Mac Os 9 booting on: xServe G4 (Detailed Posts)

[MacOS Plus]

One last thing for the night:

  Does anyone know if there is a preferential slot ordering for the RAM DIMMs in the Xserve G4?  At the moment I have three 512MB DIMMs installed starting from the left-most slot.  It probably doesn't matter hugely but I'd prefer to install them in the order it addresses them without skipping a slot.  (If we ever get access to 2MBs RAM in OS 9 I won't have to think about that because all the slots would be full anyway!)

  I'm beginning to wonder if I should be renaming this machine a "9serve" - take that, Apple!  (I claim all rights to that name!!!!  Mwahahahahah!)

[DieHard]

This is so excellent !  I scraped 3 units about 4 months ago. You guys truly suck

My lack of faith bit me in the ass on this one. This is a first for sure. Xserve running OS 9 !!!

[mrhappy]

Wow, this is crazy cool!!

[MacOS Plus]

  Hah, DieHard, you're really going to hate me when I post pics of the two machines running side-by-side!  Don't worry, we'll put the 'religion' back in you.  Perhaps I'll even test a pci expansion chassis and see just how big a monster I can make out of this!  I guess I'm now in market for more Sonnet MDX upgrades.

[MacTron]

This 66 Mhz 64 bits PCI bus, it's double the bandwidth of the most capable Macs.

Imagine a four SSDs raid with the  Seritek 64 bits SATA card (that claims that can work at 66 Mhz!).
Imagine a nVidia 4600 Ti fitted inside this machine.
Imagine the included serial port works for MIDI data.
Imagine a Sonnet 7447 @ 1.8 dual, empowering this machine.
Imagine the three ATA 133 modules working.

You may say I'm a dreamer
But I'm not the only one.
LOL.

[Metrophage]

Imagine the included serial port works for MIDI data.

Good thinking - I have never heard of people actually using the RS232 ports on these for anything. I am curious to know what might need doing to get them working

[MacOS Plus]

  I was just going to ask about that serial port myself.  Not sure it can be used as we want because it was set up as a diagnostic terminal interface that's supposed to work automatically, much as is typical on most other server platforms.  I can at least tell you OS 9 doesn't show its presence as a device at all.  I don't think it was ever meant to be freely accessable that way, but we'll see.

@MacTron - Keep dreaming big - I'm right on the same page with you!

[nanopico]

I was under the impression that the serial port was for receiving signals from UPS to be notified of shutdown so it is probably connected directly to the emu and probably not accessible.
But I could be wrong on that.

[MacOS Plus]

  The documentation said something about external management via a serial terminal.  This would allow access to the machine under certain conditions without monitor or video card.

[nanopico]

Going to investigate.

[MacOS Plus]

  More testing tonight - second PCI bus works, not that there's any reason it should not have.  I put in an Adapted AHA-2940U2W SCSI card just to see if it was recognized.  The only thing it caused, and exactly as it should have, was the start-up disk detection at boot took an extra-long time to try the FW SSD because it was obviously scanning all the SCSI IDs for possible drives.  After it got through that it went ahead and booted as normal.  The card shows up in System Profiler.  Later I will actually attach a drive and see if it behaves.

  Next up I hooked up my 4-slot 32-bit Magma PCI-PCI expansion chassis.  This is a rack-mountable version, which will be great if I decide to permanently pair it with this machine.  It is also the older type with the big low-density cable connectors.  I moved the SCSI card into the chassis and booted.  It recognizes the SCSI card there too!  The boot from the FW SSD completed exactly the same way with no indication of lag.  The only thing I don't know is if using 5-volt-only cards in the chassis is possible.  I will mess around with that more later.

  Finally, just to throw it a challenge, I put the 64-bit LSI fibre channel card this machine originally came with into the remaining slot in the riser (next to the Magma bridge card).  The machine took longer to get through the start-up disk detection, again because it was now scanning the fibre channel bus for a link and drives.  After the long pause it again booted as normal!  System Profiler shows the card present.  I don't know if it would actually communicate with an FC chassis under OS 9 though.  Unfortunately I don't have any other FC gear around any longer to test that with.

