Epox 8K3A+ Mainboard
256MB Corsair XMS3200 CAS2
AMD AthlonXP 2100+ (not overclocked)
Windows XP Pro
7.75 Catalyst (shipping) Drivers
Direct3D Performance
:
We're fast approaching the
time when just about any kind of benchmark is almost pointless.
At last we're seeing the push towards image quality and feature
sets rather than pure framerate though to some degree you can
never have one without the other. I've heard lots of gripes
about the merits of 3DMark as a true reflection of real world
performance and to be honest (flames incoming) I think this
is total bull! As an isolated test of performance 3DMark is
as valid as any game you can buy off the shelf would be. The
key to what I said is "in isolation". 3DMark is built
around an actual game engine and features polygons, textures,
overdraw and all the other things you're likely to find in your
favourite game. Yes there'll be times when your 3DMark score
goes down and the framerate in your favourite game goes up and
there'll also be times it happens the other way around but that
I'm afraid is the nature of your typical 3D engine. Nor will
it expose every bug you're likely to find in your favourite
title but conversely it may uncover a bug that doesn't affect
your favourite title. No single game, benchmark or 3 hour session
with FRAPS
enabled will give you the whole picture but as a method of feature
performance comparison it's as good as anything around.
First off I ran the default benchmark (1024x768)
with the slider set for performance. That's quite a score but
probably not as high as I'd have seen if I were running on an
Intel platform which certainly seems to favour ATi cards.
3DMark 2001SE Build330 - Default
Taking a quick look at the image quality from
the run I made at the performance setting we can see the Radeon
9700 doesn't shape up too well against the reference image.
Having said that the reference image is rendered in software
and with no idea what the equivalent settings would be we can't
draw any conclusions from this. If you take a look at the road
texture just behind the rear left wheel and the number plate
you can see the reference image is considerably sharper which
suggests it was rendered using some level of anistropic filtering
enabled. We'll go back and look at this comparison again later
when we test out the anistropic filtering performance.
Radeon9700 (Performance Setting)
Reference Image
On the lobby test where the subject is much closer
to the viewpoint image quality is very impressive indeed considering
we're using no AA or AF. .
Click For a Larger Image
Lobby Reference Comparison :
Even when we subtract the 9700's image from the
reference image we can see the two are very similar. The black
areas below show areas of no differences in the two while the
lighter areas highlight the changes from one to the other. I
have to come clean here and say this is much better the Ti4600
did when we performed the same comparison.
Click For a Larger Image
And to prove the 9700's not all looks and no muscle
here's how it performed at various resolutions, again using
the performance settings. 11677 at 1280x1024 is quite a showing
and leaves me in no doubt that the 9700 is simply untouchable
at this moment in time.
Grand Prix 4 AntiAliasing Quality :
AntiAliasing is no longer a gimmick that's included
to liven up the box specs but unlikely to be used. More and
more gamers are now at the point where they simply wouldn't
consider firing up there favourite game without some level of
AA applied. After a few days seeing it in action I have to admit
that the 9700's anti-aliasing is as good as I've seen to date
and probably better. Textures remain fairly sharp at all levels
and in action it's very effective. Here's a few screenshots
from Geoff Crammond's Grand Prix 4 to pour over.
No AA (Click)
2x AA (Click)
4x AA (Click)
6x AA (Click)
No AA
2x AA
4x AA
6x AA
I was able to pay GP4 at 1280x1024 with 4xAA and
4xAF enabled and the details maxed out with no problems at all,
and boy did it look good! I can actually notch up another point
to ATi because on the GeForce4 Ti4600 GP4 refuses to exit the
game in the normal way without dropping me into a frozen black
screen, the 9700 however caused no such problems and exited
perfectly back to the desktop. It's a long time since I've found
a good reason to praise ATi's drivers! Unfortunately they're
not all good but more on that later.
I'm guessing from looking at the gantry to the
left of the screenshots below that the 9700 isn't ani-aliasing
textures with transparency (alpha channel) data. It's also worth
noting that although there doesn't seem to be a huge amount
of difference between 4xAA and 6xAA there is a perceivable increase
in texture detail using 6xAA. You may just be able to see this
is you look at the difference between the steps (I think they
were steps) in the center of the frame. You'll see the individual
steps are more defined under 6xAA than with 4xAA.
