Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Don't quit your day job noobie. I've owned and serviced Macs since 1988.

Wow, Im spectacularly unimpressed by your response, it does appear rather infantile. If you understood how these machines worked at a deep level instead of just how to add RAM into a box, you would understand that I need a G5 for its massively parallel inflight instructions, and would have obviously tried moving them onto a newer system if it was feasible, which it evidently isnt (No G5 support in Rosetta). Although, judging by your response you didnt know what an inflight instruction was, as the best you can come up in reponse with is hilighting a part of my post where I was pointing out that you dont come across as someone with much to add to the conversation apart from telling me what I do and don't need inside a Machine. Why one would want 32GB in a G5 is not really the point, if someone has tried it and got it working and it works fully and I state an interest in wanting to know if it does actually work, it means I have a reason for wanting to know.

Also, you can have been servicing Macs since the dawn of time for all I care, you still could easily know less than me. Mainly as you dont appear to understand what is actually happening below the "Thats the CPU, Thats the RAM, Those are cards" level, otherwise you would understand there are still a lot of applications which are G5 only, especially in the scientific/medical field, and a lot of them can use 2% CPU power, but need as much RAM as you can fill a machine with, or at least, if you do know these things, your not letting on in your posts. Its very common to have applications in many fields which are tied to having a lot of RAM for large datasets to build models/images with, but the model once built may only consume 2-3% of the CPU to actually run.
 
Wow, Im spectacularly unimpressed by your response, it does appear rather infantile. If you understood how these machines worked at a deep level instead of just how to add RAM into a box, you would understand that I need a G5 for its massively parallel inflight instructions, and would have obviously tried moving them onto a newer system if it was feasible, which it evidently isnt (No G5 support in Rosetta). Although, judging by your response you didnt know what an inflight instruction was, as the best you can come up in reponse with is hilighting a part of my post where I was pointing out that you dont come across as someone with much to add to the conversation apart from telling me what I do and don't need inside a Machine. Why one would want 32GB in a G5 is not really the point, if someone has tried it and got it working and it works fully and I state an interest in wanting to know if it does actually work, it means I have a reason for wanting to know.

Also, you can have been servicing Macs since the dawn of time for all I care, you still could easily know less than me. Mainly as you dont appear to understand what is actually happening below the "Thats the CPU, Thats the RAM, Those are cards" level, otherwise you would understand there are still a lot of applications which are G5 only, especially in the scientific/medical field, and a lot of them can use 2% CPU power, but need as much RAM as you can fill a machine with, or at least, if you do know these things, your not letting on in your posts. Its very common to have applications in many fields which are tied to having a lot of RAM for large datasets to build models/images with, but the model once built may only consume 2-3% of the CPU to actually run.

Ahh, the G5 and its 216 in-flight instructions. Sorta too bad this was never really leveraged very much during the G5's lifetime :/

I actually just bought a Quad G5, but it only shipped with 1.5 GB of RAM.. a little light in the loafers. Obviously I don't need 32 GB right now (I do okay with 768 MB on my upgraded Beige G3 w/1 GHz cpu and 3 HDDs.... generally 0 page-outs), but I'd like to know I could. I have 4 GB on my Lenovo and occasionally run out (albeit only 3GB is used >_>)

So I suppose this is still unresolved.. can the G5 take advantage of 32 GB of RAM?

It looks like it's at least capable of addressing 4 GB modules... so I suppose the memory controller is capable of reading 4 Gbit chips. Whether or not it can address >16 GB still remains to be seen, however.

The Beige G3 I use for example is technically capable of addressing up to 1 GB of RAM, but it only has 3 memory slots and its memory controller (Grackle; MPC106) can't address RAM chips of greater than 256 Mbit density (256MB sticks). So 768 is the limit.
 
Why do people continue to revive year old threads just to bump?

Probably to get a response like the one from Intell to a question they still have, not insipid, pointless and unhelpful responses like the one you've just given. :rolleyes:

Thanks, Intell.
 
  • Like
Reactions: Jubadub
But this is a curiosity, is it a hard or soft limit? If the limit is soft as are some of the kernel limits in newer intel Macs then installing Linux should address the issue. If its a hard limit of the memory controller then I accept, no it can not.
 
  • Like
Reactions: Jubadub
But this is a curiosity, is it a hard or soft limit? If the limit is soft as are some of the kernel limits in newer intel Macs then installing Linux should address the issue. If its a hard limit of the memory controller then I accept, no it can not.

It's a memory controller limit. Leopard can address up to 96GB of RAM in Mac Pros.
 
