ASUS V9900 Ultra (NVIDIA GeForce FX 5800 Ultra) Video Card; AA & Anisotropy
Quality
|
CONTENTS
-
General
information
-
Video
card features
-
Test
system configuration
-
Test
results:
3Digest
-
3D
quality:
new AA 8x and 16x modes
-
3D
quality:
AA at different angles
-
3D
quality:
highest-quality modes
-
Conclusion
Spring 2003 is the hottest season for 3D game accelerators. If you remember,
in Spring 1999 3dfx launched its Voodoo3 line and NVIDIA brought out Riva
TNT2/Ultra. That spring was the only similar in the tension.
No sooner had NVIDIA started building up its GeForce FX 5800, than ATI
announced a new 3D king - RADEON 9800 PRO. NVIDIA stroke back with a whole
line of mainstream accelerators - GeForce FX 5200/5600, but they failed
to beat the current favorites of this sector - RADEON 9500/9500 PRO. Meanwhile,
ATI's new solution - RADEON 9600/PRO expected in April will be much cheaper
than 9500/PRO and much slower (according to some Net sources).
But the new solutions are still on paper, and testers simply hang around
looking for what to test. Old products of 2002 which everyone is tired
of? Or maybe something that is expected this Spring? But everything that
could be said is said; we just hope that production cards will be at least
90% similar to what we got in the tests. I mean NV31 (GeForce FX 5600 Ultra)
which is going to have clock speeds different from the reference card (for
example, ASUSTeK mentioned they would be 325/275 (550) MHz against 350/350
(700) MHz of the reference solution). It makes me think that the new-comers
will be slower than expected. It also concerns ATI when higher-rated cards
(9600) are slower than those marked 9500. But ATI has already offered DX9
accelerators for the mainstream market (RADEON 9500/PRO) though they are
still in short supply.
April and May will show who are developing just "paper" creativity and
who will actually fill up the market. At present we can see only GeForce
FX 5800 Ultra - expensive, sophisticated and greatly heating cards which
are crawling bit by bit into the market.
Some of these accelerators were already reviewed before:
The PCB of such cards is very sophisticated, that is why NVIDIA places
orders for them with third parties and supply manufacturers both with the
chips and PCBs (who only assemble the cards). Some vendors, however, demonstrate
their imagination and make masterpieces out of their NV30 cards (but we
are not able yet to estimate reliability of the cards and, in particular,
coolers).
Most manufacturers simply assemble such cards according to the reference
design including the FlowFX cooler which has gotten a great many ludicrous
names and epithets.
ASUSTeK is among those who decided to deal with such arguable and low-profit
product, and it throws in the box a copy of the reference card with its
sticker (why arguable is explained in the reviews listed above as well
as in this article).
Now let's focus on the card itself.
Card
ASUS V9900 Ultra |
|
|
ASUS V9900 Ultra |
The card has AGP x2/x4/x8 interface, 128 MB DDR
SDRAM (8 chips on both PCB sides). 128bit memory exchange interface.
Samsung K4N26323AE-GC1K
memory chips of BGA form-factor. Maximum clock speed is 550 (1100) MHz,
hence 1.8 ns access time. The memory works at 500 (1000) MHz in 3D and
at 300 (600) MHz in 2D. |
|
Comparison with the reference design, front view |
ASUS V9900 Ultra |
Reference card NVIDIA GeForce FX 5800 Ultra |
|
|
|
Comparison with the reference design, back view |
ASUS V9900 Ultra |
Reference card NVIDIA GeForce FX 5800 Ultra |
|
|
|
Since this is a copy of the reference design, we are not going to describe
it anew; but this time we will take the risk and remove the cooler so that
you can see the card naked and examine the cooling device.
Here is the processor hidden under the cooler.
The marking here is placed on the top, though earlier it was made in the
middle. Letter U on the right probably indicates the Ultra version.
The PCB design is very complicated because of DDRII memory, and NVIDIA
makes orders for the boards with third parties which have enough experience
in this sphere.
Box contents: |
ASUS V9900 Ultra |
Two user guides (one common for all ASUS video cards, and the other
for this particular card), CD with drivers and utilities, CD with ASUS
DVD player, CDs with games (mostly obsolete), S-Video-to-RCA and DVI-to-d-Sub
adapters, TV-out extenders. Unfortunately, they forgot to throw in a splitter
for external power supply (though a power supply unit may have no free
"tails"); it's not excusable for a card priced at $400-450. |
|
The video card ships in a retail package. |
ASUS V9900 Ultra |
The video card ships in a big box filled with foam
plastic by 70% which supports the card and its accessories. |
|
|
Overclocking
ASUS V9900 Ultra |
NVIDIA GeForce FX 5800 Ultra has risen its speed a little with
the new driver version. What we could obtain was 525/1050 MHz. Taking into account
that NV30 is limited mostly by the memory throughput, and the memory gains only
25 MHz (50), there is no sense in showing diagrams. Just add in your mind 3-8
fps. |
Note that
- Overclocking requires additional card cooling (for its memory, in particular):
- Overclocking depends on a certain sample, and you shouldn't extend single-card
results to the entire series or trade mark. Overclocking results are not obligatory
characteristics of video cards.
