Re: Benchmarking Quake II .NET Exposed

From: Alisdair Meredith [TeamB] (alisdair.NO.SPAM.PLEASE.meredith_at_uk.renaultf1.com)
Date: 01/19/05 

Date: 19 Jan 2005 10:30:06 -0800

Michael Anonymous wrote:

> "Test Results"
>
> My results were that the Quake II NET version was about 54% the speed
> of the original 1997 version. Also, The Quake II native Vertigo
> Software version of Quake II was about 63% the speed of the original
> 1997 version.

Thanks, interesting comparison to be aware of.

I am not sure what it proves, other than the Vergtigo quake is slower
than original quake. Unless we can compare Vertigo to the original
implementation we may never know why.

It is still true that the managed and unmanaged builds of the Vertigo
code ran at similar speeds, so the .NET *can* deliver similar
performance on your code base - we don't have a managed build of the ID
code to get a valid comparison on a more optimised codebase so that
question remains unanswered.

Personally, I would expect that optimised codebases should perform
quite differently - the nature of optimising is different on the two
platforms.

For instance, this codebase is written in C++. In C++ it is common to
create a lot of short-live temporary objects to solve certain problems,
just as you use try/finally in Delphi. As these objects are
stack-based, there is no garbage to clean up. In a GC system, using
idioms with short-lived temporaries is going to clog up the garbage
collector, so you use different techniques instead, and minimise the
number of short-lived objects in the system.

Any system heavily optimised to work in one of these environments is
going to perform poorly in the other. If the .NET claims are true
though, similar attention to targetted optimisations should produce
applications of similar speed and efficiency - in fact the .NET hype
predicts the .NET versions will conitune to improve and one day deliver
better performance as that is where the tool-development is focussed
these days.

AlisdairM(TeamB)

Relevant Pages

  • Re: Thats not the Answer to Your Question
    ... and that lack of appreciation is warranted if block processing ... horizontal and vertical processing speeds on a personal computer, ... suspect calculation in the loop might be part of the C or C++ language ... Optimising compilers have been taking unecessary stuff out of loops ...
    (sci.image.processing)
  • Re: Release Dll Problem - Optimization Issue
    ... If optimising for speed on smaller size processors, ... This limits the compiler in using fast registers for the ... and speeds up access to the variable. ...
    (microsoft.public.pocketpc.developer)
Loading