MagmaSlots output:

Code: [Select]

                       Magma Slot Peek
                       ===============

PCI Slot Name : $3x4 (Devices:device-tree:pci:pci-bridge:pci-bridge:pci-bridge)
   Card Name : pci-bridge
   Device Type : pci
   Board Version : 6
   Board Vendor ID : 1011
   Bus Slot Number: 4; on PCI Bus Number: 3

PCI Slot Name : $4x7 (Devices:device-tree:pci:pci-bridge:pci-bridge:pci-bridge:ADPT,2940U2B)
   Card Name : ADPT,2940U2B
   Device Type : scsi-2
   Board Model : ADPT,1757800-00
   Board Version : 1
   Board Vendor ID : 9005
   Bus Slot Number: 7; on PCI Bus Number: 4
Look at all those PCI bridges!

System Profiler output:

Code: [Select]

Software overview
Serial number: XB2270GNLZF
Mac OS overview
Finder: 9.2
System: 9.2.2  US
Active enabler: Mac OS ROM 10.2.1 (Generic)
QuickTime: 6.0.3
CarbonLib: 1.6.1
File sharing: is on
Multiple Users: Not installed
Note: No startup disk was selected.
Memory overview
Disk cache: 8160K
Virtual memory: is off
Built-in memory: 1.50 GB
Number of empty RAM slots: 1 (DIMM0/J21)

Location            Size           Memory type
DIMM1/J22   512 MB   PC2600U-30330
DIMM2/J23   512 MB   PC133 CL3
DIMM3/J20   512 MB   PC133 CL3
Video memory: 32 MB Built-in display in use
Backside L2 cache: 2 MB
Hardware overview
Machine ID: 406
Model name: RackMac1
Keyboard type: Apple Pro Keyboard
Processor info: PowerPC G4
Machine speed: 1000 MHz
Processors: 2
Nanokernel version: 2.28
Nanokernel pool extends: 11
Nanokernel scheduled CPUs:2
uni-n: 36
Network overview
Ethernet built-in Link: up Speed: 1 Gbps Duplex: full
Modem
Name:
Protocol:
Version:
Status: No Apple modem found.
Open Transport
Installed: Yes
Active: Yes
Version: 2.8.3
AppleTalk
Installed: Yes
Active: Yes
Version: 61
File sharing: is on
Default AppleTalk zone: Not available
Active network port(s): Ethernet built-in
This network: 65280
This node: 128
Router: <not available>
Hardware Address: 00.03.93.b5.e3.f2
TCP/IP
Installed: Yes
Active: Yes
Version: 2.8.3
Personal Web Sharing: is off
USB Printer Sharing: Not installed
Netmask: 255.255.255.0
IP address: 192.168.0.194
Default gateway address: 192.168.0.1
Domain:
Name server address: 8.8.8.8
Production information
ROM revision: $77D.45F6
Boot ROM version: $0004.44f1
Mac OS ROM file version: 10.2.1
Serial number: XB2270GN-LZF-0000
Software bundle: Not applicable
Sales order number: Not applicable
Fun!

[MacOS Plus]

Last thing for tonight:

  I stumbled upon an ATI Radeon 9200LE 128MB PCI video card I'd completely forgotten I had.  Obviously I wanted to try this in place of the 'lowly' Radeon 7000 I've been testing with so far.  But why put it in the single-slot riser when there is much more fun to be had?

  I already know that the as-built config from Apple with the 8500 PCI video card placed the card in the 2-slot riser (because a Broadcom 64-bit 66MHz gigabit ethernet card took over the single-slot location).  This meant, as it should be, that a PCI video card should be detected on any PCI bus or bridge and just work.  Performance would be less potentially, but wouldn't it make for a neat experiment to try a video card in the Magma chassis?

  So the 9200 went into the Magma.  It initiallized and displayed Open Firmware properly so I continued with the experiment.  It booted through to the desktop without a hitch!  It even supports a higher refresh rate at 1440x900 than the 7000 card did.

  I've learned the hard way in the past that one of the most effective ways of determining the stability of an expansion chassis setup is to put the primary video card in it.  Given that this worked through this older chassis even, the PCI bus seems to be no issue at all on the Xserve with OS 9.  I'll try a SCSI card again later with some heavy disk activity though so I can be certain.

