The Switch memory expansion is probably much more useful than this, for general users. No standing software is able to utilize this memory expansion, only homebrew software specifically coded for it can.
In contrast, the Switch OS manages memory for apps and would be able to allocate additional memory automatically, as long as it's been made aware.
Also, if my YouTube recommended is any indication there's a healthy Linux ecosystem for the Switch and the X1 now - some people have even shown Switch games playing on Switch via emulator in Ubuntu [0]
That's because nVidia has always had a decent support level for their Tegra chips via the Jetson program, the X1 being one of the better utilized ones.
Correct, there are some applications that can take advantage because they were built only on the devkit or for the arcade configuration. I just meant to say standard Xbox software and a good chunk of homebrew will not benefit from the expanded memory.
I also believe (but could be misinformed) that the arcade configuration and Devkit had an even larger memory space, so you might still have issues in that regard.
The dev kits had 128MB of RAM. It's been possible for decades to upgrade the Xbox to 128MB of RAM by just populating the empty RAM using the same modules as the rest of the system. It allowed booting Chihiro games, and dev builds. But this hack replaces all the original modules with ones sporting twice the capacity to allow 256MB.
Are you kidding? It's a Bethesda game, of course it won't fix any bugs. It's 2024 and the exact same sort of issues exist in Skyrim and Fallout 4 (even after the remasters/anniversary editions)!
I've run into issues recently (in June) with Fallout 4 where if your saved game data starts approaching 2GB on Xbox One x, saving the game will fail in weird ways (may hang the game, may crash the game). Once the game gets in this state, I'm not even able to launch the game without it hanging. Have to delete local game data, resync your saves from Live and pray you didn't lose more than a few minutes of gameplay.
I mean in this case, it's not a bug, it's a feature. The original Xbox had a feature where you could reboot and directly launch a binary while an image remained in the framebuffer.
Throw up a loading screen and secretly reboot.
Morrowind used this to silently reboot the Xbox to free memory.
I remember buying Skyrim VR, only for it to be unplayable without modding because the physics is tied to the FPS and anything above 60 Hz would send objects flying everywhere (including the carriage in the intro sequence).
Sounds like a Bethesda feature to me. Morrowind was a fun game in its own right, but the outright janky stuff you could do was the main reason I loved it.
Modders have done some incredible work optimizing and bug-fixing Morrowind over the years. I'm curious if someone could mash all of that together and 're-pack' an Xbox version of the game that ran way smoother.
Probably not, at least not without hacking the game. Typically only things which were built to use the extra RAM see any benefit, whether that's homebrew, leaked game betas which rely on the extra RAM that devkits had, or games made for the Sega Chihiro arcade platform which was basically an Xbox but with double the RAM.
There was a great article here a few months ago on this very subject, detailing getting Halo 2 to run in true HD, properly using the extra RAM of the devkit, and the engine hacks required to do so (since it basically statically allocated a handful of regions and couldn't dynamically go "oh look more RAM!"). It was a good read!
Since it was mentioned in the Macho Nacho video, this was quite a hurdle. But it does bring up the question, what is the theoretical maximum possible? I know the CPU was 32-bit, so 4GiB is the upper limit in that regard. I don't know if the CPU has limited address lanes or any other physical limitation though. The max I've seen discussed is 1.5GB for a similar cpu on a normal desktop and 640MB for the test design of the xbox (the literal pc version, not the dev kit).
I can't figure out if the xbox processor supports PAE or not. But references say most Intel processors after pentium pro, including Pentium III support PAE. Probably with 36 bits of physical addressing. But it's a mobile socket; no idea if they put all the address pins out on those.
OT: Like you mentioned PAE was introduced with the Pentium Pro in '95. That was the same year my family got its first PC, with a whopping 8MB of RAM. It's wild to think that Intel was looking at 4GB and saying, "This probably isn't going to be enough."
>It's wild to think that Intel was looking at 4GB and saying, "This probably isn't going to be enough."
Not really. While consumers at home had 8MB, Intel was already looking to compete with PowerPC, DEC Alpha, MIPS systems where 4GB wasn't really out of the ordinary, so they knew where things were going.