No AA
2x AA
4x AA
6x AA
No AA
2x AA
4x AA
6x AA
Despite all the advances in technology the days
of free AntiAliasing are not with us yet and even the mighty
9700 takes quite a hit when you flip the AA switch. It still
remains a fine performance but on future titles the need to
balance IQ and performance will remain a consideration.
Anistropic Filtering :
There a still a surprising number of people who
don't understand and therefor don't use anistropic filtering.
I don't want to start lecturing these people on the technicalities
involved other than to say it uses better quality textures for
surfaces that aren't square to the point of viewing (viewport).
In a corridor for example all the surfaces, floor, ceiling and
both walls are angled from the viewer's perspective and normally
lower and lower quality textures would be used as the floor,
ceiling and walls get further away from the point of viewing.
The level of anisotropic filtering set alters how far into the
distance higher quality textures are applied so with 2xAF (Anistropic
Filtering) enabled the corridor textures might be high quality
up to around 20 feet ahead of you while with 16xAF enabled the
textures might be good for as far down the corridor as you can
see.
Even a fairly humble 2xAF setting can bring considerable
benefits. Remember I mentioned the fuzzy registration plate
further up the page? Well here's what happens to it when we
enable 2xAF.
No Anistropic Filtering
2x Anistropic Filtering
Even the far fog light becomes visible on the
bumper and if you look at the ground texture bottom left you'll
notice it looks considerably better.
Below are some strips from the 3DMark image quality
screens that demonstrate the 9700's anistropic filtering. Unlike
with the 8500 it's great to see no missing shadows or disappearing
details, the only thing that alters is the texture quality.
NoAF
2xAF
4xAF
8xAF
16xAF
And if we look at the 3DMark scores with the various
levels of AF enabled we can see that the performance hot is
more than acceptable, not free but damned impressive....or is
it?
Well NVIDIA would argue quite correctly that in
the past ATi have only allowed anistropic filtering to be used
in conjunction with bilinear filtering while NVIDIA's anistropic
filtering uses higher quality trilinear filtering. Remember
the performance/quality option above the anitropic filtering
level slider? Well with that set to quality it's now possible
to use a combination of trilinear filtering and anistropic filtering
at the same time. The benefits vary depending on the type of
textures used but if you look at the face of the control panels
below, particularly the higher of the two, you should be able
to make out that the trilinear filtered panels are sharper.
Anistropic Performance
(Bilinear Filtering)
Anistropic Quality
(Trilinear Filtering)
Unfortunately when anistropic filtering is combined
with trilinear filtering the performance hit is increased by
a fair amount, in fact we see a drop of around 1000 3DMarks
at each setting which is not entirely surprising. You've still
got to be impressed with the performance though even allowing
for this!
Code Creatures :
Code Creatures is tough benchmark for any system,
in fact how the hell they hoped to build a game around it at
the time of release in anybody's guess. I still look forward
to the day when something can manage a 30FPS average across
all three resolutions but the best is never quite good enough
is it?
Radeon 9700 Pro
GeForce 4 Ti4600
VillageMark 16bit Colour :
VillageMark is a quick and clean bench designed
originally to show off the Kyro graphics card. You're taken
on a flight along village streets which by their very nature
feature a heap of overdraw (hidden scenery). Interestingly the
9700 offered identical performance at both 16 and 32bit though
I can't imagine for one minute anybody is planning on running
anything in 16bit with a card like this!
VillageMark 32bit Colour :
TempleMark 32bit Colour :
TempleMark does pretty much the same thing VillageMark
does but uses 6 layer multitexturing rather than 3.
There is some kind of multi-texturing bug buried
in the drivers somewhere though and though I saw a couple of
examples of this here's a perfect demonstration of the problem
in action. The first shot shows a scene rendered using the Radeon
8500 while the second shows the same scene rendered on the 9700.
Radeon 8500
Radeon 9700
Driver glitches happen, there's no getting away
from that but even today you get the distinct feeling that every
advance ATi's engineering team make gets wiped out by the attention
ATi's driver dev team attract from some of the rather raw code
they release. Sure Catalyst looks better and is easier to navigate
but the fact remains that the shipping drivers lacked TRUFORM
while the latest release includes TRUFORM but the SmartGart
Control Panel has been disabled. If somebody could just get
a rocket under ATi's driver development team they'd undoubtedly
gain an extra level of respect, not least from games developers.
