We suspect that you are very eager to see how the Radeon 9800 XT
looks:
ATi Radeon 9800 XT
|
Chip: |
R360
|
Manufacturing
process: |
0.15-micron
|
Transistors: |
~115 mil.
|
GPU
speed: |
412 MHz
|
Memory
speed : |
|
Pixel Shader: |
2.0
|
Vertex Shader: |
2.0
|
Pixel
Pipelines/Pixel Fillrate: |
8 / 3296 MP/s
|
TMU’s/Texel Fillrate: |
1
/ 3296 MT/s
|
RAMDAC: |
(2) 400 MHz
|
Memory
size : |
256 MB
|
Memory
type and interface: |
256-bit, DDR-SDRAM
|
Inputs
and outputs |
VGA, DVI-I, S-Video
|
Accessories: |
|
Software: |
–
|
Full
version software: |
–
|
Estimated price: |
|
Just by looking
at the specifications we see that Radeon 9800 XT is a quite puny upgrade of
Radeon 9800 PRO. The memory is 30 MHz faster and the core 32 MHz faster. Not
much really, it’s just an increase of about 8 percent in both cases. However,
this is not the whole truth. ATi has included overclocking in their new drivers,
also known as ATi OverDrive, which according to ATi will make it possible
to run the chip at 460 MHz. That is certainly a very impressive figure. However,
this is still a little unstable at the moment which you’ll see a bit later
in the review.
| Card/Chip: |
Radeon 9700 Pro / R300
|
Radeon 9800 Pro / R350
|
Radeon 9800 XT / R360
|
Maximum
Memory: |
128 MB
|
256 MB
|
256 MB
|
Temperature
sensor: |
No
|
No
|
Yes
|
| F-Buffer (Smartshader 2.1): |
No
|
Yes
|
Yes
|
Optimized
memory controller (Smoothvision 2.1): |
No
|
Yes
|
Yes
|
Core
frequency: |
325 MHz
|
380 MHz
|
412 MHz
|
Memory
frequency : |
620 MHz
|
680/700 MHz
|
730 MHz
|
Those of you
that read our Radeon 9800 preview probably remember we were moaning about
the small improvements. This time they are even smaller if you take a quick
look in the table above. We also see that ATi has dropped DDR2 in advantage
for ordinary DDR. We can probably agree on that it is not prime time for DDR2
yet, if it ever will be for graphic cards.
Front |
|
Rear |
|
ATi has abandoned
the discrete cooling devices and brought the heavy artillery. Now it is a
massive copper heatsink with a much larger fan than before. The fan speed
is variable and the speed is dynamically changed depending on the temperature.
However, neither the fan speed nor the clock frequencies are adjusted whether
you are running 2D or 3D as nVidia’s fans and clock frequencies are. ATi claimed
during the presentation of the card that the fan speed would increase and
decrease very smoothly, but that is not the case on the test card we’ve got,
at least not at higher speeds.
Cooling |
|
On the rear side
of the card there is a very thin copper heatsink for memory cooling. A larger
heatsink would have been good since this one becomes extremely hot. About
the noise level there is no worries though. Well, it sounds more than 9800
PRO but still is far away from the noise generated by FX 5900 Ultra (let’s
not even mention 5800 Ultra). It becomes strikingly more silent in the room
when I switch from 5900 Ultra to 9800 XT.
As you can see on the picture, ATi’s cooling solution still is "single
slot", it does not require any additional PCI slot. The length of the
card is identical with 9800 PRO.
In- and outputs |
Another thing
we find very nice is that ATi’s reference design nowadays has ViVo as standard.
This has according to me been a big flaw on the previous cards in the Radeon
9×00 series.
DVI, VGA and the 15-pin S-Video cable with which you connect the ViVo dongle.
That’s ok, but some people might have preferred dual DVI instead.
Overall we find R360 technically rather uninteresting. In the end, it is simply
an "overclocked" R350, and that in its turn was only a moderately
optimized R300 that was overclocked. We did hope for more progress this fall,
but as it looks right now ATi’s arch enemy nVidia will not bring anything
revolutionary this year either.
For the massive news we have to have patience enough and wait until the spring.
A very pleasant
thing with Radeon 9800 XT (and 9600 XT as well) is that ATi and partners will
be able to bundle this card with Half Life 2. As far we are concerned the
game will be released in three different versions depending on the size of
your wallet. The co-operation between ATi and Valve goes to the "medium"
version. This means that you will get the full single player game, but if
you want to play online there is some kind of upgrade needed.
We are introducing
a little upgraded test system with this graphics card review due to that a
faster system will generate bigger differences in graphics tests.