As far as I know, there were only 5 PPro computers that could take >4GB:
Axil Northbridge NX801, Data General AV8600, HP NetServer LXr Pro8, NCR Worldmark 4380, or Unisys XR/6
Interestingly, while the Intel 450GX chipset was designed to daisy-chain DRAM controllers (1 controller chip per 4GB of RAM), none of the boards that actually could take >4GB used that, instead building their own solutions. Presumably that meant that there was a bug in the chipset.
Anyway, each of those 5 computers would have run you at least $500k in 1998.
(I've been on-and-off trying to acquire one of those boxes for a few years now.)
Where I worked back in 1995, we had DEC Alpha workstations running Windows NT processing what were then huge images. We had DEC workstations running 4 GB of RAM.
I think some Alpha systems had 8 GB installed even back then.
The Alpha was 64 bit native but Windows NT was still 32 bit so I think you needed VMS if you wanted more than 4 GB. A 64 bit version of Windows 2000 for Alpha was created but never released ( this was the first 64 bit Windows ).
It was not totally impossible to see beyond 4 GB of RAM.
at the time the original xbox was designed and its same generation of intel CPUs, the absolute practical maximum you would be able to put into a single socket desktop motherboard would be 768MB of PC100 or PC133 SDRAM (3 x 256MB DIMMs). If you actually saw a system with 768MB of RAM in it, that would be some very high-end CAD workstation.
There was no practical way to get anywhere near 4GB of RAM within the dollar cost that a person purchasing a $200 CPU on a $150 motherboard would be able to bear.
As the pentium 3 used the chipset for the memory controller, it's likely limited by that, though you'll likely struggle to find specs at that sort of level publicly available. Though assuming it's the same as contemporary consumer nForce chipsets may be a good assumption [0], which may be where the 1.5gb limit discussed came from.
And the p3 also supported PAE [1], so technically could address more than 4gb of ram. Though software support may be lacking - I think on the original xbox the app ran in ring0 so did things like page table management itself, rather than the "OS", so would likely require significant per-game modifications.
True, I didn't know PAE was a thing till now. I figured something like that was possible, I just didn't know we had it around the time of the xbox. I guess that raises the current theoretical limit to around 64GiB? And yeah even the chip presented in the video was also noted that even 256mb wouldn't provide any benefits to games, only to homebrew or game mods and even that was overkill for anything that currently exists.
Thanks for the info, I'm going to look more into it later and see if I can come up with anything more concrete.
In theory the chipset max is 2GB, but the physical motherboard does not have enough address lines traced to support more than 256MB. Adding more memory would require a PCB redesign
> I know the CPU was 32-bit, so 4GiB is the upper limit in that regard.
Not necessarily. Just because a CPU has a 32-bit virtual address space doesn't mean it has a 32-bit physical address space. PAE (Physical Address Extension) is a paging mode for x86 CPUs that offers a 36-bit physical address space on 32-bit CPUs.
Amazing that this is possible. Also saw a video of upgrading a Nintendo Switch to 8GB of RAM https://www.youtube.com/watch?v=J4oXmQGZgzU
The Switch memory expansion is probably much more useful than this, for general users. No standing software is able to utilize this memory expansion, only homebrew software specifically coded for it can.
In contrast, the Switch OS manages memory for apps and would be able to allocate additional memory automatically, as long as it's been made aware.
Also, if my YouTube recommended is any indication there's a healthy Linux ecosystem for the Switch and the X1 now - some people have even shown Switch games playing on Switch via emulator in Ubuntu [0]
[0] https://youtu.be/H1gveQUBIKk?t=2332
That's because nVidia has always had a decent support level for their Tegra chips via the Jetson program, the X1 being one of the better utilized ones.
The Switch benefited greatly from that community.
I don't know for sure about 256MB, but some Xbox prototype games require a debug BIOS and the extra RAM in the debug kit.
Correct, there are some applications that can take advantage because they were built only on the devkit or for the arcade configuration. I just meant to say standard Xbox software and a good chunk of homebrew will not benefit from the expanded memory.