It's a memory controller limit. Leopard can address up to 96GB of RAM in Mac Pros.

Yeah, that's unfortunate. I imagine it's the memory controller's inability to address chips of >2 Gbit density.

They really screwed up with the MC in the G5... latency was horrid.
 
Yeah, that's unfortunate. I imagine it's the memory controller's inability to address chips of >2 Gbit density.

They really screwed up with the MC in the G5... latency was horrid.

I firmly believe they screwed up a lot of things with the G5.
 
They did, or IBM did? On paper the G5 should be as fast as a Core 2 chip but IBM couldn't bring the production costs down so manufacturing was affordable, or dye shrink in order to make them run cooler/run more of them on the same chip, or get them to run at higher clock speeds. IBM proved again why its not a high quantity chip manufacturer in the desktop market against Intel and AMD as it did with it's x86 processors bought by Cyrix later bought and currently manufactured by Via. The Power5 (G5) IS a fast CPU, the high end Power5 CPUs rates as fast as a low end Intel Core 2 CPU and are faster then intels P4s for most tasks, but IBM could never scale them up in a way that Apple was happy with, so Apple again fell behind in the CPU clock war.

In the end Apple saw all the writing on the wall particularly after trying to deal with Motorolla and losing the CPU clock race first to AMD and then to Intel before Intel overtook AMD again and left Apple even further behind.

They, thanked the open source community for writing them a kernel that would do everything on intel chips for free and then took it for themselves. A great play by Apple with the Open Source community, but it managed to royally piss everyone off in the Open Source community who ended up developing for Apple's playtoys. Now, Open Darwin is effectively dead and a lot more Open Source users have an intense dislike for Apple.

It was always Apples long held hope that IBM would get Apple back into the CPU clock race, but ironically the move to Intel CPUs has been the best thing to happen to Apple since Steve Jobs returned. Thinking differently with RISC CPUs really hurt Apple in the long run, where the transition to CISC hasn't led to any extra substantial piracy or drops in sales figures and with the benefits of the first Athlon 64 and a new CPU architecture from that point onwards we all get to benefit from better and better CPUs.
 
Last edited:
They did, or IBM did?

They did, I think.. the memory controller was their design AFAIK.

On paper the G5 should be as fast as a Core 2 chip
Considerably faster, actually.

but IBM couldn't bring the production costs down so manufacturing was affordable, or die shrink in order to make them run cooler/run more of them on the same chip, or get them to run at higher clock speeds.

Well, to be fair, 90nm was current at the time.

The Power5 (G5)

It's based on the POWER4.

IS a fast CPU, the high end Power5 CPUs rates as fast as a low end Intel Core 2 CPU

Shame we never got to see the POWER5 derivatives.. they were clocking at 4.7 GHz right out of the first batch, and were faster per clock. And forget about the POWER6. I Really wish Apple had held onto the PowerPC a bit longer.... the 5 would have blown the C2D out of the water, and the 6 would have obliterated the Core i-Series.

and are faster then intels P4s for most tasks
All tasks.


In the end Apple saw all the writing on the wall particularly after trying to deal with Motorolla and losing the CPU clock race first to AMD and then to Intel before Intel overtook AMD again and left Apple even further behind.


Thinking differently with RISC CPUs really hurt Apple in the long run, where the transition to CISC hasn't led to any extra substantial piracy or drops in sales figures and with the benefits of the first Athlon 64 and a new CPU architecture from that point onwards we all get to benefit from better and better CPUs.

RISC/CISC is basically meaningless now.. didn't mean much by the time we had the G4/G4+.

What does moving to an x86 architecture have to do with piracy? Or the Athlon 64 have to do with anything...?

----------

I firmly believe they screwed up a lot of things with the G5.

Unfortunately.
 
It's based on the POWER4.

My bad, that's what I meant to say, just a little rusty after all these years.

What does moving to an x86 architecture have to do with piracy? Or the Athlon 64 have to do with anything...?

AMD were the first to mass produce the new CPU arch we're using right now in intel Macs, that is why it's called AMD64, even if the chip is made by Intel, or Via the current 64bit CPUs being produced by these companies are all AMD64 which has largely replaced x86 which hasn't really been relevant since 686 CPUs. As opposed to IA64 (Intel Itanium 64) the other 64bit CPU standard. AMD won, so AMD has naming rights...

As for piracy, there was speculation that moving to x86 and AMD64 now would lead to a great amount of clone boxes and illegal installations of OS X, neither of which have occured I suspect greatly because of closed source drivers and lack of hardware support as compared to Windows.