Test system |
Hardware |
CPU: |
AMD Athlon XP 3200+ (400) Mhz
|
Cooling: |
Corsair HydroCool 200
|
Mainboard: |
Shuttle AN35 (nForce2 400 Ultra)
|
RAM: |
768 MB DDR333 @ 2-5-2-2 Timings:
3x 256 MB Corsair TWINX512-3200LL DDR-SDRAM
|
Graphics cards: |
GeForce FX 5900 Ultra (256 MB, 450/850)
ATi Radeon 9800 Pro
(128 MB, 380/680)
ATi Radeon 9800 XT (256 MB, 412/730)
|
HDD: |
80 GB Western Digital Caviar 7200 RPM Special Edition (8 MB cache)
|
Sound card: |
Creative Soundblaster Audigy 2 ZS Platinum Pro
|
PSU: |
Nexus NX-3000 300W
|
Ethernet: |
D-Link DFE-530TX 10/100 |
Software |
Operating system: |
Windows XP Professional (Service Pack 1 + updates) |
Video drivers: |
nVidia: Detonator FX 44.03 and 51.75
ATi: Catalyst 3.7 |
Other drivers: |
nVidia UDA Chipset Drivers v2.45 |
Benchmark applications: |
Unreal
Tournament 2003 (v2225)
Quake 3: Arena (v1.32)
Aquamark 3 (“1.0”)
Tomb Raider: Angel of Darkness (v.49)
Star Wars Jedi Knight: Jedi Academy (“1.0”)
Battlefield 1942: Secret Weapons of World War II (demo “1.0”)
Comanche 4 (demo “1.0”)
3DMark03 Professional Edition(v330)
WarCraft 3: Reign of Chaos (demo “1.0”)
Counter-Strike (v1.6)
Mafia: City of Lost Heaven (v1.1)
SPECViewperf (v7.1)
|
As you can see
we have twelve different tests for you to examine. Nine of them are actual
games. The remaining three tests are SPECViewperf that test professional 3D-performance,
Aquamark 3 that is based on the game Aquanox 2 and last but not least the
completley synthetic Directx9-test 3DMark03. All together these tests include
9 different game engines and we deliberately tried to diverse the types of
games we use so it wouldn’t be all first person shooters. We have to admit
though that we have not included any sporting games. The main reason for this
is that this writer doesn’t have the time and generally I don’t play these
types of games.
Something you might notice is that we have tested the FX-card with two different
sets of drivers. This is because the contemperary drivers from nVidia contain
optimations and "funny stuff" so it’s very hard to pick a decent
driver when one is to perform various comparisons between competing graphics
cards in a "fair" way. We will exemplify the differances between
nVidias different drivers as well as the preformance and overall quality.
In this review
you will come across a wide variety of expressions and abbreviations. If you
feel confused by these terms we suggest you take a look at the following list.
This list contains a couple of expressions exclusively used in this review
and are not necessarily generally accepted.
- AA, FSAA (Full Scene Anti Aliasing) – Edgesmoothing (reduces
“pixlyness” in 3D-graphics), all contemprary graphics cards should be able
to handle at least some form of this. AA can be applicated in various strengths
2, 4, 6 x and so on. The higher value, the better.
- AF,
Aniso (Anisotropic Filtering) – Anisotropical
texturefiltering
(increases the focus in textures/surfaces in 3D-graphis), all
contemporary
graphicscards should be able to handle AF as well. Same as AA you can
set
different strengths and again more is better.
- App (Application Preference) – If anything is marked
with this suffix it means that we have left the control of the AF to the
game instead of setting it via the graphics card control panel.
- Trilinear and biliniar filtering –
These are two different types of texture filtering that are either used
seperately or in combination with Anisotropicl filtering. The best combination
here is trilinear filtering and anisotropic filtering. At present time trilinear
filtering is standard with most new game releases and has to be considered
the minimum level of acceptance with regards to quality.
- Aliasing – Graphicsbug that depends on inadequate
information/precision.
This bug often manifests in the form of "dancing" pixels on
either
edges or inside textures. By using AA and AF you can reduce
aliasing.
If it is anything
indistinct with this text you are more than welcome to hit me with your questions.
The address is in the head of this review. (Please remember though that we
are not a non-profit support service. If you want support however I suggest
you visit our forums or get in touch with the dealer that sold you your product.)
I have promised
you all a new method of testing when it comes to our reviews of graphics cards
and this will be the virgin voyage for this new method. Among the new improvements
you will find that we have started to perform tests with FRAPS. This is a
tool that can measure FPS in any game. The results are, however, not as exact
with games that don’t have the feature to record a demo. We have tried to
work around this by measuring the FPS during longer periods of time since
this should eliminate statistical deviations. In games that have built-in
tools for testing we have chosen to use these tools. Furthermore we have decided
to test more games with fewer settings per game. Please let us know what you
think of these new routines. You’ll find the address above. (Could be both
complaints and praise.)
So… Shall
we
see what this Radeon 9800 XT really can do? Click the link to find
out.
