modded 4500 AGP with extra RAMs on the solder contacts as self-made 4800s, aswell.

    If 128 MB per chip would be possible, I think those crazy modders would have done that for a long time.

    ------------------
    Deutsch:
    Zweiter Rateversuch: Ist es eine der von Hank inspirierten Voodoo5 5500 PCI - Voodoo5 5500 PCI - 4-Wege-SLI - Mods? Und falls ja: funktioniert sie (ich schätze eher nicht)?

    Gold Leader / Thema: Ich habe nie eine VSA100 - Karte mit mehr als 32 MB SD pro Chip gesehen oder genutzt. Der originale V4 4800 Prototyp* hat, wie wir alle wissen, 64 MB, außerdem diese gemoddete Voodoo5 5500 (AGP? mit gestapelten RAMs) und natürlich die Quantum 3D Aalchemy 8164.
    * Man kennt ja die gemoddete 4500 AGP mit zusätzlichem Modulen auf den Lötkontakten als selbstgemachte 4800.

    Wenn 128 MB pro Chip möglich wären, hätten das diese verrückten Modder schon lange gemacht.

    VSA-100 Can address 128MB check this thread here as Komponent applied 8x 16MB to hos Voodoo4 4500 AGP, since a V4 AGP has space for 8 ram chips ;)
    http://www.falconfly.de/cgi-bin/yabb2/YaBB.pl?num=1232877537

    As you can see VSA-100 supports 128MB VRAM :)
    So thereby I thought of what if we can find 32MB Sized SDRAM chips so then we can apply a 5500 with 8x 32MB per chip would be 4x 32MB SDRAM which results 2x 128MB 128Bit SDRAM makes it 256MB 256Bit SDRAM in theory, due to the SLI being there.

    The only problem is finding these 32MB sized SDRAM chips, they might exist they might not I have no idea, 16MB sized SDRAM chips do exist.

    VSA-101's max RAM Size per chip is 256MB, so 128MB max VRAM size for the VSA-100 seems logic.

    3 Mal editiert, zuletzt von Gold Leader (3. Juli 2020 um 22:56)

    Very, very far back in my head there's that little voice telling me, that some of the adress lines for additional memory are needed for the SLI configuration.
    Maybe I remember that fact wrong, but something tells me, that the SLI configuration has some sort of impact on the memory interface.

    Zitat von OutOfRange

    Very, very far back in my head there's that little voice telling me, that some of the adress lines for additional memory are needed for the SLI configuration.
    Maybe I remember that fact wrong, but something tells me, that the SLI configuration has some sort of impact on the memory interface.


    Oh oke so 64MB per chip was probably the max possible without showing that issue then?
    That may explain why Quantum3D didn't use more than 64MB per VSA-100 for the AALchemy 81xx/82xx boards, I guess :gadget: :)

    Einmal editiert, zuletzt von Gold Leader (14. März 2016 um 18:45)

    As I've said, I'm not a 100% sure if I remember this right.
    Thing is also, that the 5500 with it's Master/Slave SLI configuration differs from boards who use more than two chips.

    Sry, it's just a little bit long ago...

    Zitat von OutOfRange

    As I've said, I'm not a 100% sure if I remember this right.
    Thing is also, that the 5500 with it's Master/Slave SLI configuration differs from boards who use more than two chips.

    Sry, it's just a little bit long ago...


    Hey don't be sorry I have not been in this 3dfx stuff in like ages either I just started coming back man :)
    Maybe Loschzwerg might be able to explain this, since he is the one that has done some crazy mods and fixes with V4's & V5's right?
    From what I can remember though.

    You reply does make a lot of sense, to which why Q3D also didn't surpass the 64MB per chip for their AA5 boards, but that is just my hypothesis :)

    Einmal editiert, zuletzt von Gold Leader (14. März 2016 um 18:50)

    The VSA100 does have a memory bus A and a memory bus B. Each bus features address lines for two memory chips. Additional to that the VSA100 can make use of MCS (memory chip select) and "double" control lines. So in total each memory bus can handle up to 4 memory devices and that's why the V4 AGP can handle 128MB in total by using 16MB SDRAM chips.

    I'm not 100% sure anymore when it comes to the V5, but if I remember correctly the "second" MCS connection from each VSA100 ist not connected to the PCB, simply because there was no space for additional memory and therefore no need for this "feature". In theory it should be possible to "piggy back" additional memory and connect the needed MCS pin.

    When it comes to 32MB SDRAM chips... I haven't seen memory of that kind that fits our needs :( It's hard enough to get those 16MB SDRAM.

