Geoff said:
Can you explain how to figure out it for one card and two?
I mean, when trying to determine what is right for you, me, others, how does
one slice up those performance charts into meaningful categories?
-g
First of all, two graphics cards in SLI mode or Crossfire, is
a "special needs" catagory.
Cards have two aspects two them - 2D performance and 3D performance.
The 2D part of the cards, hasn't changed in years. With AGP 8X
or PCI Express x16, there is plenty of bandwidth for updating
frame buffers, or for doing other simple 2D accelerated functions.
Just about any card will do a good job of 2D. So a $50 card with
either AGP 8X or PCI Express interface, is good enough.
In the past, 3D functions could be broken down into applications
like CAD (OpenGL) and Games (DirectX). (The two APIs do many of the
same things, but OpenGL is an open standard, suitable for other
operating systems.) Higher 3D clock rates, more memory, all
contribute to "buttery" performance, whether you are rotating a
solids model in a CAD tool, or playing your favorite first person
shooter game.
Vista will be one of the first OSes (for x86) to incorporate
3D into the desktop. In the same way that the MacOS used 3D
functionality to hold individual application windows (Quartz).
So Vista represents a change in purchasing habits, in the sense
that if you want to use the Aqua interface, then a certain level
of 3D performance would be good.
So, if the question is one not involving Vista, then any card
will do for 2D purposes. If you are a gamer, then there are
entry level, mid range, and high end cards, and each level
allows more detail to be enabled in a game's preference panel.
(Since 3D can be GPU limited, or CPU limited, putting a high
end card on a computer with a low end Celeron/Sempron on it, you
would not expect to get the benefit of the high end video card.
When selecting a card, some judgement must be made, as to "how
much card" makes sense to buy. Match the level of card, to
the speed of the processor, as otherwise the video card could
be mostly a waste of money.)
Another function hiding in video cards right now, is their
ability to accelerate video playback. (The degree of acceleration
is not earth shattering, but every bit helps.) An aid to that
function, is a high core clock. The video acceleration function
is a fixed block, so you wouldn't necessarily get feature
differentiation in that function. But a high core clock feeding
that logic block, means being able to handle higher resolution
video playback, like HD.
http://www.anandtech.com/video/showdoc.aspx?i=2798&p=3
Note that in this table, some feature differences occurred
because of a change to the video acceleration block (from one
generation of video cards to the next), while other feature
differences between cards, are performance limits. So Nvidia
would disable functions in software, that wouldn't make sense
on the lower end cards (i.e. if performance would be like a
slide show).
http://www.nvidia.com/page/purevideo_support.html
Selecting a 7xxx series Nvidia card, with a high core clock,
might be the best compromise for video acceleration, if you
cared about it. For example, in this table, there are two
7300GT cards mentioned. One has a core clock of 350MHz, and
the other lists 500MHz. For accelerating video playback, you'd
probably want to pay a few dollars more for the card with the
higher core clock.
http://www.techpowerup.com/gpudb/
Presumably, ATI has similar considerations, but ATI doesn't
document their products quite the same way as Nvidia. I
cannot quite figure out what I'm getting.
http://ati.amd.com/technology/avivo/h264.html ???
http://www.anandtech.com/showdoc.aspx?i=2536&p=5
http://www.anandtech.com/showdoc.aspx?i=2691&p=6
So, feature set considerations for a video card:
1) Output connectors. VGA with decent DAC (400MHz), and
decent output quality (look for customer reviews). DVI
output can be either single-link or dual-link, and
dual-link is needed for super-hi-resolution LCD monitors.
Svideo output, VIVO (video in, video out) are also useful
functions for an HTPC, or for inputting analog video from
a really old camcorder or a VCR.
2) 2D performance. All the same, as no one benchmarks that
stuff anymore. People used to do "scroll tests" and the
like, but no one cares any more.
3) 3D performance. Current day - used for CAD (OpenGL) or
gaming (DirectX). Future day - Vista will use it for
desktop display issues (Aero can be enabled or disabled).
More processing will be moved inside the video card,
with DirectX 10. There are very few DirectX 10 cards yet
(just 8800 series) but that will open a separate chapter
in the saga.
4) Video acceleration. Purevideo (Nvidia) or Avivo (ATI).
Perhaps more important to a person building an HTPC.
Some level of acceleration (IDCT) has existed for maybe
the last five years.
5) SLI or Crossfire ? Using two cards ? Purely dictated
by economics and benchmarks. If two particular cards
used in SLI mode, give better performance for less
money, than a similarly performing single card, then
maybe it makes sense. SLI/Crossfire bring with it
compatibility issues, which means you are at the mercy
of driver updates to more of an extent, than perhaps
trying to make a single card perform well in games.
For office use, or web surfing, only (1) and (2) matter.
A $50 card will cover most of those needs.
HTH,
Paul