Yeah, I'd definitely vote that Linus is out of touch. On the other hand, most x86 class machines out there still run 32-bit software even though they're on AMD64/EMT64 CPUs. I blame that on an inferior OS wannabe. Come on - I've been using 64-bit machines since 1994 and other people have been using them for longer; it's hardly "the edge", it's more a case of a lot of the world being backward. The 64-bit data (wide wide) word goes back to ~1961. The >32bit addressing was a much later development; with the cost of RAM even in the mid-1970's, enormous RAM was just not economical - we're talking from 10's of M (late 1970s) up to billions of $$ per GB (early 1970s). Anyone know of a multi-billion $ supercomputer from that era?
Rather than bigger physical and virtual address space, really useful for processes that need to allocate about > 4 GiB of memory (and can drive to slight decrease of performance since math on 64 bit addresses is slower), the switch to x86-64 also doubles the general purpose and mmx registers, and here comes the major performance increase. It was not the case of SPARC, where SUN suggested 32-bit for particular binaries since they could be faster than 64 bit counterparts (and no additional features apart larger address space) nor the Alpha of cource since they never deployed a 32 bit version.
His statement wasn't an indictment of 64-bit as being "new and unusual" so much as a statement about 32-bit i686 being waaay too uninteresting for the tinkerer crowd.
Even the embedded x86 crowd has 64-bit CPUs to play with now, so even those folks don't necessarily have to stay on 32-bit....
out of touch
Yeah, I'd definitely vote that Linus is out of touch. On the other hand, most x86 class machines out there still run 32-bit software even though they're on AMD64/EMT64 CPUs. I blame that on an inferior OS wannabe. Come on - I've been using 64-bit machines since 1994 and other people have been using them for longer; it's hardly "the edge", it's more a case of a lot of the world being backward. The 64-bit data (wide wide) word goes back to ~1961. The >32bit addressing was a much later development; with the cost of RAM even in the mid-1970's, enormous RAM was just not economical - we're talking from 10's of M (late 1970s) up to billions of $$ per GB (early 1970s). Anyone know of a multi-billion $ supercomputer from that era?
Rather than bigger physical
Rather than bigger physical and virtual address space, really useful for processes that need to allocate about > 4 GiB of memory (and can drive to slight decrease of performance since math on 64 bit addresses is slower), the switch to x86-64 also doubles the general purpose and mmx registers, and here comes the major performance increase. It was not the case of SPARC, where SUN suggested 32-bit for particular binaries since they could be faster than 64 bit counterparts (and no additional features apart larger address space) nor the Alpha of cource since they never deployed a 32 bit version.
Alpha
Alpha could run code in a 32-bit "TASO" (Truncated Address Space Option) mode. Dimdows NT for Alpha was only 32-bit.
Read his comment again,
Read his comment again, please, you've got it backwards.
I don't know
I'd prefer to think of 64-bit for people who dream of a better world.
I think you missed the point
Linus wasn't saying that 64-bit is unusual, but rather that 32-bit is so passé and pedestrian that the bleeding edge users long since left it, and there are no (or very few) bleeding edge users using 32-bit i686 to try strange and wonderful combinations of settings.
His statement wasn't an indictment of 64-bit as being "new and unusual" so much as a statement about 32-bit i686 being waaay too uninteresting for the tinkerer crowd.
Even the embedded x86 crowd has 64-bit CPUs to play with now, so even those folks don't necessarily have to stay on 32-bit....
--
Program Intellivision and play Space Patrol!