This page is a wiki. Please login or create an account to begin editing.

9 posts / 0 new
Last post
SkyCapt's picture
Joined: 2017 Jan 11
AGP-16x graphics acceleration & 4GB RAM on PowerMac G4 Mirror Door 2003

This is fresh, nobody else's yet made this happen. I create an obvious two fold increase in graphics render/display speed when I combine these 2 things: greater than 2GB RAM plus changing the firmware's model code from G4 to G5. I already have a G5 graphics card in my Mirror Door '03, the ATI Radeon 9600 Pro 64MB vram. And I have G5ish RAM 1GB per module 467MHz CASL2.5 (declared to be CASL3.0 at least 400MHz).

The acceleration takes effect in OS 9.2.2 & OSX 10.2.7(G5) thru OSX 10.4.11 - not in 10.2.8(G4) nor in OS 9.2.1 nor OSX 10.5.8 Leopard. See below, for more on OS9's weird behavior. I'm sure that Leopard can do this too, but there's an additional lockout to achieving it, Leopard will need another new trick discovered in order to accelerate.

I'll describe more specifically what I'm doing, and, while I've already witnessed past acceleration due to my RAM and my G5 graphics card, I don't know yet the specifics of how and why this here newest two-fold speed increase is happening. This comes as such a surprise because I thought I had been pushing the hardware as fast as it could go. fyi my screen resolution is 1920x1200.

There's a wide range what could be causing this additional two-fold leap up. I'll hailmary a first best guess. The OSX+CPU recognizes only 2GB RAM. The RAM in excess of 2GB (right now I'm succeeding with 3GB total, 4GB isn't required) RAM is available to the GPU/AGP without CPU speed taxation. The CPU could write/read the excess RAM via bank selecting, or via specific commands to the memory controller. *The GPU can access the excess RAM faster than it can its own 64MB vram. At 1920x1200xMillions of color, the 64MB vram can hold only about 7 frames, less than 7 after factoring in texture data and GPU commands etc. This new two-fold speed increase is basically the result of gaining the equivalent of 1024MB-2048MB superaccelerated vram, when there's 3GB-4GB total RAM. System Profiler now says my computer is a "Power Mac G5" but with a G4 CPU. I dedicate this discovery to ArsTechnica and everybody else who Dropped The Ball (..."there is no proof!" spoken like the trade federation in Star Wars Episode I)


SkyCapt's picture
Joined: 2017 Jan 11

This project requires the PowerMac G4 MDD2003 computer, the MDD2002/FW800 computers are not good enough because they can't install more than 2GB total RAM and, the MDD2003 has a more advanced memory controller which accepts 1GB RAM per slot therefore up to 4GB total. Apple's "Setup Guides" explicitly state 512MB per slot MDD2002 vs 1GB per slot MDD2003. **Are there more G4 models in which greater than 2GB RAM can be attached? ?

I haven't yet tested what if any acceleration happens with my G4 graphics cards (I have the MDD's native ATI Radeon 9000 Pro 64MB vram and the MDD's other native option Nvidia GeForce4 Ti 4600 128MB vram). Instead I'm using the ATI Radeon 9600 Pro 64MB vram, an OEM G5 edition. It's already been like TWICE the card the 9000 was, so I've little interest in the G4 cards even if they are supported in OS9 ... the 9600 is not utilized when booted in OS9.

G5 graphics card required a trick to make it function in this G4 computer. Specific info has circulated widely, the G5 card needed two of its AGP pins disabled, pin #s A3 and A11.

The main memory bank could take a lot of explaining. Start with knowing you need modules with 1GB capacity in each, test them installed solo one at a time and run loads looking for crashing. See if you can, still with the G4 machine id in firmware, sort your modules by their apparent performance speed, not their profiled speeds. You need 3 modules of preferably the best speed which can all work together without crashing in the three ram slots nearest the cpu.

RAM exceeding 2GB is NOT recognized by the firmware nor any version of OS X nor OS 9. Excess RAM is however recognized by the AHT (v2.0.2-v2.1.0) for PowerMac, where you see "3GB" or "4GB" colored red.

