Sony GSBox, the Sony GSCube technology prototype

Discussion in 'Rare and Obscure Gaming' started by jollyroger, Jun 16, 2018.

  1. jollyroger

    jollyroger Gutsy Member

    Joined:
    Oct 18, 2008
    Messages:
    458
    Likes Received:
    256
    Everyone, I will use this thread to report all the findings on this truly special piece of hardware that came into my possession recently.

    First of all, if you don't know what a GSCube is, there are some excellent threads here on ASSEMBLER, I will link them here in due course.
    At some point in the past I had a chance to spend a few hours with a GSCube, which I have always considered a really fascinating machine; the GSCube was similar in spirit to E&S, GE Aviation and later Lockheed Martin and SGI image generators of the past, used to create real-time 3D experiences ranging from medical training to full military simulators. In Sony's mind it was destined to represent a new avenue to use consumer-level hardware for the purpose of high-end real-time 3D visualization, dedicated to the media industry. I have a fascination with parallel and distributed system, always have, always will have.

    A few months back, a fellow collector and friend (who shall remain unnamed for now) asks me if I would be interested in purchasing what appeared to be a large Sony box, sporting a number of blades and many connectors. A friend of his apparently saw the box for sale online, and having recognized it from many years back he decided to purchase it, but after some initial attempts at restoring and working with it (crucially he had salvaged a hard disk that was pulled from another one of these back in the day), it didn't appear to work properly, and he was thus interested in trading it for something of his liking, something my friend had in his collection. My friend immediately thought of me, as I have quite a bit of experience with parallel systems and he knew I would try my best to investigate and restore its functionality.

    My mind immediately jumps to the GSCube; I knew all too well how slim the chances were that one of them would have skipped the chopping, but I decided to dig in, and after seeing a picture I was surprised. It was not a GSCube, it was a grey/silver box I had never seen before. It was not even closely as refined as a finished product needs to be. Like other people here, I have seen quite a bit in my life as far as development and prototype systems go, but this one was new to me, and the Internet had no clue either.

    We quickly agreed on a three-way deal which included cash, hardware, trading, and the machine was mine. After a couple of weeks of wait the machine gets to me in an absolutely impeccable packaging, but I found it had suffered some damage in a previous transport, likely between Japan and the US. It had clearly been dropped from a height or smashed against something else, many screws were missing, it was very dirty, it really looked salvaged from a scrapyard.

    _IMG_2240.jpg _IMG_2243.jpg _IMG_2249.jpg

    It turns out this machine is called "GSBox", and it represents one of Sony's first attempts at scaling the Playstation2 into a larger, distributed image generator, which would lay the foundations for what eventually became the GSCube.

    This particular system is a "GSBox 0.3S". As we will see there were other versions, and this box was actually part of a larger system called "GSBox 0.3D".

    To be continued...

    Jollyroger
     
  2. unclejun

    unclejun Site Supporter 2011-2014

    Joined:
    Nov 12, 2005
    Messages:
    1,919
    Likes Received:
    136
    That looks great!
    From your other thread it sounded like it came from Yahoo auctions, how much didit got sold for, by curiosity...
     
  3. Jackhead

    Jackhead Site Soldier

    Joined:
    Apr 3, 2008
    Messages:
    2,445
    Likes Received:
    628
    awesome! Love to see some inside pictures.
     
    Syclopse likes this.
  4. PixelButts

    PixelButts Site Soldier

    Joined:
    Aug 19, 2014
    Messages:
    2,730
    Likes Received:
    2,023
    Neat. Mind documenting some normal procedures in operating it?
     
  5. jollyroger

    jollyroger Gutsy Member

    Joined:
    Oct 18, 2008
    Messages:
    458
    Likes Received:
    256
    I don't remember what it sold for, I think somewhere around $1000. That was way less than its value, as the seller clearly had no idea what it was, and it was in a pretty bad shape. Prior to this box surfacing, there was no information about its existence at all, and I would wager this is probably the only one that still survives.

    Of course! I will try to write something about it every day. Stay tuned.

    For sure, I will try to explain everything.
     
  6. jollyroger

    jollyroger Gutsy Member

    Joined:
    Oct 18, 2008
    Messages:
    458
    Likes Received:
    256
    Before turning it on for the first time, I wanted to get a little familiar with the system, so I thoroughly read the system manual that describes the high-level operation of this machine.

    The GSBox is somewhat similar to a PS2 TOOL: it consists of a so-called "Communication processor", which is a full x86 PC, and four "Graphic Synthesizer Modules" or GSM, each equipped with a Emotion Engine and a Graphics Synthesizer, plus additional hardware.