  Now if only I had a functioning ProTools TDM on here.  I've really got to get my authorization disk issue sorted out.

  Afterward I moved the card from the chassis to the single-slot to compare Cinebench stats.  One of the figures had taken quite a dent from being in the lesser-capable slot prior.  Now the Stats were where they should be.  Interestingly they are virtually the same as for the 7000.  I know the 9200 is also kinda crappy too, but it has to be better in ways this particular benchmarking doesn't record.  I'm pleased it works without issue though.  It's probably a decent and affordable choice for anyone looking to outfit themselves with an Xserve G4 on OS 9 and will work without accessory power feeds or taped pins as with some of the more capable AGP cards.  (It's difficult to get hold of an AGP-AGP riser for these machines anymore anyway.)

[MacOS Plus]

  I decided since I was trying to test as many configurations as possible that I would finally take the potentially risky step of seeing if an ADC-type AGP video card would safely function in the Xserve AGP riser.  The only slot pins that pass through to the ADC port are data lines for the ADC USB bus and should not be of consequence in a 4x AGP slot where those pins aren't assigned in the generic slot specification.  I assumed this carried over to the PCI spec also such that no pins should be in conflict.  The ADC power doesn't go through the main section of the slot so it is safely physically isolated.  (Any 8x card probably wouldn't require the same pin taping/cutting as for ADC-slot machines and would likely run at 4x automatically, but they wouldn't have OS 9 support anyway.)  I went on the assumption that it was unlikely Apple did anything out of spec for this slot because the same pins were used for both riser types and had to be safe for any standard PCI card too.  Still, I've learned from experience not to trust Apple to not screw around with standards, but I considered the risk to be low in this case.

  I went with the first OS 9-compatible ADC video card I found that would physically fit, a GeForce2 MX 32MB.  It actually worked without any trouble, aside from the fact it defaults to a very low resolution until it boots once properly with drivers.  After that I could set the same resolution I needed before - 1440x900.  It is connected to the VGA output directly rather than through an adapter off the ADC port.  Cinebench shows a substantial increase in OpenGL performance with this lowly card, proving that AGP communication with acceleration enabled makes all the difference over PCI in this machine.  I guess that should come as no surprise.

  Obviously the only thing to keep in mind is that you can't directly connect an ADC monitor without an adapter to power it and provide the USB link.  Conversion to DVI or VGA with adapters should work because the ADC power and USB lines aren't needed for that.  This means a vastly increased pool of accessible video cards for anyone trying to piece together one of these Xserves, especially for those looking at supporting both OS's in a dual-boot configuration.

  Anyone else ready to try their Xserve yet?

[DieHard]

 

 

All of mine were sent to the scrapper, but the CPU boards live in my MDDs. 

If I get another one in I will test Cubase and some other DAWs.

[MacTron]

All of mine were sent to the scrapper, but the CPU boards live in my MDDs. 

If I get another one in I will test Cubase and some other DAWs.

I'm anxious buy an xServe also. But after buying the two HP's displays my budget to this kind of stuff is empty for a couple of months at least ... 

While, I will wait to if a solution for the ATA 133 module is found.

This is by far, the more important unsupported G4 Project.

Congratulations and thanks to ALL.

[MacOS Plus]

  Keep in mind that while the ATA100 version of the Promise chip, PDC20270, had a functional OS 9 equivalent ROM, the ATA133 PDC20271 never did.  This would make it far more likely that a stock Xserve "Tray-load" with the ATA100 can be made to work than the faster one in the Xserve "Slot-load".  Not that it's impossible, but Sonnet did go with a completely different vendor's chipset for their ATA133 RAID card for some reason.  At the very least you might be able to simply swap an ATA100 bridge board into a "Slot-load" machine if we can get that controller working in the first place.  Otherwise just go with a PCI card for now.

  On that note, a reminder to all that we're still seeking a Sonnet Tempo RAID ATA100 PCI card for testing on this project.  Also, if anyone scraps an Xserve "Slot-load" I'm interested in acquiring the front bezel plate and possibly some other bits.