Before we start
with today’s tests we want to inform you about the colour system we use in
today’s review. Since we know that many of our readers (and we ourselves too
for that matter) are worried about the development of aggressive optimizations
lately, we’ve simply chosen to show less reliable test results in grey. It
can be about lowered picture quality or simply that we’ve found other reasons
to believe that it is a matter of too aggressive optimizations. Unnaturally
low results (i.e. because of a bug) are also grey.
We are using
the resolution 1280×1024 with 4x AA (Full Scene Anti Aliasing) and 8x AF (Anisotropic
filtering) in every test, if nothing else is stated. We’ve chosen the resolution
1280×1024 (or 1280×960 in games where 1280×1024 isn’t an alternative) because
it is a reasonable resolution concerning performance, but also because most
of our readers should have monitors able to handle this resolution.
In the tests where 1280×1024-4xAA/8xAF appeared to be too demanding we’ve
chosen to first lower the resolution to 1024×768 and then finally turned AA/AF
off until a playable level of performance is reached.
After we’ve done
the virtual performance tests we’ve spent about 30 minutes playing the game
because we want to see how it works practically.
Notice that below
every diagram there is a small table with screenshots from the different cards/drivers.
|
Quake 3: Arena
|
We test the Open GL-game Quake 3 to evaluate the performance
of older titles. A large amount of titles are built on the “Q3”-engine.
We use the test demo four.dm_67 in the test utility Q3Bench.
|
nVidia clearly
has a performance advantage in the old Quake 3. Radeon 9800 XT naturally succeeds
in taking the lead over the 9800 Pro but nVidia still has a hard grip of the
first place.
When it comes
to picture quality you directly see that ATi is the one with an advantage.
Exactly as expected their anti aliasing is of higher quality which appears
on the large towers right side, among others. Concerning both manufacturers
anisotropic filtering, they look almost identical.
Subjective analysis:
Quake 3 still works, precisely as expected, incredibly well with all of the
three cards we’ve tested. The performance differences are principally impossible
to notice during gameplay. The differences in anti aliasing are more noticeable
but Quake 3 generally don’t include that many moments with high contrast (the
moments when “aliasing” is most noticeable) so there are no gigantic
differences we are talking about. Especially not when it’s such fast gameplay
we are dealing with.
|
Unreal Tournament 2003
|
UT2003 is a DirectX 8-game which puts the graphic cards
under a lot of stress with large textures, high amount of polygons
and more. A multiple of games are built on this engine. We are using
the more graphic demanding flyby-test. We test two different maps
in the game: Bifrost and Inferno.
|
ATi has a marginal
advantage in UT2003 with 9800 Pro and XT gives them a very healthy boost past
nVidia. We’ve chosen to show the results from nVidia’s cards with Detonator
51.75 in grey since the picture quality with AF activated in the control panel
suffers too much for a fair comparison. The big difference between 9800 Pro
and XT is there to a good deal due to the larger amount of memory on the XT-card.
I can already reveal that this is the biggest difference you will see in this
review.
|
Picture
quality :
|
|
Control
panel :
|
|
|
|
|
Application:
|
|
|
|
The picture quality
in UT2003 has been the reason for some of the criticism against nVidia recently.
nVidia has chose to lower the picture quality in this game when it concerns
anisotropic filtering. Most clear we see nVidia’s lower picture quality in
the 51.75-driver when AF is activated in the control panel. The “plate”
in the front of the picture is significantly more blurred than on the other
pictures. ATi has also introduced certain optimizations in their drivers;
however, these are working generally for anisotropic filtering. The picture
quality doesn’t suffer when activating AF in the control panel here though.
nVidia’s optimizations consists of that they, in 44.03, applies a mix of trilinear
and bilinear filtering. In 51.75 they have, as long as AF is activated in
the control panel, went even further and quite simply turned off the trilinear
filtering. This results in visible transitions between the mipmap-levels.
To see this you must watch the game “in action” though. A still
picture doesn’t show the entire dilemma here.
The FSAA-quality is yet again ATi’s triumph and UT2003, unlike Quake 3, includes
a lot of edges with high contrast which makes the anti aliasing pretty important.
Subjective analysis:
UT2003 has been a hard beast to master for the graphics card manufacturers.
Personally I think that nVidia’s optimizations are hard to see while playing
the game. However, on certain maps, especially the ones with flat ground and
far sight, there are very visible quality deteriorations. Concerning the performance,
all three cards works approximately equally good, but it feels like the nVidia-card
suffered a bit more of what is called “mouse lag” (e.g. the mouse
feels stubborn) when AA is activated then what ATi’s two cards did.
|
WarCraft 3: Reign of Chaos
|
WarCraft 3 is one of this year’s best-sellers
which make it a good object for testing. Even if the graphics lacks
extravagant technique the game is quite demanding. The performance
tests are made on the first map in the WC3-demo with FRAPS.
|
ATi has a clear
lead in this test. nVidia has succeeded in raising the performance without
visible quality losses with Detonator 51.75 but it’s still not enough to take
the first place.
