AMD prepares SOI shift in PC processors with initial 0.13-micron process; .13 micron Athlon 4's THIS YEAR!!!

[BurntKooshie]

<< BurntKooshie

ok 64bit programs are more complex right so if we had SEE for a 64bit processor with 64bit programs we will get much more horse power right
i mean it is like pentium pro with 16bit programs it sucks but when we use 32bit programs the pentium pro shines right >>



64-bit programs don't have to be more complex. They simply allow larger integers to be used. This means that if you need numbers that are greater than 32-bits, you'll basically need a 64-bit architecture.

You'd get much higher theoretical performance. Alpha chips, and UltraSparcs (not sure about other vendors) already have SIMD -- they just call them different things (Sun calls it graphics visualization instructions, or something like that -- don't quote me on that though). The problem is that SIMD isn't for everything. It requires you to have two (or more) pieces of data, that are not dependant upon each other, and are supposed to have the same instruction performned on them.

Of course, when you increase the number of pieces of data that a processor can crunch through, the more important data issuing becomes (meaning it needs to be able to fetch more and more pieces of data from caches in a cycle), so you push the bottleneck around.

The Ppro had poor 16-bit performance for entirely different reasons (I don't remember exactly why) -- the problem was fixed (or at least, mosty fixed) in the P2, but of course, by that time, far fewer 16-bit programs were run (and win9x was out with partial 32-bit compliance).

 

[SCUBA]

so dosnt it has any potential power for the futur at least is 32bit binary limit all what we need as humans

 

[Sohcan]

<< The Ppro had poor 16-bit performance for entirely different reasons (I don't remember exactly why) >>

I think it was because the PPro lacked segmented registers.

Scuba: As BK and Noriaki said, the push to 64-bit processors is mostly to have flat addressing for more than 4GB of memory. The ability to do 64-bit arithmetic operations is not a big deal, since 32-bit operands is currently all that's needed for most applications, and SSE/3DNow already can do 64-bit and 128-bit vector arithmetic.

Yes, a 64-bit processor could do 64-bit non-vector arithmatic twice as fast as a 32-bit processor could do the 64-bit non-vector arithmetic, but that wouldn't necessarily translate to twice the performance. Arithmetic instructions usually only comprise about 50% of the instruction mix for most programs, and probably very little of the arithmetic code for the 32-bit processor would be comprised of 64-bit arithmetic. In actuality, the initial shift to 64-bit desktop processors would likely cut performance by 5-10%, due to the increased bandwidth requirements and because the caches will hold around half the number of instructions and data words. Once the system bandwidth increases and caches enlarge, the deficit will be nullified.

 

[SCUBA]

so if we had 64bit aplications with 64bit processor we will get twice as fast preformans :frown::frown:

 

[BurntKooshie]

No, because it's only faster at 64-bit data manipulation, not faster at 32-bit data manipulation. Only if every single integer were 64-bits, (which is ludicrous for the VAST majority of programs) would it be &quot;twice as fast.&quot;

Basic point: 64-bit computing isn't the solution to having programs run faster.

 

[SCUBA]

i got ur point thanks evry one

 

[BFG10K]

Nice. At the very least TBirds will run cooler and I won't need skyscraper-sized heatsinks on them.

 

[ElFenix]

tbirds won't run cooler. the athlon line will, but tbird is a .18 core that puts out a lot of heat.

 

[grant2]

but they most likely need these methods more than Intel because of the Athlons shorter pipelines which limits the Athlons ability to scale to higher speeds.

Huge pipelines aren't the wonderful panacea that some people make it out to be. This is just like the RISC vs. CISC argument that crapintosh users tried to foist on everyone back in the 90's.

 

[grant2]

ok guys lets stop thinking about those old 32bit cpus and start thinking of the new 64bit ones they will just duble speed

You'll end up with 50% larger programs that run 15% faster... at least, that's about the increase some programmers reported in going from 16bit to 32bit code 12 years ago =)