RE: [GD-Windows] DirectX Pure Device / HW Vertex Processing Stuff

[Tom F. <to...@mu...>]

(1) This is an ideal question for the DX list:

-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-
 FAQ: http://msdn.microsoft.com/library/techart/DirectX8faq.htm
 Web Interface: http://DISCUSS.MICROSOFT.COM/archives/DIRECTXDEV.html
 Problems/Suggestions: DIR...@di...
 Use the Web Interface (above) to unsubscribe from the list.
 Use plain-text only. HTML is not accepted. Attachments are removed
 MSDN DirectX Developer Centre:  http://msdn.microsoft.com/DirectX
-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-

For a start, nVidia people hang out on that list and are more than happy to
answer these sorts of questions.


(2) Note that a fast CPU (e.g. 1.5GHz or above) can beat the GF3 moderately
easily in raw vertex-processing speed. In a real game, you probably still
want to use hardware VS - the CPU usually has some other things to do
instead. But in terms of raw speed - CPUs are currently faster (a Radeon8500
on the other hand will happily beat the CPU in any benchmark I've thrown at
it - I'm rather impressed).


(3) Remember that the usage rules for buffers are pretty detailed. You need
to be using a DISCARD/NOOVERWRITE scheme, or things are going to go very
slowly. Also note that reading from AGP VBs is _seriously_ bad, and that not
writing the whole of the AGP buffer can also be slow (the write-combining
only works if you overwrite whole 64-bit chunks, IIRC - if you just update a
single DWORD then it all goes slowly). So this may explain your massive drop
when doing processing yourself.


Tom Forsyth - Muckyfoot bloke.

What's he up to now (and can I have a go)?
http://www.eidosinteractive.com/downloads/search.html?gmid=86

> -----Original Message-----
> From: Vek [mailto:ve...@ho...]
> Sent: 23 September 2001 06:35
> To: gam...@li...
> Subject: [GD-Windows] DirectX Pure Device / HW Vertex Processing Stuff
> 
> 
> I guess this is the best place to ask this - since DX is 
> windows specific...
> and its not really about Algos or such.
> 
> I recently got ahold of Geforce3 hardware to test my vertex 
> shaders / pixel
> shaders in, on my Game Engine.
> To my surprise though, simply flipping the engine into 
> 'hardware vertex
> processing mode' cut my framerate to about 50% of what it is 
> in software
> vertex processing mode.  This is when dumping raw vertices to 
> the card to
> transform and light and render via a vertex shader... using 
> vertex index
> buffer and vertex buffers... no pixel shaders involved.
> 
> It also drops to 5% of what it is in software 
> vertexprocessing mode if I
> additionally mess with those vertexes (Transforming them by my program
> manually instead of by a vertex shader).   Are there AGP specifics I'm
> missing out on?  Or some situations which I'm not aware of?  
> I'm setting the
> device up almost exactly like the NV demos, etc... except those go at
> several hundred frames a second in the same situations.
> 
> I would have thought that hardware vertex processing would be 
> the same speed
> or faster... not 5% as fast.  500 polygons are enough to bring it to a
> crawl.
> 
> If I can't mangle this soon, I'll take the renderer code out 
> and place it
> into a nice 'clean room' to mess with it, and find out whats 
> going on by
> brute force ;) But I was hoping someone out there already ran 
> into this.
> -Nick Lawson
> ve...@ho...
> project: http://members.home.net/vektuz/gw2k
> 
> hooray.
> 
> 
> _______________________________________________
> Gamedevlists-windows mailing list
> Gam...@li...
> https://lists.sourceforge.net/lists/listinfo/gamedevlists-windows
>