The major difference
in picture quality in WC3 is the anti aliasing. Due to some reason it seems
like nVidia’s cards have problems with horizontal edges in this game above
anything else. Check i.e. the large log in the fire; it looks like it is principally
without AA.
Subjective analysis:
WarCraft 3 works equally good concerning performance on all three cards we’ve
tested. The performance differences are not noticeable while playing the game.
ATi’s better anti aliasing makes the picture a bit softer to look at though.
Quake 3 is so
far the only game where nVidia still has the advance, but we have only done
a third of all of our tests. Next game is Mafia, a personal favourite I spend
a lot of hours in front of.
|
Mafia:
The City of lost heaven
|
Mafia is based on a Direct3D engine (developed
by the Mafia team) which uses large amounts of low-quality objects
to get a high detail level. Similar 3D engines can be found in i.e.
GTA. We have used Free Rides first level with FRAPS to measure the
performance.
|
The Radeon cards
have an great advantage in Mafia, no doubt about it. With 51.75 nVidia has
shut down anistropical filtering. Detonator 44.03 suffers from another problem
because it simply won’t cooperate with FRAPS in Mafia. In other words we do
not have any approved figures of how nVidia works in this test. The test results
generated by FRAPS can differ some because in a game without demos it is rather
hard to get identical game rounds. It might be possible that XT has a slight
advantage.
It maybe gets
rather boring to hear (read), but we can not lie; ATi’s high quality FSAA
win over nVidia once more. As we can see Detonator 51.75 is totally out of
the game when it seems like AF is being disabled.
Subjective analysis:
Mafia runs smoother and looks better with ATi, it is sometimes that simple.
When we play the game without FRAPS it is actually OK with 44.03, not as good
as with ATi but rather close. Any difference between Pro and XT could not
be observed.
|
Comanche 4
|
Comanche 4 is based on a self-developed
Direct3D engine which uses Pixel and Vertex Shaders together with
high resolution textures. The game is one of few which really “needs”
a graphic card with 256 MB. We’re using the downloadable demo’s
benchmark tool.
|
Once more ATi
takes the lead, 9800 Pro gets along this time too. Detonator 51.75 lowers
(once more) the quality of texture filtering. It is not any gigantic differences.
Should I let
you guess? ATi has better FSAA, when it comes to AF the cards are rather similar
as long as nVidia uses the 44.03 driver.
Subjective analysis:
Comanche works very good with all cards. Maybe the differences would have
been bigger if we had a faster CPU (like a Pentium 4 Extreme Edition). The
FSAA quality is more salient in this game than many other because of all the
mountains with the heaven as background, therefore ATi gets the final advantage.
|
Counter-Strike
|
Counter-Strike does not need any presentation. The new
1.6 version is told to be more demanding than before, therefore
we chose to test it. The test consists of a demo from the map cs_havana
and a total of 18 players.
|
The differences
here are insignificant. Though, it is interesting to see that nVidia has got
their image quality higher with 51.75 at the same time as the performance
is better.
ATi’s AF is somewhat
better, and of course also the AA. If you look at a house you can see that
it looks better on the ATi card. nVidia’s Detonator 44.03 has a rather smoothy
picture, but that is fixed in 51.75.
Subjective analysis:
Personally I think that ATi and nVidia’s cards are equal in Counter-Strike.
The image quality is a little bit higher with ATi but for the most people
CS is like breathing, and you won’t actually see the small things. If I would
have to choose XT is my choice.
ATi is still
in the lead after two thirds of the tests.
|
Battlefield 1942:
Secret Weapons of WWII
|
Developed in Sweden, Battlefield 1942 is
still a very popular online multiplayer game and therefore we
feel that it’s important to test it. Once again, we use FRAPS
and test the downloadable demo.
|
Radeon kicks
some ass in BF1942, nVidia doesn’t stand a chance. Once again, Detonator 51.75
lowers the texture quality so much that we don’t find the results comparable.
It’s not the
first time we see Detonator 51.75 fiddle with texture quality in this review.
Detonator 44.03 produces good results when it comes to image quality. This
is one more game filled with high contrast edges and that makes the differences
in FSAA quality rather large (to ATi’s advantage).
Subjective analysis:
I have had the bad luck to play BF1942 with a 5900 Ultra quite a lot, and
I can’t say anything else than that it plays a lot better and looks
better on a Radeon 9800. The BF player should buy the Radeon without hesitation.
