I would suggest waiting for a working 64-bit Windows to make valid conclusions about the performance difference.
Or you could just use Linux, an OS that does work =)

I doubt this place has even tried Linux, let alone use it for benchmarking. (Its like 90% of hardware sites out there, all Windows in benchmarking and nothing else.)

Its odd though, it looks like this site uses FreeBSD...Well, at least it isn't all bad, is it? :)


The one thing we can rule out in this article...Sisoft is dodgy as hell, and the current AMD64/EMT64 implementation in Windows ain't that good. (Not surprising, considering the driver support ain't as good as the 32bit version...I wonder wtf is going on there?)

Better off building two separate Linux distros and compiling them for max optimizations for the relevant architectures...Takes a few hours, but hey! They're fast CPUs. ;)

A more interesting test would be to try out specific AMD64/EMT64 functions, and find out where each implementation is good and bad. (although they are functionally equivalent, they are very different in implementation in regards to certain functions).

What about 64bit games?

Nice to finally see some tests of 64-bit performance. I agree, would have been nice to see some linux-based tests, and some more indepth analysis of where AMD64 and EMT64 differs in performance, and why though.

Hi,

I'm curious to know how GCC (GNU C Compiler) compile times stack up, when compiling under Linux. I think Linux (Gentoo for example), is a far better native 64-bit enviornment than pre-compiled XP-64.

A simple test would involve removing all optimizations from the USE flags, retaining only -pipe -march and -O2.

Cheers,

S-N

I'd like to see some results on a linux 64-bit platform!!

> or that would use over 2GB of RAM, which is the maximum that 32-bit operating systems would allocate for a single process.

please, DO NOT consider windows limitations as the ultimate limit!
it's even possibel to get 3.5GB per process (on linux, and certainly many os).

Could you please post wich ATI drivers did you use for this teste? I'm having some issues with ATI and XP x64 and I'm trying some different drivers, so I would like to kwno wich one worked ok for you. Thanks!

What about instruction scheduling? Aren't there any slight differences between Intel's and AMD's implementations, which would require a compiler to be aware of them and generate code accordingly?

After all, haven't all the programs tested were compiled for AMD? (since they were first with 64bit)

Basically, your article starts off with rhetoric about 64bits and ends with a don't care.

Pls compare Conroe and AMD performance during a full moon and explain any deviation from tesing during 1/2 moon. There might be a full moon problem.

What a great article!

I'd like to see another benchmark: QuickPi. It has both 32-bit and 64-bit native versions. Available here:

http://home.istar.ca/~lyster/pi/qpi40a.zip

at last a real test :)

Your use of the word efficient is rather dubious. Saying AMD is more efficient in 64 bit mode when it is beat in all but one of your benchmarks makes no sense.

You are calculating efficiency from the performance gain based on 2 different 32bit baselines (one for Core 2 and one for FX-62). Hypothetically, if by some case a Pentium 4 gained 20% going from 32 to 64 bits, would you claim that is is 10% more efficient in 64bits than Conroe?

One point missed here is that the Intel parts are much more dependent on their caches for performance than the AMD parts. 64-bit code means 64-bit pointers and for some programs that means half as much data fits in the cache. Both will slow down because of it, but possibly Intel's parts by relatively more.

Core 2 Duo kicks out AMD processors! Great work Intel

anybody else notice that the core 2 duo used was 130Mhz per core (total of 260 mhz)higher clock speed then the X2? couldn't they have overclocked the X2 to make it more even?

and to anyone who says the 260 mhz is a minimal difference.... I am typing this on a 266mhz laptop....

also, I would be VERY interested to see Linux benchmarks... windows 64 is know to have terrible bugs and problems....

very good artical indeed

que guia de mierda, es todo mentira, Intel es lo mejor que hay y punto, el conroe va a ser lo mejor del mundo

I do find article to be overly simplistic and analysis to be rather unreliable. Running benchmarks is not a good test for processor 64 bit operations.

It is well known fact that P4 does 64 bit multiply every second processor cycle, how about that on Core 2?

Pls check performance of specific 64 bit operations at assembly level:

add, sub, mul, div etc etc etc and than draw conclusions.

Currently I see no point in claiming that Core 2 works well in 64 bit mode since Intel always had problems with it.

You may use GMP library which has some assembly level opitmizations for 64 bit arithmetic for AMD64 which should work on EM64T as well. Run their speed tests (you will get number of cycles for different instructions and operations) and publish results.