Actually, my ISO of RC1 installs the PAE kernel by default, even if your CPU doesn’t support it (for example, my P4 2.6C Northwood doesn’t support PAE, and never has; there are, however, S478 CPUs from Intel that do support PAE based on the Prescott core). This leads me to believe that PAE is modular, not monolithic, which is a quite-sensible approach, because processor cores such as Northwood and Prescott are odd ducks even in terms of Linux, let alone Win32 (odd in the sense of their age, not reliability; Prescott last shipped in S478 trim in late 2005, and replaced Northwood-C). Contrariwise, there are few (if any) LGA775-based CPUs that don’t support PAE. (All Core-architecture CPUs support PAE, for example.)
Why didn’t I mention AMD in the PAE-stakes? Two reasons, actually:
I have not run AMD-powered anything since the days of the Original Athlon (and even then it was strictly a classroom PC, as opposed to one I owned or built personally).
I have even less (1.5 GB), and a processor that doesn’t even support PAE (P4 2.6 Northwood-C), which is why I have 2 GB of swap. Apparently, the effect on performance is niggling, or processor non-support of PAE is very low to non-existent in the testing pool (the latter makes all sorts of sense, as Intel adopted PAE with the Prescott core clear back in early 2005, and AMD had adopted it even earlier).
One of the kernel gurus should pitch in; however, I suspect that the PAE support, while present, is like the SMP/multicore support; it’s used when detected on startup by the kernel (modular, in other words). SMP/multicore support went modular starting with the 2.4 kernels (primarily because of HT’s mainstreaming by Intel). The current state of PAE support on the CPU front is wider than SMP/multicore support has ever been, even in the heyday of HT (for example, there are lots of single-core Celerons that support PAE, while the first multicore Celeron was just launched this year) while Northwood-C supports HT, but does not support PAE.
The pae kernel is now enabled by default on pae-enabled systems, the old “default” kernel is available as an option for those systems.
The pae kernel not only supports >4GB memory on 32-bit systems, but also supports additional features like the “no execute” memory protection to try and prevent things like buffer overflows. The devs decided that the pae kernel was the best option for systems that supported it, so it replaces the old -bigsmp kernel option and will be selected in lieu of default when applicable.
No performance hits or anything else to be concerned with.
Thanks for that post. … It suddenly woke me up to the fact I may need to go back and update the audio-troubleshooting and alsa-update openSUSE wiki’s, reflecting the fact that there can be a pae kernel, and one has to pay attention to that when updating their alsa version.
Bizarre I didn’t wake up to this before, until I read your comment about replacing the old bigsmp kernel.