XT actually differed rather noticeable compared to a 9800 Pro 128 MB, especially
during the more action filled situations, the XT tended to slow down less.
|
Tomb Raider:
Angel of Darkness
|
Tomb Raider is the first commercial game using DirectX
9.0 fully for rendering graphic, and that makes it an interesting
test object. The test is performed using the latest patch with
built-in tools. The level Prague3a was chosen for the test.
|
ATi rips nVidia
into pieces in Tomb Raider. If this is an indication of upcoming DirectX9
performance (and it seems like that when looking at Half Life 2 tests) nVidia
won’t have a lot to say the coming 6 months. Detonator 51.75 increases performance,
but also causes some strange bugs.
nVidia’s present
drivers doesn’t allow for floating point precision for render targets etc.
resulting in lower quality. Detonator 51.75 also makes the Depth Of Field
effect to run amok. The main character is repeatedly erased, despite the fact
that the effect is supposed to erase things the longer away they are, for
example. As expected, AA works better with ATi’s card. Anisotropic filtering
also looks better on ATi’s card, since this kind of filtering causes “”texture
aliasing” (floating pixels) on nVidia’s cards.
Subjective analysis:
There’s no question about it, nVidia’s card isn’t even close to being playable.
Despite that, ironically, Tomb Raider is part of nVidia’s so called “The
Way It’s Meant To Be Played” program. I noticed no differences between
XT and Pro when playing the game.
|
Star Wars Jedi Knight:
Jedi Academy
|
Jedi Academy is the follow-up to the popular Jedi Knight
II. It’s based on the Q3 engine but has high resolution textures
and more light effects. We have tested a demo recorded by ourselves
on the map Traspir where we met 7 opponents.
|
ATi doesn’t win
unopposed, but wins nonetheless.
|
Image
quality :
|
|
Control
panel :
|
|
|
|
|
Application:
|
|
|
|
It’s a bit strange
to see that ATi’s image quality and performance increases when we apply AF
through the application instead of from the control panel. On the mountain
in the very middle of the picture it’s clear that the image quality once again
is better with ATi’s card.
Subjective analysis:
I am very confused with the test results from Jedi Academy. The strange
thing is that the game runs a lot better (and looks better)
on the ATi card, but in the performance tests it seems like if nVidia beat
the 9800 Pro, which doesn’t reflect the impression we got when actually playing
the game.
Finally, we find
among our gaming related tests this preview’s two synthetic tests: Aquamark
3 and 3DMark03 build 330. The reason why we chose to test the previously named
controversial software is simple. We have seen indications through Halo, Half-life
2 and Tomb Rider that 3DMark03 does not at all show as wacky results as nVidia
wanted to point out. Fact is that they seem to fit quite well with the above
mentioned DX9 games.
3Dmark03 has
a couple of months in the back, and with that reason we have chosen to do
this test a little more demanding by switch on post processing-effects which
raises the picture quality a lot.
|
3DMark03
|
3DMark03 was the first synthetic test where DX9-performance
could give us an insight in how advanced Shaders work on different
graphics cards. We have chosen to test GT4 "Mother nature"
and Pixel Shader 2.0 because they include DX9 effects.
|
If we take a
look at the 3DMark03 performance with Detonator 51.75 we can see that it has
been given a good increase. Although we have great suspicions that these optimizations
are more or less exclusively application specific, and that it might be optimizations
that is by a nature that is not applicable on a real game. Therefore we have
chosen to give 51.75 a gray pillar because we don´t believe it is giving
an accurate result.
With Detonator
44.03 that is "approved" by Futuremark you can see that nVidia really
has not got a chance against ATi’s cards. As said: nVidia says it is misleading,
but all of the DX9 games so far shows that Futuremark actually succeeded in
creating a pretty good prophecy with their 3DMark03.
If we take a
closer look at the even more synthetic Pixel Shader 2.0-test we see that the
FX-card’s weak spot is DX9.
In plain quality
there is not exactly something that separates the cards. If you really want
to look at the picture really close it looks like the water is slightly different
on the two pictures. We also see that ATi eventually has a minimal advance
in texture filtering if you look very close at the large rock’s right corner.
These are very small things, and you would not notice this in a real game.
Last out for
today is performance tests in Aquamark 3.
|
Aquamark 3
|
Aquamark 3 is a synthetic test based on the
game Aquanox 2. High polygon/particle count combined with shaders
mostly done by DX8 – But also DX9 class.
|
We got very weird
results in Aquamark 3. nVidia’s Detonator 44.03 shows a very low performance,
but the driver generates an output that is completely approved Detonator 51.75
on the other hand has noticeable losses of picture quality on a couple of
areas: Mist/smoke has been heavily reduced in at least one scene, texture
filtering is a lot worse, some things look like the are rendered with lower
precision, and finally many lightning effects are a little suspect or not
even there. Right now under these circumstances we can only point away from
the results from detonator 50. And under those circumstances ATi has an unbelievable
lead ahead nVidia.
Sadly enough
we have not found a reliable way of taking screenshots in the freeware of
Aquamark 3, and as Massive has not sent us a copy of the full-version you
will just have to trust our judgment instead of picture quality. If our readers
eventually ask for it, we might purchase the full-version. But as it looks
now we don’t know if its worth the trouble.