AHT software on the 10.2.7 G5 disc, AHT v2.1.0 looks more appropriate for the Mirror Door 2003 than the one supplied with Mirror Door (v2.0.2). But it turns out AHT can't run with the Mirror Door in G5 machine id mode, you must switch to G4 machine id or use an edited AHT disc/image to run AHT. The AHT v2.1.0 must be edited to run on Mirror Door. Fortunately, such an edited AHT has been posted on mg for a couple years, the incredibly awkward "AHT All-In-One beta1" has all 6 versions of AHT for mirror doors, on one bootable disc and a menu for choosing them. EDIT: AHT "All-In-One for PowerMac" nicely updated, found here:

This "insane" graphics/video acceleration (of thread topic) can be seen, and without modifying the firmware, by booting into the G5-specific version of OSX 10.2.7 (Build number 6S80) found here:

Unfortunately, unless you use a G5 to do the 10.2.7 install and then offer that disk to the Mirror Door, or edit the 10.2.7 Install Disc (see next link below) -- you are best-off changing your Mirror Door's machine id in firmware from G4 ("PowerMac3,6") to G5 ("PowerMac7,2") - some firmware instructions included in this link:

I havent offered how to make the firmware's machine id become semi permanent, I could do that later, here below.

After updating OS X 10.2.7 (G5) to become 10.2.8 (G5) (build number 6S90) or if you apply a QuickTime update, the graphics slows back down to G4-video mode, but only if the firmware's machine id is G4. So adopting the counterfeit G5 id is what allows 'insane' speed within OSX 10.2.8(G5 build) and higher.

note: Always remain aware of the performance and limitations of your Mirror Door's air cooling system. Increasing performance such as the topic of this thread, comes with increased demand for cooling, and thus may necessitate improvements to one's cooling system, avoid hardware damage due to overheating.

OSX 10.2.7(G5) runs with 'insane' graphics speed on MDD2003 and you don't need to know any firmware commands to experience it. When Apple/internet tried to discredit/destroy OSX 10.2.7/10.2.8 for G5 by posthumously naming it "Smeagol" (nee Jaguar) and saying it had been (successfully) recalled, I thought they were trying to hide something rather than be helpful. Now I'm certain. The G5 graphics card lock-out, needing to disable two AGP pins to make cards work, is the same way - it hides this G4's abilities.

SkyCapt's picture
Joined: 2017 Jan 11

are those crickets I hear?

In a few days I'll take down my "AHT All-In-One beta1" & put up version 0.2 which will be for PowerMac G4/G5 only.

The concept in beta1, ALL new world PPC AHT on one CD, requires a firmware driven menu (a ton of programming work) and begins to exceed CD size so would require a DVD (not legacy friendly) and a ton of work in general. beta1 diagnostics driven menu is flawed in that even tho my hardware runs all relevant AHT v1 with a single v1 diag & all relevant v2 with a single v2 diag, OTHER hardware exists where this is not the case and each AHT should utilize the diags which comes with them (requiring a firmware driven menu to load like that). But on the PowerMac Mirror Door 2003 I still want TWELVE versions of AHT , so the very easy diagnostics-driven AHT-AIO-CD/image will live on as a more completed looking version 0.2 made for PowerMac-only. It'll have this:

A) AHT v2.1.0 on G4 (ignores hw id re G4/G5)
B) AHT v2.1.0 on G5 (ignores hw id re G4/G5)
C) AHT v2.0.2 on G4 (ignores hw id re G4/G5)
D) AHT v2.0.2 on G5 (ignores hw id re G4/G5)
E) AHT v2.0.1
F) AHT v2.0.1 (edited to be RAM-friendly)
G) AHT v2.0.0
H) AHT v2.0.0 (edited to be RAM-friendly)
J) AHT v1.2.7
K) AHT v1.2.7 (edited to be RAM-friendly)
L) AHT v1.2.6
M) AHT v1.2.6 (edited to be RAM-friendly)

The preferred AHT for MDD2003 will be choice A and gotten without "wiggling the mouse'. AHT is offered way back to v1.2.6 because, never forget, up until only a couple years ago, from 2003 to 2016, the internet was telling people that v1.2.6 was the correct version for ALL "Mirrored Drive Doors" systems, AHT v2.0.2 was nowhere to be found because it was married to a 4GB disc image, AHT v2.1.0 was nowhere because no one would have suggested running G5 AHT on the G4-MDD.

