64 bit performance (maybe someday AH will have 64bit version)

[artik]

Hello,

After long doubghts I've decided to test myself what is the advatage of 64bit SW over 32bit...

http://www.geocities.com/artyomtnk/bench-amd-64-32.pdf

I've did the comparison tests using pure 64/32 environment on Ubuntu Dapper Linux running on AMD Athlon 64 3000+ Venice Core with 1GB RAM

I was really impressed - the peformance was gained in about 150% in average and specially in math applications.
I had never realized that the difference might be such significant.

For heavy games like AH it can be very important...

It will be very nice if AH is released as 64bit binary together with 32bit one.

[artik]

So it there any chances for 64bit binary for AH?

It can improve the performance of the game.

[Spatula]

Im sure you'd see some benefits from a 64bit AH distribution, but i wouldnt expect 150% improvement, as i would suspect AH is more constrained by your video card's performance, than by the processor and memory-addressing scheme.
Would be interesting to see how much dif it made compared back to back on the same system.

[artik]

Any game that uses big amout of calculations will gain a performance.

All 3D requires lot's of math calculations, linear algebra and others things...

What I have seen that the difference in math calculations was about 180% of the original... (according to the test I've did)

Even the perofrmace of the acceleration is very important but CPU as well.

[smash]

I benchmarked WinXP 64 vs the 32 and saw only about a 1 - 1.5% improvement in most areas.  Absolutely no improvement in the video stats.  Wasn't worth it considering there aren't drivers for over half my periph's.

You tested on Linux.... not really the same thing.

[artik]

Which software had you tested? Was it compiled in 64 bit mode?
Most of software you run under XP 64 is still 32bit so it does not benifit from the 64 bit archetecture advantage.

You need to compare SW compiled as 64 bit with same version of SW compiled as 32 bit.

It does not relay to OS because the specific program takes all the CPU and not OS.

The only difference that Linux cames with exactly same software prepared in 32 and 64 bit mode - thus it makes it very easy to test...

Edit:
The fact the XP64 is not stable now does not change the fact 64bit architecture is the future:

Now 1GB of memory is normal in several years 2GB and 4GB will be normal and it is the top limit of 32bit systems.

[straffo]

This as nothing to do with the trasition from 32 to 64 bits, in 64bits there is more registers, less legacy opcode.

In short a better CPU lead to better benchmark.

I'll add that your bench is CPU oriented ,and doesn't take into account drivers.

[artik]

Originally posted by straffo
I'll add that your bench is CPU oriented ,and doesn't take into account drivers.

Unfortunatly couldn't get nVidia video drivers installed on the LiveCD - for this I need normal installation thus I couldn't bring any kind of "Gaming" benchmarks.

What else drivers can be tested???

But anyway I've tested several kind of heavy typical processing from different areas (mostly multimedia connected):

• Image Processing
• Video Processing
• Audio Processing
• Mathematical Calculations

For my work on PC that is heavy multimedia oriented this benchmarks are quite improtant that lead me to desision - I'll install 64 bit linux version instead of 32.