Finishing words about the game tests |
The game tests
that is lacking from this preview in our meaning is Halo, H-L 2 and eventually
Doom 3. To judge by the tests we have seen in Halo and Half-life 2 ATi has
a great lead here. Even when nVidia lowers the picture quality in H-L 2 they
still wont manage to keep ut with ATi.
Many will probably
remember the Doom III tests that was seen in relation to FX 5900 Ultra release.
We have discussed these with ATi and their advice is to wait until the next
"generation" of tests arrive until we judge any of them. We will
also have to wait with H-L 2 if we are going to believe nVidia.
To sum it all
up we have seen that ATi even with Radeon 9800 Pro has an advantage both when
it comes to performance and quality. The interesting is (we’re probably not
alone with this) that the less "obvious" tests (ex. BF1942, Mafia,
WC3, Tomb Raider, etc) performs much better with ATi’s cards than nVidia’s.
I have always
thought that nVidia did not overtake the "throne" from ATi with
their 5900 Ultra, they were on the same place. As for now Radeon 9800 XT 256
MB rules in the graphic cards market, it will be very interesting to see if
GeForce 5950 Ultra together with the final version of Detonator50 can change
these facts.
(As we have seen Detonator 51.75 is absolutely not THE driver many have waited
for.)
As we have mentioned
in the specifications and test system page ATi has a triumph card left. ATi
OverDrive in Catalyst 3.8 will be able to overclock the card dynamical depending
on the core’s temperature. This will raise the core’s speed with a maximum
of 460 MHz. This would have given ATi much better scores in the tests we saw
today, therefore the question is why they even sent out a test-product were
this feature was not included.
Before we leave
this preview’s performance tests we will take a look at how the card performs
in professional applications. Time for some SPECViewperf results.
The last of
our tests are SPECViewperf. This is as a “de facto” standard when
testing graphic cards when it comes to professional 3D. It has now and then
been some disturbance considering specific optimizations with this benchmark
and to be honest we have had little opportunity to examine what really happens
when the test is run. We haven’t been able to supply any pictures for comparison
but we can certify that what we saw on the screen looked identical no matter
which card or driver was used.
|
SPECViewperf 7.1
|
We use SPECViewperf to measure professional 3D-performance.
The test includes “parts” of different popular tools for
professional 3D-graphic artists and it gives a pretty good overview
of how the card performs in a row of different applications.
|
The results
are as clear as they can be. ATi are slaughtered and humiliated by nVidia
in five out of six tests. nVidia has succeeded in conjuring a significant
increase in performance in the dx-08 test with its Detonator 51.75, and
we have as of yet neither heard nor seen anything that would indicate that
these results wouldn’t be achieved in a fair manner. In the last test “ugs-03”
ATi gets a little revenge though, as it succeeds in taking the lead with
quite a good margin.
The Swedish
division of NH previously published a guide for installing FireGL-drivers
(FireGL is ATi’s professional series of graphic cards) on a Radeon 9700
Pro. This resulted in an enormous increase of performance in many of the
SPECViewperf-tests. That being said it isn’t the hardware being the limit
for the ATi cards in the SPECViewperf tests. This could likely be done because
they want those who needs the performance to buy a “real” professional
graphics card instead of buying a more regular one.
As we see
it, the most likely case is that it is the drivers that are limiting the
ATi cards, especially as the XT shows a near identical performance as the
9800 Pro.
Anyhow it
does not look like ATi is the obvious choice for all you 3D-graphics users
out there. This is if they don’t want to play with hacked FireGL-drivers.
This is not a good move made by ATi, as we have seen that they can create
better drivers and as they already exist it should be just to “move”
over most of the work from the FireGL-drivers as possible.
One thing that
probably comes as a very positive surprise is that ATi from now on will
include an overclocking tool in their drivers. The first card that was equipped
with the technique was the Radeon 9600 Mobility and as we said, this software
has now reached the stationary computers too. It’s also pretty fun to notice
that ATi also makes a big deal of how overclocking-friendly their new cards
are. This is a quite big step forward for ATi which earlier has been a little
bit like Intel when it comes to overclocking and their products while nVidia
has been more like AMD.
One nice
thing is that ATi offers dynamic overclocking. In other words the card is
being clocked on the fly while you are gaming depending how hot it gets.
Since the card all the time read from the heat sensor and clocks by it ATi
can trust this overclocking so much that it is covered by the warranty. We
presume that this also will apply on ATi’s third-party manufacturers.
We are not
really certain if ATi will offer the so called OverDrive software for all
their cards or if it’s only for the cards with the built-in temperature
sensor. One thing for sure is that Catalyst 3.8 is meant to ship the 8th
of October and that it shall contain the OverDrive software. Though we can
give you a picture of how the Radeon 9800 XT performs under maximum overclocking.