    The Communication processor runs a custom variant of Red Hat Linux, and contains special drivers to communicate with the system backplane, which connected via a PCI slot.

    The backplane hosts the connectors for the four blades, and a custom processor that performs merging of four video streams, more on this later.

    With enough information on the machine, I started the restoration process. First of all I had to remove and dump the hard disk, to salvage anything present in there and to have the ability to experiment without compromising what was in there.
    The amount of dust in the machine was unreal!

    _IMG_2260.jpg _IMG_2317.jpg

    After creating a raw image of the entire hard disk, I cleaned the disk, put it in an antistatic bag and stored it.
    As expected, the disk contained a few partitions, all coming from an installation of Red Hat Linux 6.1. I then sourced an ISO of the original OS and downloaded it, in case I would need it at some point.

    I then proceeded to write the disk image onto a new IDE hard disk and installed it in the machine for a very first trial run.
    After connecting power to the two power supplies (one for the PC side, the other for the backplane and blades), a keyboard (using a PS/2 dongle) and a VGA monitor to the PC side VGA port, I started it and it turned on properly.

    The activity panel at the front of the machine had only three of the four LED columns illuminated, so I decided to turn everything off and investigate.

    I gently pulled out blade 0 using the latches at the back of the machine, and peeking inside, I could see two wires dangling inside the chassis. Checking blade 1, I took note that two identical wires were connected to it, so I connected the first two wires to blade 0 and slotted it back in.

    _IMG_2264.jpg
    _IMG_2265.jpg

    Then I turned the machine back on, success! All four LED bars were now working properly...

    _IMG_2275.jpg

    More to come...

    Jollyroger
     
  7. wisi

    wisi Rising Member

    Joined:
    Apr 16, 2016
    Messages:
    56
    Likes Received:
    76
    Truly remarkable!
    Some observations:
    The early GMAIN 2.0 PS2 TOOL PCB (MiniRa ... whatever that means) has placement points for those five connectors that can be seen around the CXD9553GB IOP, but this one can be seen having the connectors themselves too. Maybe they were used for debugging and performance analysis of the IOP and peripheral hardware. The later TOOL PCBs lack them.
    http://psxmedia.ign.com/media/news/image/ps2motherboardbig.jpg

    Curiously each blade can be seen having its own Sound Interface on the SIF-1 PCB. Odd why SCE used such abbreviations, as it matches with the IOP<->EE SIF Sub-system Interface, but then again there are also the VIF (Video Interface) board and the VIF (Vecor unit interface), which have nothing in common.
    Each blade can also be seen having an EE SIO port, AVmulti port and a VGA port.
    A connector similar to the one for the VIF board on the TOOL for the S-Video input is also present, so perhaps there was an option to add such input.
    There is a PLX PCI bridge chip on each blade, as with the GSCube.
    EDIT:
    Wrong: It seems like instead of the Xilinx FPGA (found on the backplane of TOOLs), an Altera FPGA (on the back of the PCB) is used. It and the four FIFO chips below it, together with the PLX PCI bridge make-up the PIF PC(PCI)<->IOP interface.
    Correction:
    The Xilinx FPGA for the PIF/MRP is visible under the sound interface PCB (see the second page of this thread), while the Altera FPGA on the bottom side of the (GSM-1) board is for the AIF. The through-hole mounting points for the AIF HDD PATA port and RTC are visible between the sound interface (SIF-1) board and the Processors Module (PM-1) board.
    There doesn't seem to be a PCMCIA slot and Dev9C under the SIF-1 board.
    END_OF_EDIT

    I guess the only way to actually upload data to the PS2s is through the PC->PIF. Could it be that this was the slow way of uploading data that was discussed in the GSCube thread, as being too slow to be useful, and thus required some special cables to connect to external workstation to achieve higher speeds.
    From: https://assemblergames.com/threads/gscube-information.18036/
    https://assemblergames.com/threads/gscube.9569/

    The main PCB appears to be marked GSM-1.
    The IOP has SIO2 (Controllers & MC), USB and iLink integrated, so the main PCB probably has an option for having them connected and made available on the back panel, similar to the S-Video input interface.

    Please excuse me interrupting your story.
     