Unfortunately.

I wish Apple had have stuck with Power chips as well, but we can only really acknowledge some of the screw ups that came with the G5. I've still got a G5 Xserve and I'll be keeping it for as long as it continues to serve my Apple TV 3.
 
Last edited:
It's a memory controller limit. Leopard can address up to 96GB of RAM in Mac Pros.

It seams that the northbridge can handle 64 GB of RAM.

From wikipedia:

Northbridges

There are two dedicated northbridges for PowerPC 970-based computers, both manufactured by IBM:
CPC925 — Designed by Apple[10] and called the U3 or the U3H (which supports ECC memory). It is capable of supporting up to two PowerPC 970s or PowerPC 970FXs and has two 550 MHz unidirectional processor buses, a 400 MHz DDR memory controller, x8 AGP and a 400 MHz 16-bit HyperTransport tunnel. It fabricated on a 130 nm process.
CPC945 — Designed by IBM and called U4 by Apple, it is capable of supporting two PowerPC 970MPs and has two 625 MHz unidirectional processor buses, two memory controllers that support up to 64 GB of 533 MHz DDR2 SDRAM with ECC capability and has a x16 PCIe lane and an 800 MHz 16-bit HyperTransport tunnel. It is fabricated on a 90 nm process.
There was also a cancelled CPC965 northbridge. Slated for release in 2007, it was to be a uniprocessor-only northbridge. Its features were a 533 MHz DDR2 controller that supported up to 8 GB ECC memory, a 8x PCIe bus, integrated four-port Gigabit Ethernet with IPv4 TCP/UDP offloading, USB 2.0 ports, a Flash-interface. The northbridge contains an integrated PowerPC 405 core to provide system management and configuration capabilities.[8]

Is this not the case?
 
It seams that the northbridge can handle 64 GB of RAM.

From wikipedia:

Northbridges

There are two dedicated northbridges for PowerPC 970-based computers, both manufactured by IBM:
CPC925 — Designed by Apple[10] and called the U3 or the U3H (which supports ECC memory). It is capable of supporting up to two PowerPC 970s or PowerPC 970FXs and has two 550 MHz unidirectional processor buses, a 400 MHz DDR memory controller, x8 AGP and a 400 MHz 16-bit HyperTransport tunnel. It fabricated on a 130 nm process.
CPC945 — Designed by IBM and called U4 by Apple, it is capable of supporting two PowerPC 970MPs and has two 625 MHz unidirectional processor buses, two memory controllers that support up to 64 GB of 533 MHz DDR2 SDRAM with ECC capability and has a x16 PCIe lane and an 800 MHz 16-bit HyperTransport tunnel. It is fabricated on a 90 nm process.
There was also a cancelled CPC965 northbridge. Slated for release in 2007, it was to be a uniprocessor-only northbridge. Its features were a 533 MHz DDR2 controller that supported up to 8 GB ECC memory, a 8x PCIe bus, integrated four-port Gigabit Ethernet with IPv4 TCP/UDP offloading, USB 2.0 ports, a Flash-interface. The northbridge contains an integrated PowerPC 405 core to provide system management and configuration capabilities.[8]

Is this not the case?


dude I wanna make a show about you called leave it to rabidz :p
 
Wishful thinking

Those wishing our PowerPC G5 Quads can handle 32GB this is just a pipe dream.. While the G5 architecture supports up to 64GB of memory, Apple only limited it to support 16GB and this thread is quite old to be honest.

The Intel Macs, like the Mac Pro 1,1 only support 16GB according to Apple, but the memory controller is by Intel and the maximum memory Woodcrest can support is 32GB undisclosed through 3rd party memory companies.

I too wish the G5 Quad and Dual-Core machines could support 32GB, but a total rewrite of the memory controller would need to be done and that would take a lot of time and patience + disassembling the thousands and thousands of assembly code and machine code just to get it to work and be recognized.

For most tasks today, I would think 16GB is sufficient.
 
Those wishing our PowerPC G5 Quads can handle 32GB this is just a pipe dream.. While the G5 architecture supports up to 64GB of memory, Apple only limited it to support 16GB and this thread is quite old to be honest.

The Intel Macs, like the Mac Pro 1,1 only support 16GB according to Apple, but the memory controller is by Intel and the maximum memory Woodcrest can support is 32GB undisclosed through 3rd party memory companies.