The computer
we test the card in is “in the free” so to speak. In other words we have no
case for the test system. The room temperature was at approx. 18 degrees
when we ran the tests. The tool which was used was Powerstrip where the
latest stable version had support for the XT card.
|
Standard
|
Overclocked
|
Percent
|
|
412/730
|
|
|
What is there
to say… to say that I was disappointed is a big understatement. In the
presentation of 9800 XT ATi bragged about how good the card overclocked
and how their OverDrive would take the core to 460 MHz. Let me say this:
at 460 MHz the computer hard locked after something like three frames with
major graphic corruption had been rendered. We cannot increase the core
with even 3 percent, which is very pathetic. You often say that there is
10-15 % “overhead” left on most of the graphics cards. The memory
is a little better but it doesn’t impress in any way either.
We might have
been given a broken card, who knows. Maybe the cooling wasn’t fitted like
it should. If not, this is how the 9800 XT will overclock “for real”.
You can really ask yourself what OverDrive would be good for at all, especially
if it’s not going to handle the memory frequencies.
We assume that
you want to see some performance tests at maximum overclocking. For this
we have chosen Quake 3 where ATi got most beaten and Tomb Raider where ATi
was the leader. Furthermore these two games represent something like two
different “schools”. Tomb Raider is the new generation of games
where shaders and the core’s frequency is of importance while the old Quake
3 generally likes everything and won’t say no to more memory bandwidth.
Additionally we have AA/AF in Quake 3 and not in Tomb Raider. Shortly said
it should give a quite good picture of how overclocking affects the card.
Sure there is
some performance to gain, but it’s hardly any impressive results.
This was a
disappointment to say the least. Rumours about bad yields has floated around
and it has been said that ATi’s intention was to ship XT with a 425 MHz
core but simply they haven’t been able to raise the core speed to that level.
If you then look at our results, it looks like there might actually have
been some truth in the rumours we have heard. We will confirm this when
the retail cards reach the market, but as it is now the XT is nothing that
will impress any enthusiasts at least when it comes to overclocking.
Before we end
the review we have some things to tell you about the little brother 9600
XT, turn the page for more information.
ATi isn’t only
releasing the Radeon 9800XT, in the horizon we also have the Radeon 9600XT.
This card is, rather naturally, an upgrade of the Radeon 9600 Pro and from
what we have seen it looks very promising. Here’s a comparison with it’s older
versions:
| Card/Chip: |
Radeon 9500 Pro / R300
|
Radeon 9600 Pro / RV350
|
Radeon 9600 XT / RV360
|
Max
memory quantity: |
128 MB
|
256 MB
|
256 MB
|
0.13-micro
core : |
No
|
Yes
|
Yes
|
Optimized
memory controller (Smoothvision 2.1): |
No
|
Yes
|
Yes
|
Low
K-manufacturing process: |
No
|
No
|
Yes
|
Core
frequency: |
275 MHz
|
400 MHz
|
500+ MHz
|
Pixel
pipelines/Pixel Fill rate: |
8 / 2200 MP/s
|
4 / 1600 MP/s
|
4 / 2000+ MP/s
|
Memory
frequency: |
540 MHz
|
600 MHz
|
600+ MHz
|
As we can se
the Radeon 9500 Pro is still king when it comes to fill rates, at
least theoretically. If ATi has succeeded in getting the Radeon 9600XT up
to 550MHz core frequency then this would match the 9500 Pro. This is because
that the 9600-series has half the amount of pipelines as the 9500 pro.
A short explanation
of what Low-K is might be justified though. Low-K Dielectric is a specific
kind of material used to encapsulate the components in a microprocessor. By
encapsulating every single path in a chip one prevents disturbance which otherwise
might occur inside the chip. Thanks to the more pure signals that does not
disturb each other one can clock the processor higher while getting lower
heat development/power consumption.
ATi is the first to deliver a graphics card manufactured in this way (followed
by S3’s DeltaChrome shortly) and romours tells that nVidia has tried to master
this technique since before they released their GeForce FX 5800 Ultra. If
nothing else then this is a real blow against nVidia’s honour. First ATi succeeded
in delivering a 0.13-micro based graphics card (Radeon 9600 Pro) without having
any problems (which nVidia has with it’s first FX card) and then succeeding
in mastering the Low K-technique before they do.
Since the Low-K-technique
results in more pure signals less power has to be applied as there is less
"waste". Thus resulting in that the Radeon 9600XT, despite its powerful
specifications, will not need to be supplied with additional power.
Another news
which we are not that certain is a real news or not is that ATi now can boast
with 24x z-buffering. From our knowledge though, all Radeon 9×00 over 9500
has this. But the concept is rather fuzzy so we are not really sure of what
is true or not. In fact there was similar confusion at the release of the
Radeon 9600 Pro, where it proved to be a miss in the presentation by ATi that
caused the headache.