    Last edited: Jun 20, 2018
    AKuHAK, pool7 and Shuji like this.
  8. jollyroger

    jollyroger Gutsy Member

    Joined:
    Oct 18, 2008
    Messages:
    458
    Likes Received:
    256
    Those connectors are very clearly Logic Analyzer probe attachment points, so in my opinion they are there to debug the hardware.
    There are several other LA connectors on the backplane too.
    I will post better pictures of everything.
     
  9. Bad_Ad84

    Bad_Ad84 The Tick

    Joined:
    May 26, 2011
    Messages:
    8,618
    Likes Received:
    1,387
    Are you saying they are logic analyser connector points based purely on the type of connector used?

    Ive seen all types of connectors used for all types of purposes. If you have actually investigated them, then disregard. Otherwise my advice is don't just assume
     
    DeChief likes this.
  10. jollyroger

    jollyroger Gutsy Member

    Joined:
    Oct 18, 2008
    Messages:
    458
    Likes Received:
    256
    Well, I cannot be 100% sure, but having worked with some LA systems in the past, I recognized the Tektronix high-density multi-channel probe connectors... ;)
     
    Bad_Ad84 likes this.
  11. jollyroger

    jollyroger Gutsy Member

    Joined:
    Oct 18, 2008
    Messages:
    458
    Likes Received:
    256
    After starting, the BIOS stopped right away. It turns out the BIOS backup battery was dead (of course), the configuration reverted to the default behavior of stopping for any error, and it couldn't find the keyboard. After replacing the keyboard with a true PS/2 keyboard, I fixed the BIOS clock, and the BIOS proceeded fine.

    The Red Hat Linux OS booted fine, besides complaining of the ethernet port being disconnected, and I got to the login prompt.

    Only the root user exists, so I use the password provided and login correctly and start inspecting the content of the hard disk to find the SDK, and sure enough I found it. The Communication Processor can be used as a complete workstation, as the full SDK is present, including all PS2 compilers, tools and libraries, very similar to the standard PS2 SDK version 1.4.5.

    I launch one of the GSBox specific examples, and... nothing. The example seems to start correctly from the debug messages, and the activity LED strips at the front of the machine light up correctly, but no signal comes out from the system's main VGA video output port.

    _IMG_2280.jpg
    _IMG_2281.jpg

    Time to investigate further. Each GSM blade has several connectors at the back: a serial port, a AV multi-output port (only analog audio is enabled), 4-channel digital audio connectors and a VGA port that shows a 640x480 video output for that GSM.

    I turn the machine back off, disconnect the VGA cable from the main system VGA output and connect it to blade 0 VGA port, then turn everything back on and re-launch the demo, sure enough it works!

    _IMG_2284.jpg

    I tested one by one the other GSMs, and all of them produced a valid output, so it was the "Merger" part of the system that was not working properly.
    Time to do some analysis on the system's main VGA port...
     
    AKuHAK, sa1, DeChief and 7 others like this.
  12. HI_Ricky

    HI_Ricky Intrepid Member

    Joined:
    Jun 7, 2007
    Messages:
    670
    Likes Received:
    213
    wow , that cool :)
     
  13. sp193

    sp193 Site Soldier

    Joined:
    Mar 29, 2012
    Messages:
    2,235
    Likes Received:
    1,077
    This is a very nice finding!
    Did you ever figure out what the diskette drive was meant for? No DTL-T10000 unit ever had it. I heard that Sony might have used it to program in the board device model details into the DEV1 flash ROM, but no TOOL console had it.

    Have you figured out whether the device was meant to be logged into? Unlike what seems to be happening around here, nobody was meant to log into the TOOL units...
    Arguably, it might have been initially designed to be a usable PC (and this might have been part of that project), but the late Sony decision was to not allow the built-in PC to be used as a workstation.

    Do you know when this prototype might have been built? I believe that they made the final boot ROM image (17/01/2000) with SDK v1.3.4, due to a Sony comment that the TOOL's test mode will provide DECI2 functionality similar to SDK release 1.3.4.
    Hence SDK v1.4.5 could indicate that it was made after the standard boot ROM, but it doesn't mean that the device is that new.

    Ridge Racer has IOPRP14E.IMG (perhaps release v1.4.4), dated internally within the comment as 2000/02/17.
    IOPRP16.IMG is dated 2000/05/16.

    We can visibly see that this device has a CXD9553 as an IOP, much like the SCPH-10000 and the DTL-T10000. But even the DTL-T10000H units had the first-generation chips (EE, GS and IOP).
    Does this unit have a PC CARD slot?

    You mentioned that the HDD has some samples. Can you elaborate on what samples there are?
    Other than us knowing that this is able to render at a high resolution, there must be some 3D demo to showcase its power?
     
