VOGONS


First post, by fitzpatr

User metadata
Rank Member
Rank
Member

Hi, everyone,

I came across a really unique board that I was given, so I decided to document it here!

It is a Fujitsu AG-10E, and belongs to, I believe, Fujitsu HALStation 375 and 385 SPARC64 II machines.

From NetBSD wiki

Fujitsu AG-10e […]
Show full quote

Fujitsu AG-10e

This is Frankenstein's monster of graphics cards. It's a double-wide, double-decked SBus device with three different graphics chips - each with its own memory, a high end RAMDAC and a DSP.
There is:

a GLint 300SX with 2x 6MB framebuffer ( that's what the firmware says, the chip itself claims to have 16MB framebuffer ) and 16MB 'local buffer' - whatever that is, 3D Labs claims they can't find any docs but I doubt they even checked. This chip provides a 24bit true colour visual.
an Imagine I128 with 4MB RAM, this chip provides an 8 bit visual.
a Weitek P9100 with 2MB framebuffer, this chip provides WIDs
an IBM RGB561 RAMDAC which supports a truckload of features and WID control over pretty much everything
an Analog Devices DSP with its own memory, we have no clue what it's being used for

Evidently there is some sort of SBus-to-PCI bridge on the card, from the graphics chip listed above only the P9100 is known to exist as an ?SBus variant, and even that was probably an in-house hack done by Tadpole. We have a kernel driver for it ( agten at ?sbus ) which uses the I128 to provide an accelerated console and switches to the GLint's 24bit visual when X runs. The ?xf86-video-ag10e driver is mostly glue code to map framebuffer and registers via /dev/fb*, run the ?xf86-video-glint driver's 300SX support and deal with the different DAC.

Attachments

  • Fujitsu AG-10E 2.jpg
    Filename
    Fujitsu AG-10E 2.jpg
    File size
    1.34 MiB
    Views
    328 views
    File license
    Fair use/fair dealing exception
  • Fujitsu AG-10E 1.jpg
    Filename
    Fujitsu AG-10E 1.jpg
    File size
    1.65 MiB
    Views
    328 views
    File license
    Fair use/fair dealing exception

MT-32 Old, CM-32L, CM-500, SC-55mkII, SC-88Pro, SC-D70, MU2000EX
K6-III+/450/GA-5AX/G400 Max/Voodoo2 SLI/CT1750/MPU-401AT/Audigy 2ZS
486 Build

Reply 1 of 5, by dionb

User metadata
Rank l33t
Rank
l33t

That's one seriously weird piece of kit.

Do you have any documentation for it or the Fujitsu HALStations?

Possibly it might run in a regular SparcStation. Two big technical questions:
- whether Fujitsu implemented the clock-doubling on the SBus that Sun introduced with the SparcStation Ultra systems or this thing still uses the legacy single-clock SBus.
- whether this thing is only supported in SPARC64/OS or it might also work in Solaris. SPARC64/OS is based on Solaris 2.4 with 64b extensions, the difference in name is purely related to licensing and trademarks.

Note that Solaris itself supported Ultrasparc from 2.5 onwards, but Solaris 7 (2.7 in old numbering) was Sun's first 64b Solaris.

Either way, I'd suspect that lacking a HALStation, a SparcStation Ultra 1 or (better) 2 would be best chance of getting this monster to do anything. It might even run SPARC64/OS if anyone could find the software, as HAL's Sparc64 implements the same Sparc v9 instruction set als Ultrasparc, and some OSs (FreeBSD) run on both CPUs with same build.

Reply 2 of 5, by cb88

User metadata
Rank Newbie
Rank
Newbie
fitzpatr wrote:
Hi, everyone, […]
Show full quote

Hi, everyone,

I came across a really unique board that I was given, so I decided to document it here!

It is a Fujitsu AG-10E, and belongs to, I believe, Fujitsu HALStation 375 and 385 SPARC64 II machines.