So, NOT having to partition, install separately, or burn TWELVE aht versions, but instead have all 12 at one's fingertips with one partition... sweet!

SkyCapt's picture
Joined: 2017 Jan 11

New AHT-AIO for PowerMac (G4) is almost finished, just documentation and more testing to do, see it in about one week. It is stinking awesome. It has 7 AHTs:

A) 2.1 for G4
B) 2.1 for G5
C) 2.0.2
D) 2.0.0
E) 2.0.1
F) 1.2.6
G) 1.2.7

All seven can run on any hardware and all seven contain numerous bug fixes to the AHT itself. The nastiest "bug" (Apple did it deliberately but disguised it as an honest mistake) bug exists in all AHT v1.2 thru v2.1+ for all hardware, this "bug" displays the error message "invalid memory access" without really performing AHT. Because of the seriousness of this bug fix, I've decided to not even include the choice of running unmodified-AHT from the AIO image, even tho unmodified-AHT can be found inside the AIO image. Also, the idea of AHT v2.0.2 for G5 was dropped because i found it has no meaning, the G5 test suite file found on apple's G4 v2.0.2 disc is just an older version of the one on the AHT v2.1 and this older version G5 test requires AHT v2.1 regardless.

When AHT crashes saying "invalid memory access", it's a lie if you believe you have a bad DIMM. This is what AHT does when it encounters a DIMM with "unsupported (within AHT)" SPuD codes. The SPuD is the tiny (less than 5mm X 5mm square) chip in one corner of the DIMM. I have (and use) DIMMs which crash AHT v1.2 thru v2.0.1 but don't crash v2.0.2/v2.1 ... but 2.0.2/2.1 don't contain the bug fix rather they contain "support" for more variety module types. THIS below is AHTs lawsuit-inviting bug, and its fix, wrong with All AHT sourcecode:

AHT bug
Search the AHT_script file for ascii string "+20"
It's always found 3 times and all 3 lines are in error

AHT bug fix. (I will upload and notify)
1. "strcpy( s2+20 , s2 );" should be "strncpy( s2+20 , s2 ,20);. // patched"
2. "strcpy( s3+20 , s3 );" should be "strncpy( s3+20 , s3 ,20);. // patched"
3. "strcpy( s4+20 , s4 );" should be "strncpy( s4+20 , s4 ,20);. // patched"

Now instead of crashing when any "unsupported" DIMM is met, AHT runs and then profile reports that particular DIMM is "unsupported". Finally, it turns out Every AHT 1.2.6-2.1 reports my G4's 3GB RAM correctly (in red color text).

And oh yeah, when I mentioned above that internet has always said AHT v1.2.6 was good for the MDD2003 model, I forgot to emphasize how horribly wrong they are. They've been doing it because they keep playing out the false narrative that MDD2003 is identical to MDD2002 and AHT v1.2.6 is what shipped with MDD2002. MDD2003 came with AHT v2.0.2 and on MDD2003 AHT v2.1 is even better.

SkyCapt's picture
Joined: 2017 Jan 11

AHT for Power Mac, All-In-One.

AND it works well in conjunction with the Mirror Door model sporting the G5 hardware id. Don't have to change id and change back again just to run AHT.

Nobody's asked how to make the hw id change semipermanent. Y'all can verify it all already when you type the id change into firmware ahead of each reboot, but that's a heck a lot of typing all the time - and I don't believe y'all know enough about firmware to do it semipermanently yourselves.

SkyCapt's picture
Joined: 2017 Jan 11

I have achieved this 'insane' acceleration in OS9. With 3GB RAM and the G5 hardware id, OS 9.2.1 did not speed up because it's too old (year 2001, long before G5 existed). Even the initial OS 9.2.2 update doesn't speed up, it's also too old (November 2001).

To accelerate OS9, I take my best OS 9.2.1 volume, and add only the System file 9.2.2 taken from MDD2002 or after. I also must have the Extension "QuickDraw 3D Rave" version 1.8.1 (year 2002). And I find QuickDraw3D v1.6 a little better than QD3D v1.5.4

Without the G5+RAM trick, System file 9.2.1 had been faster than 9.2.2 and QD3D 1.5.4 had been faster than QD3D 1.6 ~ whereas now I see the opposite. QuickTime v6.0.0 remains faster than v6.0.3 and other OS 9.2.1 modules remain faster than a complete OS 9.2.2 install. Nevertheless I'm closer to being on the same page as the rest of you, using System file 9.2.2 and QD3D 1.6 with massive video acceleration.