Okay, time for some pictures... 'cause it happened!  Check out the epic unintended lens flares:

Xserve meets 9serve - they seem to have the same taste in websites!  (TenFourFox vs Classilla - PPC browsing at its finest.)  They also like similar keyboards.


Xserve "Tray-load" - but wait, swapped in a vastly better slot-load DVD drive.  Four-channel fan speed controller tamed the noisy beast!  Bluetooth keyboard on ancient Kensington USB BT dongle.


9serve "Tray-load" - also with a slot-load drive (no faceplate - 'borrowed' from Powerbook Ti).  Booting SSD from front firewire.  Lacie USB floppy and Macally 'floating-apple' mouse.


Bestest desktop picture ever!   "G4 WITH VELOCITY ENGINE"  Really awesome but makes no sense at all - just like most Apple marketing!


Xserve "About This Mac":


9serve "About This Computer":

[nanopico]

  Keep in mind that while the ATA100 version of the Promise chip, PDC20270, had a functional OS 9 equivalent ROM, the ATA133 PDC20271 never did.  This would make it far more likely that a stock Xserve "Tray-load" with the ATA100 can be made to work than the faster one in the Xserve "Slot-load".  Not that it's impossible, but Sonnet did go with a completely different vendor's chipset for their ATA133 RAID card for some reason.  At the very least you might be able to simply swap an ATA100 bridge board into a "Slot-load" machine if we can get that controller working in the first place.  Otherwise just go with a PCI card for now.

I was digging around on this, I couldn't find anything on the Sonnet ATA Raid 100 on their site, but I did find the ATA Raid 133 and it was supported as a bootable device in OS 9.  So their may be hope yet on both of those.

[MacTron]

At the very least you might be able to simply swap an ATA100 bridge board into a "Slot-load" machine if we can get that controller working in the first place.

Interesting idea.

I was digging around on this, I couldn't find anything on the Sonnet ATA Raid 100 on their site, but I did find the ATA Raid 133 and it was supported as a bootable device in OS 9.  So their may be hope yet on both of those.

Even better idea

We have to face a several issues, find the firmware with Mac os 9 drivers compatible with this ATA133 bridge board  and know how to install them inside the board  (I guess we have to reflash an EPROM or something similar)

[nanopico]

At the very least you might be able to simply swap an ATA100 bridge board into a "Slot-load" machine if we can get that controller working in the first place.

Interesting idea.

I was digging around on this, I couldn't find anything on the Sonnet ATA Raid 100 on their site, but I did find the ATA Raid 133 and it was supported as a bootable device in OS 9.  So their may be hope yet on both of those.

Even better idea

We have to face a several issues, find the firmware with Mac os 9 drivers compatible with this ATA133 bridge board  and know how to install them inside the board  (I guess we have to reflash an EPROM or something similar)

Sonnet has a firmware update for the ATA 133 Raid.  I am considering trying to flash that one, hoping that it will fix the firmware properties.

The firmware properties seem to be the only issue so that it can be initialized properly.  Once booted I'm pretty sure drivers won't be an issue.

[MacOS Plus]

  As I mentioned, the Sonnet Tempo RAID ATA133 doesn't use a Promise chipset.  It's a completely different foundation.  Just to confuse the issue, the Sonnet Tempo ATA133 (non-RAID) did use a Promise chip but not the same as the one in either revision of the Xserve.  The boot code block within the firmware is probably the same between the RAID and non-RAID chips however.  This might mean 'grafting' that block of code into the another firmware file and possibly recalculating the checksum.  This isn't my area of expertise but I'm sure there is someone out there with experience editing video card ROMs who could help.

  You are most likely right that the firmware is probably the only issue, not drivers.  The only thing I'm unsure of is how this plays along with the fact there are actually two Promise controllers in the Xserve, one on top of and one on the bottom of the bridge board.  It may not matter at all or it might mean only one can be enabled at any time.  We will see.

  It shouldn't affect OS X support after the change, but I'd like to get a spare bridge board to experiment with so I don't permanently screw my current one.  If it somehow creates a situation where OS X doesn't like it any more then at least I would have the option of swapping bridge boards at any time.