Testbed and drivers
Testbed:
-
Pentium 4 3066 MHz based computer:
-
Intel Pentium 4 3066 MHz;
-
ASUS P4G8X (iE7205) mainboard;
-
1024 MB DDR SDRAM;
-
Seagate Barracuda IV 40GB Hard Drive;
-
Windows XP SP1;
-
ViewSonic P810 (21") and ViewSonic P817 (21")
monitors.
-
NVIDIA drivers v43.40 (tests in the heavy modes for this review) and v42.82
(tests in 3Digest).
VSync off in drivers, texture compression off in applications. Texture
detail - High Quality.
Test results
Together with the ViewSonic P817 monitor and BNC Bargo cable it showed
excellent 2D quality at the following resolutions and clock speeds: |
ASUS V9900 Ultra |
1600x1200x85Hz, 1280x1024x120Hz, 1024x768x160Hz |
Test Results: 3Digest
Test applications:
-
Return to Castle Wolfenstein (MultiPlayer) (id Software/Activision) - OpenGL,
multitexturing, Checkpoint-demo,
maximum test settings, S3TC OFF, the configs are available here
-
Serious Sam: The Second Encounter v.1.05 (Croteam/GodGames) - OpenGL, multitexturing,
Grand Cathedral demo, test settings: quality, S3TC OFF
-
Codecreatures Benchmark Pro (Codecult) - Direct3D, Shaders, Hardware T&L,
Dot3, cube texturing, highest quality
-
Unreal Tournament 2003 Demo v.1077 (Final Release) (Digital Extreme/Epic
Games) - Direct3D, Vertex Shaders, Hardware T&L, Dot3, cube texturing,
default quality
-
3DMark2001 SE Pro (MadOnion/Remedy), Game2 "Dragothic" - DirectX
8.0, Hardware TCL, multitexturing, LOW Details, DXTC
OFF, double buffering, 24-bit Z buffer
-
3DMark2001 Pro (MadOnion/Remedy) - DirectX 8.0, Hardware
TCL, Game1, Game2, Game3, Game4, Low, High detail levels
-
RightMark 3D (one of the game scenes) - DirectX 8.1, Dot3, cube texturing,
shadow buffers, vertex and pixel shaders (1.1, 1.4).
February 2003 summary diagrams of videocards performance with latest drivers
Overclocked cards are marked red, clock speeds follow the 'o/c' sign.
For the summary diagrams we used drivers v42.82 for NVIDIA cards,
v6.292 - for ATI cards, v3.10.51 for SIS cards and v1.03.01.002 for Matrox
cards.
-
1. Return to Castle Wolfenstein -
-
2. Unreal Tournament 2003 DEMO -
-
3. Codecreatures Benchmark Pro
-
4. 3DMark2001, Game2 Low Details (Dragothic)
-
5. Serious Sam: The Second Encounter
-
6. 3DMark03
-
4.6.1. 3DMark03 standard tests
-
Windows
XP (January 2003)
-
Tests on Pentium 4 3066 MHz, 1024x768, 3D Marks
-
Tests on Pentium 4 3066 MHz, 1024x768, Game Test 1
-
Tests on Pentium 4 3066 MHz, 1024x768, Game Test 2
-
Tests on Pentium 4 3066 MHz, 1024x768, Game Test 3
-
Tests on Pentium 4 3066 MHz, 1024x768, Game Test 4
-
Tests on Pentium 4 3066 MHz, 1280x1024, 3D Marks
-
Tests on Pentium 4 3066 MHz, 1280x1024, Game Test 1
-
Tests on Pentium 4 3066 MHz, 1280x1024, Game Test 2
-
Tests on Pentium 4 3066 MHz, 1280x1024, Game Test 3
-
Tests on Pentium 4 3066 MHz, 1280x1024, Game Test 4
-
Tests on Pentium 4 3066 MHz, 1600x1200, 3D Marks
-
Tests on Pentium 4 3066 MHz, 1600x1200, Game Test 1
-
Tests on Pentium 4 3066 MHz, 1600x1200, Game Test 2
-
Tests on Pentium 4 3066 MHz, 1600x1200, Game Test 3
-
Tests on Pentium 4 3066 MHz, 1600x1200, Game Test 4
-
7. RightMark 3D
3D quality: new AA 8x and 16x modes
In the drivers v43.* and higher GeForce FX gets new AA modes: 8x and 16x
which are composed of MSAA and SSAA (MSAA 4x is combined with SSAA 2x2
and 4x4, respectively).