This confirms something I've always suspected, that "System 9.2.2" comes in many versions. If the November 2001 release is 9.2.2 and doesn't have G5 acceleration, then maybe the 9.2.2 with G5 acceleration ought to have been OS 9.2.3 or OS 9.3

My MDD2003 computer is a G5 system (with just a G4 CPU) and I boot OS9 natively. !! What I got now really is OS9 running on G5 hardware, regardless that I don't have 3D drivers for it !

Don't everybody rush to comment at once...

greystash's picture
Joined: 2009 Oct 19

That's fantastic!! Great results! Smile I've been mucking around with video drivers trying to get unsupported chipsets working in OS9 but haven't had any luck. I have OS9 Booting natively on a PowerBook G4 17" 1.33Ghz but there's no acceleration since it's an unsupported model. Need to dig in a bit more...

SkyCapt's picture
Joined: 2017 Jan 11

greets. Bear in mind, 2D vs 3D acceleration are handled differently. The graphics card drivers are necessary for OpenGL flavored 3D acceleration plus needed for DVD Player, however, some or much of 2D acceleration (due to one "unsupported" card being faster than another) 2D accel can occur naturally and without any (additional) drivers.

This is awesome, for me, in OS9 without drivers for my Radeon 9600. I play this massive 3D game "Tomb Raider III" in OS9 without any GPU. When I play using the 9.2.2 System file from 9.2.2's debut in 2001, I receive very choppy TR3 game like about 10 fps. Now when I play using the MDD's own 9.2.2 System file, combined with my 3rd GB of RAM in my G4: looks like it basically doubles in speed to around 20 fps, both cases game resolution set to 1300ish X 844ish on my 1920x1200 display monitor.

SkyCapt's picture
Joined: 2017 Jan 11

Still going strong. Looks like this is permanent changes to my mdd2003 system. Firmware model id set to G5 (PowerMac7,2) because that's what I really have, an alternate version of the 7,2 debut G5. RAM greater in capacity than 2GB. AHT software, made compatible, has been linked above.

AHT version 2.2.5 will soon be provided as edited for PMG4. It will be the highest version AHT for mdd2003. Apple programmed it to state the speed of all mdd2003 RAM as being "unknown". You get less amount of useful information however this change, in the highest version of AHT, might be for the best, because the speed of all mdd2003 RAM really is "unknown"!, the speeds stated in older versions of AHT aren't necessarily the real speed. AHT v2.2.5 was programmed to describe PC2100 RAM as is (because no mdd2003 had PC2100) this means AHT v2.2.5 describes older RAM dimms with declared speed, newer PowerMac G5 RAM dimms with declared speed, but uniquely the mdd2003 is always RAM speed=unknown according to v2.2.5 their highest version of AHT. So it is, ram speed = unknown to AHT (& MacOS (& Firmware)).

OS9 boots with no sound capability, whenever there is one or more 1GB dimm installed. By 'wiggling' the sound output volume slider, actually thru having the Sound control panel in my OS9 Startup Items, sound capability becomes restored. I own both 400mhz & 467mhz 1GB dimms, and both 400&467 mhz <1GB dimms, so I confirm, sound output failure is not dimm speed related, it is tied to dimm capacity ... = 1GB dimm size brings on the OS9 sound level fail.

I located a 467 MHz dimm that has only 256MB capacity, ram chips on only one side of this dimm. I get the same twofold 'insane' graphics acceleration, when I create my ram bank to be 2.25GB total. Uniform speed increase with only 256MB excess ram suggests the excess ram really is substitute vram how I've proposed in this o.p. It's interesting to note, the 256MB dimm only accelerates graphics when it is used in the third sequential ramslot, when I have 1GB+1GB+256MB dimms. If I swap dimms so it's 256MB+1GB+1GB or 1GB+256MB+1GB then mdd2003's firmware/post system cuts the 3rd dimm in half and declares total RAM is 1.75GB and RAM in excess of 2GB is voided.