    Last edited: Jun 18, 2018
    uyjulian, AKuHAK, sa1 and 1 other person like this.
  14. wisi

    wisi Rising Member

    Joined:
    Apr 16, 2016
    Messages:
    56
    Likes Received:
    76
    Can't wait to find-out how the images from the four GS chips are merged. :)

    This may explain what SP193 mentioned in the other thread: https://assemblergames.com/threads/sneak-peek-of-the-rare-system-i-am-restoring.69410/#post-973254
    How a GS can have use for high-resolution image output while not having sufficient memory for that - due to it only rendering 1/4 of the output image. Although I am not sure exactly how this would work, given that the PCRTC would have to be set to display more than the display buffer, so wouldn't that result in the draw, Z-buffer and textures also appearing in the image of a single GS, which is not seen in the output image of a single GSM above.
    Are the images from the GSMs merged according to the alpha values stored in them, or are they combined with a fixed mask?
    The GS has some unknown bit-fields in its registers. I wonder if the have any use in this case. The only known use of the GS digital video input is on the DTL-T10000, although what that use was is uncertain. It has a test program for it though.

    It would indeed make sense to have a capability to add something connected to the Dev9 SSBUSC channel, and/or the CDVD SSBUSC channel, so there might be connectors for that, although in the above photos such can't be seen. Although there do seem to be connectors for some among the S-Video input, USB ports, iLink, maybe controllers and MC ports, although those that are present seem to be covered on the back.
    It is interesting that the EE SIO ports are on the back of each blade, rather than connected to the PC communication processor, like on the TOOL. I guess it would have became quite difficult when the blades got to 16, to connect all of them to the PC. :D
     
    pool7 likes this.
  15. speedyink

    speedyink Site Supporter 2016

    Joined:
    Apr 10, 2015
    Messages:
    1,453
    Likes Received:
    647
    This is all so fascinating!
     
  16. jollyroger

    jollyroger Gutsy Member

    Joined:
    Oct 18, 2008
    Messages:
    458
    Likes Received:
    256
    Everyone, please hold your thoughts, I would like to first finish the "story" of how I got it to work, and post the corresponding pictures.
    By all means, then ask away all the questions you want!
     
    segantleman likes this.
  17. iCEQB

    iCEQB Peppy Member

    Joined:
    Feb 22, 2008
    Messages:
    322
    Likes Received:
    37
    If only @sequent_blender was still around :(
    He would probably lick his screen now
     
  18. Mattetch

    Mattetch rebuilding collection one cartridge at a time

    Joined:
    Dec 29, 2013
    Messages:
    32
    Likes Received:
    6
    realistically what can be done with this machine? i really like the hardware but i'm puzzled what can actually be done with it.
     
  19. jollyroger

    jollyroger Gutsy Member

    Joined:
    Oct 18, 2008
    Messages:
    458
    Likes Received:
    256
    The most basic answer is "rendering scenes with four times as many polygons as a regular PS2 TOOL". With a game source code, it would be relatively straightforward to make it run at twice the resolution of a PS2...
    It is an image generator intended for rendering high-resolution 3D media in real time, but it was not powerful enough even in the 16 blades configuration, so ultimately it was never offered for sale.
     
    DeChief, spinksy and Shuji like this.
  20. jollyroger

    jollyroger Gutsy Member

    Joined:
    Oct 18, 2008
    Messages:
    458
    Likes Received:
    256
    Then I started to analyze the VGA output.
    I equipped myself with a Digital Oscilloscope and Logic Analyzer, to try to find the issue backwards from the VGA port.

    _IMG_2339.jpg

    To start with, I connected the DSO to the VGA port of one of the blades, to get a reference known-good signal, given that connecting any of the blades to a VGA monitor produced a valid picture.

    _IMG_2374.jpg

    Most pins seemed to work, but HSYNC was not emitting anything, so I thought I may have found the culprit.
    And sure enough, I found that the main VGA connector had probably taken a hit, and three of the traces were mildly detached from the board, right under the connector.
    A quick soldering iron job, and presto! all signals were now present.

    _IMG_2375.jpg
    _IMG_2378.jpg

    Reconnecting the VGA monitor, and success! The system finally displayed what it was supposed to: a properly composited mix of the four outputs coming from the GSMs.

    _IMG_2377.jpg
    _IMG_2385.jpg

    Now that operation was confirmed, it was time to appropriately restore the chassis, which was both extremely dirty and badly bent in numerous places...
     

Share This Page