64 Bits are Overrated

"I think they are doing a marketing gimmick. There's zero benefit a consumer gets from that. Predominantly... you need it for memory addressability beyond 4GB. That's it. You don't really need it for performance, and the kinds of applications that 64-bit get used in mostly are large, server-class applications." -- Anand Chandrasekher

most desktop apps never use more than 1 gigabyte of memory (mostly due to the von-neumann bottleneck). And 32bit still allows using more than 4 gigs of memory:
http://en.wikipedia.org/wiki/Physical_Address_Extension

So basically 64 bits just waste transistors.

>>40

verbz 4 fagz

>>40
It's not a sentence, it's a very short poem.

>>42
A haiku?

Prolapsed Anus
Extension
a shit

roses are red
violets are blue
hax my anus
eat da poo

>>43
Wow, that actually makes sense. Someone has a prolapsed anus which means that there is an extension out of said anus, and that extension is, obviously, covered in shit (because it is extended out of that person's anus and he hadn't bothered with a bidet).

hax my anus

>>34

It didn't suck.

Yes, it did. Constantly churning page mappings to deal with a limited virtual address space is cumbersome and inefficient.

DOS is still faster than Windows

Pigs don't stop being pigs just because enough applying enough thrust will make them fly.

>>47

Yes, it did. Constantly churning page mappings to deal with a limited virtual address space is cumbersome and inefficient.

Actually, kept everything localized and cache friendly.

>>48
PAE isn't free. Requiring the OS to remap pages it is using for itself means you are forced to invalidate the PTEs and incur TLB misses every time the OS must access memory that isn't mapped in.

>>49
Neither is EPA. You can't just make the RT write to invalid TULs unless the BLT is fully consumed and the FS is well-aligned, since that would incur TRU remaps to access memory.

>>50
IHBT

>>49
Am I missing something here? 32 bit Linux reserves half of the virtual address space for the kernel, and 32 bit Windows reserves a quarter. PAE or not, why would they need to swap those kernel pages when switching processes?

>>52
It's not about swapping or application pages at all. TLB entries always have to be invalidated when context switching and PAE doesn't change that.

The problem with PAE is that the kernel itself doesn't have enough address space to access all of the physical memory in the system at once. If the kernel needs to access something that isn't inside the 4 GB space that's currently mapped, the page mappings must be changed and something else must be evicted from the current mapping to make room. The kernel incurs TLB miss and page table walking penalties every time this is done. IIRC Linus once cited a case where workloads that use a lot of data in kernel high memory (e.g., lots of filesystem stuff) would spend 25% of their time just doing kmap / kunmap.

Why couldn't ey just make 40-bit systems?

This post was typed with the Colemak keyboard layout

Why couldn't ey just make 18-bit systems like I do?

This post was typed with my experimental 27-key keyboard layout.

64-bit is only good for FP

Dwarf Fortress needs more than 4GB now, otherwise it starts to run out of memory after a while on 6x6 or even 5x5 maps.

>>57 I 2x4gb planks just because some programs and games start swapping with 4gb. Of course i normally stay below 4gb, and enjoy the benefits of completely disabled swap file.