Generic 32 bit kernel should be compiled with 64 GB memory support

I just upgraded my memory to 4GB and am very disappointed that the kernel only recognizes 3GB. With cheap memory and the growing popularity of virtual machines, I think it is important that Ubuntu support the larger memory sizes.

I am going to try the server version, but if it will not support restricted modules as suggested above, then I will not be able to use the nvidia driver and without that my generic kernel does not support ¨suspend to RAM¨.

This is a very unsettling situation. This is a very basic hardware support issue.

Still no 4GB+ support in Gutsy.

There SHOULD be a desktop-version of the kernel supporting HIMEM64G (64 GB memory).

Sorry, but this is not technically possible - if it was it would have been implemented a long time ago. The only way to get 4GB ram support is to use a 64-bit OS.

Not at all impossible.

The kernelconfig is:

HIGMEM64G
and
HIGHMEM

Ubuntu Server is compiled with both these flags but not the desktop version.

Using more than 4 GB ram has NOTHING to do with 32 vs 63 bit OS. Sure, one process can use more than 4 GB ram but that is no big issue (yet).

Here is it in Brainstorm:

http://brainstorm.ubuntu.com/idea/1553/

Any comment?

Another one:

http://brainstorm.ubuntu.com/idea/2828/

And another one:

http://brainstorm.ubuntu.com/idea/2516/

And:

http://brainstorm.ubuntu.com/idea/2238/

This bug seems to be a duplicate of bug #156804.

More like bug #156804 is duplicate of this.

Compiling kernel with PAE support indeed gives the 32-bit kernel ability to use 64 gigabytes of memory. However, single processes can still use only max 3 gigabytes. There will be 3-6% performance hit for memory operations. The worst part is that non-PAE supporting hardware will fail booting with the kernel. That is a blocker for making PAE kernels default.

In any case, if you need more, switch to 64-bit platform. You can still run 32-bit applications there if you must by using compatibility libraries. They do work.

>non-PAE supporting hardware will fail booting with the kernel

Ouch, I admit that hurts.

Especially for non-x64-hardware where you have no "workaround" in switching to the amd64 distro.

I guest PAE cannot be "loaded" lateron or switched active with a kernel line parameter?
So what's left is either to eventually provide a some "generic-bigmem" kernel (and all the accompaning packages, especially several module packages for GPUs, WLAN cards, ...).

How about a PAE-enable kernel as option after installation?

The bug sure isn't invalid.

>How about a PAE-enable kernel as option after installation?
>The bug sure isn't invalid.

It exists already. It is the server kernel.

>>How about a PAE-enable kernel as option after installation?
>>The bug sure isn't invalid.

>It exists already. It is the server kernel.

At the time we came up with this issue (feisty), several module packages had not been available for the -server kernel, therefore some things (especially GPUs and WLAN cards) could not have been run easily on a server kernel ;-((

This did not change with gutsy, but with hardy seems to have changed.
Was there an official change in the policies regarding e.g. the linux-restricted-modules and kernel packages?

Anyway, did anybody actually try to use e.g. nvidia-glx-*, fglrx*, ... together with hardy's server kernel?

Does the server kernel has any other differences like timeslice length etc?

Extra special needs non-corforming with the baseline users - instructions about the advanced situations for the newbies please.

So why is it in Debian 4 (etch) i can do this

apt-get install linux-image-bigmem.

And also in older versions of Ubuntu.
Why is now that we are forced to use a 64bit kernel with workstation workloads and wrestle with
flash ans skype and all those things.. Why not at least put a linux-image-bigmem kernel like
Debian does and older versions of Ubuntu had. ? Why? Is there any reason to refuse to
provide a linux-image-bigmem kernel in the repositories?

Compiz does not work with the server kernel and install nvidia drivers from the website complains about
installing the drivers on a xen kernel even though I installed linux-image-server. What gives?