I too wish the G5 Quad and Dual-Core machines could support 32GB, but a total rewrite of the memory controller would need to be done and that would take a lot of time and patience + disassembling the thousands and thousands of assembly code and machine code just to get it to work and be recognized.

For most tasks today, I would think 16GB is sufficient.

16GB is more than enough.
 
I have yet to find 4Gb modules that are recognised by the G5. Tried Crucial and Hynix 4Gb UDIMM pairs rated for Intel - nothing happens, not detected though they work fine in Intel mobos. Tried setting the address length to 0x800 in ReggieSE - to no avail. 16Gb is likely the hard limit.
 
I have yet to find 4Gb modules that are recognised by the G5. Tried Crucial and Hynix 4Gb UDIMM pairs rated for Intel - nothing happens, not detected though they work fine in Intel mobos. Tried setting the address length to 0x800 in ReggieSE - to no avail. 16Gb is likely the hard limit.
Hey that's not true.


(Do note that the above command is for my configuration - 2x1GB, 6x4GB - if you want something different you'll have to adjust it)

You have to check first if the module numbers are recognised though.

This is an OpenFirmware hack btw.

1667527360582.png


It's not fully 32 because it needs at least 2 sticks of not 4 GB to boot the system (I have 2x1 - but you can in theory put 2x2 as well).

However the speed may be low (I'm not sure - they are 266 for me for the 4GB ones). You can check this info with:

dev /memory

.properties

It all came from here btw:


It's a 64 bit offset - then 32 bit size. You basically need to query it first from .properties and add up the additional sticks.

You want to keep the offsets aligned though and and the size at maximum 0xF0000000 (you are losing like 0x10000000 bytes for each 4GB stick).

If you don't align them akin to my post - the system will crash after peeking past the original low capacity x2 sticks.

And overall I suggest learning Forth:


(But you can just use my command as well - I currently have an ramsc script in boot-command)

\ comment

Your extend ram command

boot sd0:,\\:tbxi



So it extends the ram and then it boots from whatever.

Your boot-command should be:

setenv boot-command boot sd0:,\ramsc

Where ramsc is the file you have put your script in.

NVRAMRC - doesn't work here. You need to do it after that.

And also if you have linux (and use unsupported OS X GPU) I suggest making 2 scripts - one without the boot in the end so you can easily source yourself and boot other OS from the OpenFirmware terminal if needed.

If you have any questions - do please ask but as I've stated in the beginning first post your

" dimm-info" get-active-property

.s

(Check the address here)

ffa322b8 400 dump

(Where ffa322b8 is from above)

This will tell you the ram models - check if they are correct.


TL;DR:

You boot from 2x <4GB sticks (on first 2 slots - which are the one closest to each other on both sides in the center) (hope that succeeds and you don't hear the telephone beep - you may need to try multiple times then - or buy Samsung sticks (or the same on as I have because they worked for me - models here (in the hex dump)- https://forums.macrumors.com/threads/overclocking-the-g5-cpu.2363051/post-31622390 ))

Rest of the sticks are 4GB sticks.

You then have (preferably) a boot-command that runs a script on your disk which extend the registered ram and boot OS.

Then you (preferably) start compiling Webkit on all cores on Gentoo and test if it works (if it hangs and crashes (after you have exceeded your original low capacity sticks size) you probably haven't aligned the offsets and sizes).
 
Last edited:
So why you need more than 16GB you say - well running `make -j32` on gcc just ran out of memory:

1667538781880.png


Which might have actually been for the test (because I only have 4 cores but I wanted to see if I'll ran out of memory).
 
  • Like
Reactions: Amethyst1
That is true for the Sawtooth, Gigabit Ethernet and MDD. The Digital Audio and Quicksilver can only hold 1.5GB. The Yikes G4 can only hold 1GB.

I will never understand in a million years why Apple allowed a 2GB limit in the first 2 G4 towers with AGP and then limited the next 2 models to 1.5GB. This is the main reason I run a Sawtooth. In my experience the extra 512MB (which is 33% more) outweighs any advantage the slightly faster bus offers.
the really strange thing is when you think about the architecture of the memory controller, because 2gb max would mean that it is a 31-bit memory controller (32-bit can address 4gb)
and 1gb max would be a 30-bit memory controller
so where do they get 1.5gb from? you cant have half a bit
which makes me think that 1.5gb is not the actual max, rather is it the fact that it has 3 ram slots, so you would need 3x666.66666gb sticks to get 2gb (which dont exist)

31-bit is also a very strange choice, why they didnt just do 32-bit like every other PC system at the time, i do not know.
 
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.