[bf1942] BF2 linux server at launch

"Einar S. Idsø" esi at itk.ntnu.no
Tue May 3 04:08:41 EDT 2005


Andreas Fredriksson wrote:
> At the time the code base wasn't very optimized yet, so the only
> observation I could make at the time was that the AMD-64/Linux version
> was much faster than the 32-bit version running on windows.
> 
> It might be related to the nature of that particular machine (a dual
> Opteron) which has a NUMA-style memory architecture with local memory
> busses for each CPU. 32-bit Intel boxes have nothing like this.
> 
> Quite frankly I believe that most of the speed boost comes from the
> improved memory bandwidth rather than the CPUs themselves, so I expect
> the final, optimized code to narrow the performance gap when the
> product is released.
> 
> It's great value for hosting providers though, because it means better
> server performance any way you look at it.
> 
> // Andreas
> 
> 

Great, thanks! I suppose that single AMD 64-boxes will have a similar
advantage over 32-bit Pentiums?

We're currently looking into obtaining a dual CPU box - using the
NUMA-architecture instead of plain SMP sounds like it's definitely the
way to go, right? I've read a couple of head-to-head tests between dual
Opteron and dual Xeon Nocona (EMT64) systems, and both CPUs seem to have
an upper hand in differnt areas. We don't care about rendering speeds in
Lightwave or compilation times with GCC - what we're interested in is a
good CPU for BF2. Is there, to your knowledge, any noticeable difference
between a dual Xeon Nocona and a dual Opteron system wrt. BF2? A
dual-Core CPU would also be highly interesting, but I suppose there is
very little you can say about systems based on this new architecture?

Einar



More information about the Bf1942 mailing list