GPGPU physics makes no sense, just from an efficiency and cost
staindpoint. Buying a second graphics card jus to run physics, at 300-400
dollars, drawing about 40-80 watts of power, vs. 200 dollar physics card
drawing 25 watts of power. Until GPU's are ridiculously overpowered, this
stuff is just going to be a pipedream.
Blig Merk - 29 Oct 2005 18:26 GMT
> GPGPU physics makes no sense, just from an efficiency and cost
> staindpoint. Buying a second graphics card jus to run physics, at 300-400
> dollars, drawing about 40-80 watts of power, vs. 200 dollar physics card
> drawing 25 watts of power. Until GPU's are ridiculously overpowered, this
> stuff is just going to be a pipedream.
It is hardly a pipedream if Havok FX is going to have it anyway plus
Novadex has indicated doing it plus ATI just came out with a press
release supporting it.
The real issue is about scaling. People probably remember having to
manually configure settings to get a game running on their system
limitations. It has been gradual, but game developers are now usually
including game scaling utilities, some better than others. They are all
now usually including some kind of automated or semi-automated scaling
in their Options to determine what your system can handle. The latest
rather amusing example is Steam evaluating whether your system is up to
snuff or not.
The main factor with GPU based physics acceleration is that a
developer can almost depend on a PC to have a graphics card, but it can
be real hit-or-miss whether it might have a separate physics
accelerator card. So, in conjunction with scaling and graphics card
physics acceleration, tests can be done to see if the combination will
work for a particular system and it is up to the individual whether
they use it or not.
As for the loading and number of GPU's, there are already several
dual-GPU cards out with more on the way, a lot less power than two
separate cards. But there was even an article a couple days ago about
dual dual-GPU card systems, yes, a quad SLI. Expensive as hell? Yes,
but then everything starts out that way.
Walter Mitty - 29 Oct 2005 18:36 GMT
> GPGPU physics makes no sense, just from an efficiency and cost
> staindpoint. Buying a second graphics card jus to run physics, at 300-400
> dollars, drawing about 40-80 watts of power, vs. 200 dollar physics card
> drawing 25 watts of power. Until GPU's are ridiculously overpowered, this
> stuff is just going to be a pipedream.
Just a pipedream? So these people supporting it are making a big mistake
in your opinion?
Magnulus - 29 Oct 2005 19:56 GMT
They are feeling the waters and trying to steal the spotlight.
From what I have heard the PhysX Card from Asus will start selling at
around 200 dollars. And they'll have actual game support. At the moment
this GPGPU stuff is all academic (literally).
I'd be interested in GPGPU stuff, nontheless, but only if they can show
how it can be done without buying yet another 400 dollar graphics card or an
SLI setup.
Kroagnon - 31 Oct 2005 05:54 GMT
> They are feeling the waters and trying to steal the spotlight.
>
[quoted text clipped - 5 lines]
> how it can be done without buying yet another 400 dollar graphics card or
> an SLI setup.
Just how much will they accelerate this? I would seem more feasable to buy
an Athlon x2 and have the game use the 2nd core for the physics instead of
taking cycles from the GPU.