ARM Says 2015 Will Be The Year Of 64-Bit Chips

[walkthetalk]

PIII Xeon
Introduced October 25, 1999
Number of transistors: 9.5 million at 0.25 ?m or 28 million at 0.18 ?m
L2 cache is 256 KB, 1 MB, or 2 MB Advanced Transfer Cache (Integrated)
Processor Package Style is Single Edge Contact Cartridge (S.E.C.C.2) or SC330
System Bus clock rate 133 MHz (256 KB L2 cache) or 100 MHz (1–2 MB L2 cache)
System Bus width 64 bits
Addressable memory 64 GB

 

[walkthetalk]

http://en.wikipedia.org/wiki/3_GB_barrier

Physical address limits[edit]
It is a common misconception[according to whom?] that 32-bit processors and operating systems are limited to 4 GB (232 bytes) of RAM,[3][4] as were the original 80386DX and other early IA-32 CPUs. Since the 1995 Pentium Pro,[5] almost all modern x86 processors can in fact already address up to 64 GB (236 bytes) RAM via physical address extension (PAE). PAE is a modification of the protected mode address translation scheme. It allows virtual or linear addresses to be translated to 36-bit physical addresses, instead of the 32-bit addresses available without PAE.[6] The CPU pinouts likewise provide 36 bits of physical address lines to the motherboard. [6]

Many x86 operating systems, including any version of Linux with a PAE kernel and some versions of Windows Server and Mac OS, support the use of PAE to address up to 64 GB of RAM on an x86 system.[7][8][9]

Use of PAE to address RAM above the 4 GB point allows use of more than 3 GB.[clarification needed] There are, however, factors that limit this ability, and lead to the "3 GB barrier" under certain circumstances, even though the processor fully supports PAE. These are described in the following sections.

 

[bit_user]

4Gb Limit with 32 bits? NO. That was a Microsoft issue. If you ran Windows (desktop) there was a 4Gb limit. If you ran Windows Server, Linux or OS X, you could address more than 4Gb of RAM.

The explanation is that Intel added a stop-gap to x86 chips, to give 32-bit a bit more longevity until 64-bit CPUs arrived on the scene. The solution was called PAE (http://en.wikipedia.org/wiki/Physical_Address_Extension), which adds another 4 bits to the physical addresses in the page table, but processes would still be limited to a 32-bit address space. Many Linux distros switched over to this as the default, for 32-bits, because all Intel processors had this feature (except the older Pentium M's) from the Pentium 3 onward.

I actually upgraded the CPU in an old Pentium M laptop, so that I could update to the latest 32-bit Ubuntu. I've built my own kernel, before, but I found a good price on the CPU and figured the additional speed and power saving benefits would be worth it. I also dropped in a SSD and maxed out the RAM, so it was a good upgrade.

Just calculate 2^32 = 536,870,912 bytes. Microsoft was the problem... again!

I don't really know what to say about this, except that perhaps you should try a different calculator.

 

[IInuyasha74]

http://en.wikipedia.org/wiki/3_GB_barrier

Physical address limits[edit]
It is a common misconception[according to whom?] that 32-bit processors and operating systems are limited to 4 GB (232 bytes) of RAM,[3][4] as were the original 80386DX and other early IA-32 CPUs. Since the 1995 Pentium Pro,[5] almost all modern x86 processors can in fact already address up to 64 GB (236 bytes) RAM via physical address extension (PAE). PAE is a modification of the protected mode address translation scheme. It allows virtual or linear addresses to be translated to 36-bit physical addresses, instead of the 32-bit addresses available without PAE.[6] The CPU pinouts likewise provide 36 bits of physical address lines to the motherboard. [6]

Many x86 operating systems, including any version of Linux with a PAE kernel and some versions of Windows Server and Mac OS, support the use of PAE to address up to 64 GB of RAM on an x86 system.[7][8][9]

Use of PAE to address RAM above the 4 GB point allows use of more than 3 GB.[clarification needed] There are, however, factors that limit this ability, and lead to the "3 GB barrier" under certain circumstances, even though the processor fully supports PAE. These are described in the following sections.

I am kind of amazed you could post 4 times in a row, saying 32-bit isn't limited to that amount of RAM, and then in the final post post evidence that it was. 32-bit is locked down to 4GB. This chips extended the memory address space with additional registers to overcome it. In other words, a 32-bit address space is limited to 4GB, but the CPUs used with the PAE system use 36-bit or more for additional RAM, meaning from the memory side of things its a 36-bit CPU, with the execution units and other aspects of the core are 32-bit. Its not a wholly homogeneous 32-bit core.

 

[tygrus]

Some phones and tablets now have 3GB RAM. To fully use 4GB RAM even Linux needs 64bit addresses because of partitioning and addresses used for mapping access to system devices.

 

[bit_user]

meaning from the memory side of things its a 36-bit CPU, with the execution units and other aspects of the core are 32-bit. Its not a wholly homogeneous 32-bit core.

Pfft. It was a 32-bit core in exactly the same way that 8086 was a 16-bit core. Both had segmented memory for accessing windows of a larger address space than you could address with a single, word-length integer.

What determines the word-length of a CPU is the largest data size on which its primary contingent of instructions can natively operate. So, just because a chip might support 64-bit floating point or have a 128-bit vector register tucked away, somewhere, doesn't mean it's a 64-bit chip or a 128-bit chip. Similarly, the 8088's 8-bit bus width didn't suddenly make it an 8-bit chip, nor did the Northwood Pentium 4's 128-bit bus width make it a 128-bit chip. And just because the page table can map into a 36-bit address space doesn't somehow make Pentium 3 a 36-bit chip.

 

[bit_user]

I am kind of amazed you could post 4 times in a row, saying 32-bit isn't limited to that amount of RAM, and then in the final post post evidence that it was.

At least @walkthetalk self-corrected. It would have been better to use the edit button, or at least point out the self-correction, but I think @walkthetalk deserves some credit for that.