From NetBSD wiki

Fujitsu AG-10e […]
Show full quote

Fujitsu AG-10e

This is Frankenstein's monster of graphics cards. It's a double-wide, double-decked SBus device with three different graphics chips - each with its own memory, a high end RAMDAC and a DSP.
There is:

a GLint 300SX with 2x 6MB framebuffer ( that's what the firmware says, the chip itself claims to have 16MB framebuffer ) and 16MB 'local buffer' - whatever that is, 3D Labs claims they can't find any docs but I doubt they even checked. This chip provides a 24bit true colour visual.
an Imagine I128 with 4MB RAM, this chip provides an 8 bit visual.
a Weitek P9100 with 2MB framebuffer, this chip provides WIDs
an IBM RGB561 RAMDAC which supports a truckload of features and WID control over pretty much everything
an Analog Devices DSP with its own memory, we have no clue what it's being used for

Evidently there is some sort of SBus-to-PCI bridge on the card, from the graphics chip listed above only the P9100 is known to exist as an ?SBus variant, and even that was probably an in-house hack done by Tadpole. We have a kernel driver for it ( agten at ?sbus ) which uses the I128 to provide an accelerated console and switches to the GLint's 24bit visual when X runs. The ?xf86-video-ag10e driver is mostly glue code to map framebuffer and registers via /dev/fb*, run the ?xf86-video-glint driver's 300SX support and deal with the different DAC.

Hi, I'm interested in the card I saw your amibay post but was having trouble registering on amibay to buy this card from you so checked my other vintage computer haunts for your username and here you are!

If the admin at amibay gets this sorted I'll buy it there otherwise you can PM me here and we can work something out, I have been hunting an AG10E for years (google an AG10e and you get abuot half a page of my forum and mailing list posts).

The card is intended for sbus machines of course so an LX is probably the minimum machine that could run it and a halstation or hyperstation 30, or early Ultra machine would be the top of the line that could be compatible with it. I'm hoping to find it a home in a hyperstatio 30 board with a custom case.

Note Hyperstation.de also documents the card. And apparently abuot a year before I started looking for one there had been a relatively large liquidation of them on ebay for not much money... but such is life. Again definitly PM me and we'll work something out!

Last edited by cb88 on 2019-09-08, 08:35. Edited 1 time in total.

Reply 4 of 5, by cb88

User metadata
Rank Newbie
Rank
Newbie
derSammler wrote:

Would buy this immediately if it would work in my SPARCstation 5. But I guess it won't.

You could put this in an SS5, however an S24 or TCX (same card different names) is a better match for that machine... and it will get you a decent 24bit color framebuffer. The AFX bus in your SS5 later turned into the UPA bus on the Ultra series.

The main problem on the SS5 with this card is that it would occupy 2 of your 3 sbus slots assuming it would fit (never tried my ZX in my SS5 but it would be similar). An SS20 has 4 sbus slots so can get by with an addon like this a bit better.

Also I believe there are probably still a few rasterflex HRs left for sale on the german ebay last I checked they are the fastest 2D cards on solaris... you may get better use out of them that this franken card for practical use. Note the rasterflex HR is only supported on solaris... but the S24 is widely supported on sparc operating systems and QEMU even emulates it via the TCX option.

RasterFlex HR:
https://www.ebay.de/itm/Vitec-RasterFlex-HR-2 … iIAAOSwQJ5UR2ij

S24:
https://www.ebay.de/itm/SUN-S24-24bit-Hi-Res- … 3QAAOSwxp5dUKms

Reply 5 of 5, by fitzpatr

User metadata
Rank Member
Rank
Member

I have never had personal experience with Solaris machines other than at my school. I have absolutely no documentation or informal on them at all! This was just a part that caught my eye, and seemed like there should be some record made about it, and then sold at a reasonable price if someone could actually use it!

I had the same problem at Amibay the first time that I tried to buy things. Good detective work!

MT-32 Old, CM-32L, CM-500, SC-55mkII, SC-88Pro, SC-D70, MU2000EX
K6-III+/450/GA-5AX/G400 Max/Voodoo2 SLI/CT1750/MPU-401AT/Audigy 2ZS
486 Build