I also believe (but could be misinformed) that the arcade configuration and Devkit had an even larger memory space, so you might still have issues in that regard.
The dev kits had 128MB of RAM. It's been possible for decades to upgrade the Xbox to 128MB of RAM by just populating the empty RAM using the same modules as the rest of the system. It allowed booting Chihiro games, and dev builds. But this hack replaces all the original modules with ones sporting twice the capacity to allow 256MB.
Gotcha, for some reason I was under the impression that the Arcade kit or DevKit had an odd memory setup of 640mb or something like that.
Thanks for the correction!
Stock switch OS plays games faster / loads faster?
The real question is if this will stop Morrowind from secretly rebooting the XBOX at certain load screens.
Are you kidding? It's a Bethesda game, of course it won't fix any bugs. It's 2024 and the exact same sort of issues exist in Skyrim and Fallout 4 (even after the remasters/anniversary editions)!
I've run into issues recently (in June) with Fallout 4 where if your saved game data starts approaching 2GB on Xbox One x, saving the game will fail in weird ways (may hang the game, may crash the game). Once the game gets in this state, I'm not even able to launch the game without it hanging. Have to delete local game data, resync your saves from Live and pray you didn't lose more than a few minutes of gameplay.
I mean in this case, it's not a bug, it's a feature. The original Xbox had a feature where you could reboot and directly launch a binary while an image remained in the framebuffer.
Throw up a loading screen and secretly reboot.
Morrowind used this to silently reboot the Xbox to free memory.
https://www.youtube.com/watch?v=x0TKwPnHc-M
I remember buying Skyrim VR, only for it to be unplayable without modding because the physics is tied to the FPS and anything above 60 Hz would send objects flying everywhere (including the carriage in the intro sequence).
Sounds like a Bethesda feature to me. Morrowind was a fun game in its own right, but the outright janky stuff you could do was the main reason I loved it.
Modders have done some incredible work optimizing and bug-fixing Morrowind over the years. I'm curious if someone could mash all of that together and 're-pack' an Xbox version of the game that ran way smoother.
Completely pointless? Yes. Interesting? Also yes!
Probably not, at least not without hacking the game. Typically only things which were built to use the extra RAM see any benefit, whether that's homebrew, leaked game betas which rely on the extra RAM that devkits had, or games made for the Sega Chihiro arcade platform which was basically an Xbox but with double the RAM.
There was a great article here a few months ago on this very subject, detailing getting Halo 2 to run in true HD, properly using the extra RAM of the devkit, and the engine hacks required to do so (since it basically statically allocated a handful of regions and couldn't dynamically go "oh look more RAM!"). It was a good read!
https://kotaku.com/morrowind-completely-rebooted-your-xbox-d...
For those who don't know
Being Morrowind, this xbox is only about 2TB away from handling the memory leaks. ;)
Since it was mentioned in the Macho Nacho video, this was quite a hurdle. But it does bring up the question, what is the theoretical maximum possible? I know the CPU was 32-bit, so 4GiB is the upper limit in that regard. I don't know if the CPU has limited address lanes or any other physical limitation though. The max I've seen discussed is 1.5GB for a similar cpu on a normal desktop and 640MB for the test design of the xbox (the literal pc version, not the dev kit).
I can't figure out if the xbox processor supports PAE or not. But references say most Intel processors after pentium pro, including Pentium III support PAE. Probably with 36 bits of physical addressing. But it's a mobile socket; no idea if they put all the address pins out on those.
OT: Like you mentioned PAE was introduced with the Pentium Pro in '95. That was the same year my family got its first PC, with a whopping 8MB of RAM. It's wild to think that Intel was looking at 4GB and saying, "This probably isn't going to be enough."
>It's wild to think that Intel was looking at 4GB and saying, "This probably isn't going to be enough."
Not really. While consumers at home had 8MB, Intel was already looking to compete with PowerPC, DEC Alpha, MIPS systems where 4GB wasn't really out of the ordinary, so they knew where things were going.