    Edit:
    Oh and one thing when it comes to this memory discussion (more than 64MB on a single VSA100): The additional memory gets only useful when the screen resolutions is very high and FSAA is applied. BUT in these kind of settings the VSA100 (no matter what frequency) is missing the needed raw power to produce enough fps for real gaming ;)

    Einmal editiert, zuletzt von Löschzwerg (15. März 2016 um 18:22)

    Zitat


    Memory Architecture: The frame buffer controller of Napalm has a 128-bit wide datapath to RGB, alpha/depth-buffer, 2D desktop, video, and texture memory with support for up to 200 MHz SGRAMs or SDRAMS. For 2D fills using the standard 2D bitBLT engine, 8 16-bit pixels are written per clock, resulting in a 800 Mpixel/sec peak fill rate. For screen clears using the color expansion capabilities specific to SGRAM, 64 bytes are written per clock, resulting in a 12.8 Gbytes/sec peak fill rate. For Gouraud-shaded or textured-mapped polygons with depth buffering enabled, one pixel is written per clock – this results in a 166 Mpixels/sec peak fill rate. The minimum amount of memory supported by Napalm is 4 Mbytes, with a maximum of 64 Mbytes supported. Storing texture bitmaps, the texture memory controller of Napalm must share the 128-bit wide Datapath to Napalm memory. The texture unit uses sophisticated caching to reduce the required bandwidth of memory to perform bilinear texture filtering with no performance penalty. The amount of texture memory is only limited by the maximum amount of Napalm frame buffer memory.

    Host Bus Addressing Schemes: Napalm occupies a combined 256 Mbytes of memory mapped address space, using two PCI memory base address pointers. Napalm also occupies 256 bytes of I/O mapped address space for video and initialization registers. The register space of Napalm occupies 6 Mbytes of address space, the linear frame buffer occupies 128 Mbytes of address space, the ordered texture download port occupies 2 Mbytes of address space, and the 3D pipeline linear frame buffer takes 8 Mbytes of address space.

    Oh wow remember these?
    [Blockierte Grafik: http://abload.de/img/8164s7tf8z.jpg]

    Quantum3D Aalchemy 8264's hehe , look closely... they have 8 SDRAM chips per VSA-100, 4 on the top side 4 on the bottom side meaning they use 8x 8MB SDRAM...Imagine replacing all the ram chips with 8x 16MB Per VSA-100 the endless possibilities...
    128MB per chip hmm it would be in theory a Quantum3D Aalchemy 82128 1024MB :)

    This is possible after I said to my friend ketfri8002 the following:

    Hmm the Quantum3D Aalchemy boards, if they have 64MB per chip this may be why Q3D never went past the 64MB fram buffer mark...
    He was like well they DO have 8 ram chips per chip, 4 on the top side 4 on the rear side..
    Then I was like OMG man you know what I am thinking... 8x 16MB per chip ..
    This would make it possible for such a card to have 128MB per chip, if comparing to how it was done with Komponent's V4 4500 AGP 128MB as it too has 8 ram slots for it's VSA-100, it's the same kind of theory.

    All that got me to this crazy idea, anyways this is what you would get:

    Obi's Quantum3D Aalchemy 81128
    8 Chips - 1 Cycle per Clock - 128MB per chip

    Obi's Quantum3D Aalchemy 82128
    8 Chips - 2 Cycles per Clock - 128MB per chip

    Obi's Quantum3D Aalchemy 84128
    8 Chips - 4 Cycles per Clock - 128MB per chip

    So I even wonder is this even possible? Well Physically it is but will this really work?

    I Ask this since the V4 4500 that this was tried on also has 8 ram slots and Komponent successfully managed to get 8x 16MB on that thing and making it post as a 3dfx Voodoo4 4500 AGP 128MB
    So in theory this might work since these AA5 boards have 8 Ram chips per VSA-100 just like a V4 4500 AGP does..

    This just came in to my mind as a what if :topmodel:
    Anyways I Wonder what you all think of this, I Posted this here since it is the same kind of topic , but just more crazy :D

    3 Mal editiert, zuletzt von Gold Leader (30. April 2016 um 02:10)

    A very intresting idea, would surely be nice to get a "little" extra framebuffer. Really hopes that somebody may want to try it out. But it require a lot of patience. Wich i had the skills to do that stuff, i think if i had the tools would be possible. Have a heatgun, just needs somthing to prick on the solder.

    But other people who have done it before would ofcourse be way better, only says this cuz i have a 8264 with malfunctioning 3d part, everything looks like it works, until you start the 3d demo. Then does it ether chrash, or show some funny stuff on the monitor. Osckhar tested it, and told it may be some defective TSOP, so yea replacing the memory might not be the most stupid thing to do.
    But shit like this is what happens when you buy "untested" stuff.

    "Why so serious!"

    3 Mal editiert, zuletzt von ketfri8002 (30. April 2016 um 03:03)

    Hmm, is it not possible to replace those defective parts with new working parts?
    Unless the costs would be too high :pinch:

    But then again you can't know what works or what doesn't it's the risk you take when buying untested things,
    Even though in my opinion what a seller sells should always be tested and just noted as functional or non functional.

    This should be a rule @ places like ebay, just to avoid complications, so it's not your fault for not knowing :)

    2 Mal editiert, zuletzt von Gold Leader (30. April 2016 um 03:17)