In all other
aspects there is not much new except for the tempsensor that we have already
described on the Radeon 9800XT.
What the card
lack compared to the Radeon 9800-series is the F-buffer, Smartshader 2.1.
We have not seen any implementation of ATi’s F-buffer yet though, and we are
in doubt that it will prove much useful for us normal mortals. More reasonably
this is a finesse to be used by the professional 3D-graphic handlers working
with advanced shaders.
Radeon 9600 and 9800 "non-XT" |
What will then
happen to the present cards? Well, the Radeon 9800 Pro and 9600 Pro will be
the "non-XT" variants of the two new cards. Simply said the current
cards will still be there but the price will probably be lowered by a bit.
It sounds as a rather sensible move by ATi and personally it would not surprise
me if nVidia does the same thing when they launch the GeForce FX 5950 and
the 5700 Ultra later this year.
Radeon 9700 Pro
was an incredible card when released, Radeon 9800 Pro was a minor upgrade
of the former and here is now the Radeon 9800 XT that in turn just serves
as a minor upgrade for Radeon 9800 Pro. In the end we sit with a Radeon 9700
Pro on steroids, so to speak. The few existing architectural differences are
small, even when comparing 9700 Pro to 9800 XT. The largest difference is
shortly based on clock frequencies.
On one hand this
shows just how wickedly good the Radeon 9700 Pro was when it was released
last year, but on the other hand it shows that ATi hasn’t accomplished many
perceivable progresses regarding technology itself. They have simply raised
the clock frequencies. This doesn’t matter all that much as the card actually
performs very well, but we had wished that ATi perhaps would have optimized
their core a bit more than they did.
As mentioned,
this is all about clock frequencies. With this in mind its an utmost shame
that ATi hasn’t succeeded in presenting the upcoming drivers with the OverDrive
technology. As earlier mentioned, OverDrive is supposed to be able to overclock
the card all the way up to 460 Mhz core speed, without voiding the warranty.
Regarding that this actually doubles the advantage in clock frequencies compared
to 9800 Pro makes it rather pointless that we can’t give you any figures with
OverDrive activated. Furthermore we are not fully informed whether OverDrive
will also be able to overclock the card’s memory.
As we saw in the overclocking tests though, ATi has got a lot to work on if
they are to reach their 460 Mhz core speed. As of present they aren’t even
by far close. This is a pretty large disappointment, especially since ATi
have told us on several occasions how great an overclocker the 9800 XT is
going to be.
What I want said
is: The performance results we have been able to present to you does not have
to be, and probably aren’t, corresponding to those of a "real" Radeon
9800 XT. If I’m to be honest, the whole thing seems to be a stupid move from
ATi, as they shoot themselves in the feet by not having Catalyst 3.8 with
OverDrive ready for these previews that are posted today. Some people have
speculated in the fact that the reason for this may be that Valve’s and ATi’s
combined event takes place today the 30:th. That is: The launch of Radeon
9800 XT and Half-Life 2. We don’t know exactly what is behind it all, but
as mentioned, it does seem a bit strange that ATi voluntarily presents their
cards in a light lesser than that which is going to reach the costumers.
Anyhow, Radeon
9800 XT is the fastest card on the market, and judging from our forecasts
it looks like it will get an ever greater advantage as more DirectX 9 based
games are released. The card has a phenomenal image quality (of course), and
here also better than any opponent. The only sand in my eye is a too small
margin in the upgrade of a 9800 Pro to get really enthusiastic.
What might be
a problem for ATi is the price. Presently ATi’s recommended retail price is
a lot higher than that which nVidia claims their GeForce FX 5950 Ultra will
have. But this is something we will have to take a closer look at when the
retail cards finally shows up.
As we mentioned
in the introduction, ATi has got themselves two important partners as of late:
ASUS and MSI. Hopefully these large manufacturers will create some life in
the land of ATi, where all cards till now have been each other alike. If one
is to judge by this
hot picture one can see that in any case ASUS have some nice things going
on.
As of now ATi
leave us with a pretty ambiguous impression. The promised OverDrive might
possible make the XT to a by far more impressive product. But then we also
have to wait for the cards that overclock better than the one we tested. Unfortunately
we therefore have to wait with the final judgment until the retail cards reaches
us, which they will do at the same time as nVidia send out their GeForce FX
5950 Ultra, which plans for some pretty exciting reviews!
Otherwise if
you want a 9800XT in the shape reviewed today, then just run out and buy a
Radeon 9800 Pro 256Mb and overclock it to 412/730 Mhz. It’ll be cheaper and
the card will make less noise (and have less weight).
What probably
will be rewarding commercially for ATi with the XT-series is of course the
fact that Half-Life 2 comes free with the card. Without doubt a smart move
from ATi to team up with Valve; it will sell them hundreds of thousands of
cards.
