Ben Scott wrote: > On Thu, Apr 23, 2009 at 6:22 PM, Alex Hewitt <hewitt_t...@comcast.net> wrote: > >> "Linux hostname.localdomain 2.6.27.5.117.fc10.i686.PAE #1 SMP Tue Nov 18 >> 12:08:10 EST 2008 i686 athlon i386 GNU/Linux" >> > > The "i386" indicates the running kernel is for the i386 > architecture. In other words, 32-bit. A kernel for x86-64 will > identify the architecture as "amd64" or "x86-64" or something like > that. (In an interesting historical twist, AMD invented "Long Mode" > and Intel copied AMD.) > > >> When I issue the free command I see all 8 gb of RAM. Does that mean that >> the distro installed itself as a 64 bit version? >> > > Nope. As Ted says, PAE means the processor can see more than 4 GiB > of RAM, even when in 32-bit protected mode. Most processors made > since 1997 or so support PAE. Many (but not all) motherboards do as > well. > > With PAE supported and enabled: > > * The physical address bus has 36 lines. These are the actual pins > coming out of the processor. > * The physical address word is 36 bits wide. > * The page table structures change to support the larger physical address > word. > * A third level of page table indirection is added to support the > larger page tables. > * The processor can address up to 64 GiB of RAM or other hardware. > * The virtual address word (point size) is still 32-bit. > * Each process is still limited to a 4 GiB virtual address space. > * Each process is still limited to a 3 GiB user virtual address space > (kernel reserves 1 GiB). > > The common scenario where PAE is of benefit is a multiple-process > workload, where no single process needs more than 3 GiB of memory, but > the aggregate memory use of all processes is greater than 4 GiB. The > kernel and MMU can map different RAM pages into each process's virtual > address space. > > It is possible for an OS to support bank switching, to enable a > single process to make use of more than 4 GiB of RAM. At the > process's request, the kernel can change the memory mapping for the > process. For example, say the process writes 1 GiB of data into > memory, and then tells the kernel to switch that with a new 1 GiB > block. The kernel unmaps that 1 GiB of RAM, but leaves it allocated. > The drawback is the application has to do its own memory management. > I don't know if Linux implements this. Microsoft does for Windows, > but they kind of had to, because they were so late to the 64-bit > party. > > -- Ben > _______________________________________________ > gnhlug-discuss mailing list > gnhlug-discuss@mail.gnhlug.org > http://mail.gnhlug.org/mailman/listinfo/gnhlug-discuss/ > >
Just an update - the system that I was trying to install various 64 bit Linux distros also wouldn't install Vista 64. Turns out the processor I was using has some kind of TLB bug (AMD Phenom X4 9600). I RMA'd the processor after finding that I couldn't easily communicate with anyone at AMD. AMD support requires that you register with them and no matter how much I tried I couldn't get the registration completed. One might suspect that they don't actually want to talk with their customers. Under the circumstances I RMA'd to my supplier who granted an exception return under a lack of compatibility category. I ordered AMD's new Phenom II 940 processor which seems to get excellent reviews. I really need a 64 bit platform because I'm running software that manipulates large images. -Alex _______________________________________________ gnhlug-discuss mailing list gnhlug-discuss@mail.gnhlug.org http://mail.gnhlug.org/mailman/listinfo/gnhlug-discuss/