As far as I know, there were only 5 PPro computers that could take >4GB: Axil Northbridge NX801, Data General AV8600, HP NetServer LXr Pro8, NCR Worldmark 4380, or Unisys XR/6
Interestingly, while the Intel 450GX chipset was designed to daisy-chain DRAM controllers (1 controller chip per 4GB of RAM), none of the boards that actually could take >4GB used that, instead building their own solutions. Presumably that meant that there was a bug in the chipset.
Anyway, each of those 5 computers would have run you at least $500k in 1998.
(I've been on-and-off trying to acquire one of those boxes for a few years now.)
Had fun googling those system names. NX801: https://www.tpc.org/results/individual_results/axil/axil.nx8...
200 9.1GB SCSI disks for 1.8TB!
And still only 4GB RAM on that SQL Server 6.5 box they used for TPC-C. Wild.
And yeah, $770k for that server.
Edit: I guess whilst I'm at it...
Data General AV8600: https://www.tpc.org/results/individual_results/dg/dg.8600.es...
HP NetServer LXr Pro8: https://www.1000bit.it/ad/bro/hp/netserverlxrpor8.pdf
Where I worked back in 1995, we had DEC Alpha workstations running Windows NT processing what were then huge images. We had DEC workstations running 4 GB of RAM.
I think some Alpha systems had 8 GB installed even back then.
The Alpha was 64 bit native but Windows NT was still 32 bit so I think you needed VMS if you wanted more than 4 GB. A 64 bit version of Windows 2000 for Alpha was created but never released ( this was the first 64 bit Windows ).
It was not totally impossible to see beyond 4 GB of RAM.
at the time the original xbox was designed and its same generation of intel CPUs, the absolute practical maximum you would be able to put into a single socket desktop motherboard would be 768MB of PC100 or PC133 SDRAM (3 x 256MB DIMMs). If you actually saw a system with 768MB of RAM in it, that would be some very high-end CAD workstation.
There was no practical way to get anywhere near 4GB of RAM within the dollar cost that a person purchasing a $200 CPU on a $150 motherboard would be able to bear.
As the pentium 3 used the chipset for the memory controller, it's likely limited by that, though you'll likely struggle to find specs at that sort of level publicly available. Though assuming it's the same as contemporary consumer nForce chipsets may be a good assumption [0], which may be where the 1.5gb limit discussed came from.
And the p3 also supported PAE [1], so technically could address more than 4gb of ram. Though software support may be lacking - I think on the original xbox the app ran in ring0 so did things like page table management itself, rather than the "OS", so would likely require significant per-game modifications.
[0] https://en.wikipedia.org/wiki/Comparison_of_Nvidia_nForce_ch...
[1] https://en.wikipedia.org/wiki/Physical_Address_Extension
True, I didn't know PAE was a thing till now. I figured something like that was possible, I just didn't know we had it around the time of the xbox. I guess that raises the current theoretical limit to around 64GiB? And yeah even the chip presented in the video was also noted that even 256mb wouldn't provide any benefits to games, only to homebrew or game mods and even that was overkill for anything that currently exists.
Thanks for the info, I'm going to look more into it later and see if I can come up with anything more concrete.
you'll likely struggle to find specs at that sort of level publicly available
Intel was far more open at that time (up through the Pentium IV era); here's the datasheet for the PIII:
https://download.intel.com/design/PentiumIII/datashts/245264...
...and yes, it does expose all 36 address bits on the FSB. The FSB protocol is actually the same as the one first used by the P6.
In theory the chipset max is 2GB, but the physical motherboard does not have enough address lines traced to support more than 256MB. Adding more memory would require a PCB redesign
... or one crazy rework.
> I know the CPU was 32-bit, so 4GiB is the upper limit in that regard.
Not necessarily. Just because a CPU has a 32-bit virtual address space doesn't mean it has a 32-bit physical address space. PAE (Physical Address Extension) is a paging mode for x86 CPUs that offers a 36-bit physical address space on 32-bit CPUs.
Neat! I've been loving the Xbox Arms Race recently.
100
[flagged]