Why isnt 64bit twice as fast as 32bit?

[Scorpiuscat]

I understand the basic idea of how a 32bit cpu works. Processing 32bits of data per cycle (and lets say a 500mhz CPU) times 500 million cycles per second would equal 16 billion bits of data per second.

A 64 bit cpu should be twice as fast, processing 64bits of data per cycle, so the same (for sake of argument) 500mhz 64-CPU would process 32 billion bits of data per second.

Clearly, 64 bit CPU's with 64bit O/S's are not twice as fast as thier 32bit cousins.

Why?

Will the day come when a 64bit computer is twice as fast as a 32bit one? And what will make this change happen?

 

[AndyMc]

According to
bit specifications Definition: TechEncyclopedia from TechWeb

CPU
The size of the computer's internal registers. This is the computer's "word" size, which is the amount of data the CPU can compute at the same time. Theoretically, if the clock rates were the same (800MHz, 1GHz, 2GHz, etc.) and the basic architectures were equal, a 32-bit computer would work twice as fast internally as a 16-bit computer. In practice, 8-bit, 16-bit, 32-bit and 64-bit architectures are rarely identical to each other even from the same manufacturer. Thus, a 64-bit computer may be (internally) less than twice as fast or more than twice as fast as a 32-bit computer.

In order to take advantage of a CPU with larger words, operating systems and applications must be recompiled with a compiler that supports the larger word size. If not, the older software may actually run slower in the bigger CPU, but this is totally dependent on the mix of instructions used in the program.

Most important, this measurement does not result in twice as much actual work being done for the user, as the computer's cache size and bus and disk speeds are all part of the performance equation.

So to sum up... its more complicated than 64 > 32 etc.

 

[AndyMc]

I've just ran some benchmarks in Sisoft Sandra:

Running both 32 and 64 bit tests.
According to the inbuilt results running 64 bit tests on a typical Pentium D 531 3.0 Ghz (1MB L2, Single core) it comes up with

7561 MIPS
7564 MFLOPS

For 32 Bit tests on a Pentium D 530 3.0 1MB L2
5884 MIPS
4828 MFLOPS

No it isn't quite twice as fast I'll agree, more like 25%. In terms of the 530 vs 531 chip, the 531 is exactly the same chip but has EMT64 64 bit optimisations in it, I couldnt find two chips exactly the same in the list to bench or I would have done so.

My CPU benches out at:

50113 MIPS
36009 MFLOPS

In both the 32 and 64 bit tests. It too is a 3.0 Ghz and is significantly faster than both 3.0 P4's in 32 and 64 bit testing. Even taking into account the fact that it has 4 cores I would suggest that it is rather more than 2x as fast as either of the other 3.0 GHz chips.

Yes I know all testing is subjective


I guess what I'm trying to say is that there's too many possible variables to take into account. My PC Benchmarks much more than 2x Faster than either of the two 3.0 Ghz Machines I have mentioned, although I suspect that is down to architechtural differences rather than JUST being 64 bit on a 64 bit OS. And as the
article states "All things being equal" yes 64 bit would be twice as fast, in reality, sometimes it is, sometimes it isnt.

I hope this helps

Andrew

 

[rajinisback]

The biggest difference I ever seen on 64 bit is when your dealing with Citrix Servers. It smokes 32 bit Windows by far. I'm using 64 bit Vista Ultimate at home. Long as it is stable like it has been I'll never go back to 32 bit.

 

[Scotteq]

I've run both 32 and 64 bit versions of Ultimate on my computer - In and of itself, Vista 64 is easily much faster than Vista 32. In my experience it's just not a question at all.

Having said that: Most consumer applications are *not* 64 bit. Vista 64 identifies these as such (the x86 folder), and runs them using a set of 32 bit libraries. I'm not a computer science (or even trained tech), but my high level understanding is this amounts to running the 32 bit apps in a 'container' which does the translating from 32 bit to 64 bit commands. Any time you have to do that there is going to be a performance penallty since you are (more or less) doubling the number of calls. On the positive side, common 32 bit instructions which are less than the full length can often be grouped and run in pairs. The rule being that there has to be enough room in a 64 bit instruction for both commands and the necessary tags to differentiate the two.

That's the theory - In practice, I can detect no discernable difference between 32 bit *applications* on 32 bit or 64 bit Vista. This leads me to believe the libraries and instructions Vista uses to run 32 bit apps are developed enough that the performance overhead from translation is offset by the added efficiency of grouping instructions.

Can I "prove' it?? No. I'm not a comp sci grad, nor am I a trained tech. But in usage, I can say for sure I have not noticed any difference on an application level between the two. By itself, Vista 64 is faster than Vista 32. 64 bit applications should run faster than their 32 bit counterparts.... But very very few people are writing native 64 bit apps because that particular market is so small, and because XP64/Vista 64 can run 32 bit applications anyhow.

 

[skunksmash]

i i beg to differ
after working with x32 ever since i touched a keyboard IT IS 50% quicker
if not more my current rig just seems flawless with ultimate x64 i cant seem to slow it down no matter what i have open

apps running at once

avg scan
spybot s&d scan
game installation
gpu benchmark fur renderer

while still surfing the web WITH NO LAG
i used to dream of being able to do all that with x32 XP ONLY LAST WEEK
i love this OS and its only gonna get better SWEEEEET

 

[whyteo]

It can be if the software is written correctly!

Apparently 64 bit Windows chess is twice as fast as 32 bit Windows chess - A nice example of what 64 bit can do - And will do in the future. Unfortunately we are still in the dusk of 32 bit, and full 64 bit is not being pushed right now. Consumer 64 bit OS's are juggling 32 and 64 bit - they have 2 jobs to do while the 32 is faded out.

I don't know much about computer hardware, but I think that Motherboard architecture and buses between components also slows things down for the present.

It will get better.

I understand the basic idea of how a 32bit cpu works. Processing 32bits of data per cycle (and lets say a 500mhz CPU) times 500 million cycles per second would equal 16 billion bits of data per second.

A 64 bit cpu should be twice as fast, processing 64bits of data per cycle, so the same (for sake of argument) 500mhz 64-CPU would process 32 billion bits of data per second.

Clearly, 64 bit CPU's with 64bit O/S's are not twice as fast as thier 32bit cousins.

Why?

Will the day come when a 64bit computer is twice as fast as a 32bit one? And what will make this change happen?

 

[rajinisback]

I agree whyteo. Even if the hardware/os is there for 64 bit the applications have to be written to take full advantage of it.