Such modes reduce the speed markedly; here are some examples for AA
8x (I think it's useless to draw diagrams for AA 16x working only in 800x600
or in lower resolutions):
And what about quality? Have a look at the screenshots: |
AA |
Application |
Balanced |
Performance (Aggressive) |
Serious Sam: TSE |
AA 4x |
|
|
|
AA 8x |
|
|
|
AA |
Application |
Balanced |
Performance |
Serious Sam: TSE |
With colored MIP levels |
AA 4x |
|
|
|
AA 8x |
|
|
|
AA 8x quality removes jaggies better, but font blurring and considerable
speed falls of SSAA make this mode unwanted (it's needed only for professional
accelerators where quality is much more vital than speed).
The difference between Application, Balanced and Performance modes is
based on adjustment of trilinear filtering quality and MIP levels (LOD).
Besides, the Performance mode enables S3TC/DXTC technology which lifts
up speed at the expense of quality (look at the sky in the screenshots
above).
3D quality: AA at different angles
Here we are going to examine anti-aliasing once again. Fist of all, look
at it at different images.
AA |
Cards |
~10 degrees |
~30 degrees |
~45 degrees |
Serious Sam: TSE |
No AA |
GeForce FX |
|
|
|
AA 4x |
GeForce FX |
|
|
|
RADEON 9700 |
|
|
|
AA 6x |
RADEON 9700 |
|
|
|
The worst angles for GeForce FX are those which are close to zero.
AA quality of RADEON 9700 is definitely higher in this case. But the difference
vanishes away as the angle approaches 45 degrees. In AA 6x mode RADEON
9700 shows even worse pictures (at 45 degrees).
3D quality: highest quality modes
Let's see whether NVIDIA's new solution benefits from the AA mode on the
whole. Also, we will have a look at the highest-quality modes in case of
AA and anisotropy (where possible).
Anti-aliasing in large
AA |
GeForce FX |
RADEON 9700 |
3DMark03: Game4, Example 1 |
No AA |
|
- |
AA 4x |
|
|
AA 4xS |
|
- |
AA 6xS/6x |
|
|
AA 8xS |
|
- |
3DMark03: Game4, Example 2 |
No AA |
|
- |
AA 4x |
|
|
AA 4xS |
|
- |
AA 6xS/6x |
|
|
AA 8xS |
|
- |
3DMark03: Game4, Example 3 |
No AA |
|
- |
AA 4x |
|
|
AA 4xS |
|
- |
AA 6xS/6x |
|
|
AA 8xS |
|
- |
Elder Scrolls III: Morrowind, Example 1 |
No AA |
|
- |
AA 4x |
Unsupported |
|
AA 4xS |
|
- |
AA 6xS/6x |
Unsupported |
|
AA 8xS |
Unsupported |
- |
Elder Scrolls III: Morrowind, Example 2 |
No AA |
|
- |
AA 4x |
Unsupported |
|
AA 4xS |
|
- |
AA 6xS/6x |
Unsupported |
|
AA 8xS |
Unsupported |
- |
On the whole, GeForce FX fails. It shows inferior AA and lack of support
for some AA modes in some games. The mode of 8xS is almost useless as it
doesn't improve quality. That is why we will use AA 6xS as the maximum
AA mode in Direct3D.
Anisotropy
Anisotropy |
GeForce FX |
RADEON 9700 |
3DMark03: Game4, Example 1 |
No ANISO |
|
- |
ANISO 8x Application |
|
- |
ANISO 8x Balanced/16x Quality |
|
|
ANISO 8x/16x Performance |
|
|
3DMark03: Game4, Example 2 |
No ANISO |
|
- |
ANISO 8x Application |
|
- |
ANISO 8x Balanced/16x Quality |
|
|
ANISO 8x/16x Performance |
|
|
3DMark03: Game4, Example 3 |
No ANISO |
|
- |
ANISO 8x Application |
|
- |
ANISO 8x Balanced/16x Quality |
|
|
ANISO 8x/16x Performance |
|
|
Elder Scrolls III: Morrowind, Example 1 |
No ANISO |
|
- |
ANISO 8x Application |
|
- |
ANISO 8x Balanced/16x Quality |
|
|
ANISO 8x/16x Performance |
|
|
Elder Scrolls III: Morrowind, Example 2 |
No ANISO |
|
- |
ANISO 8x Application |
|
- |
ANISO 8x Balanced/16x Quality |
|
|
ANISO 8x/16x Performance |
|
|
Although the fans of NVIDIA and ATI are passionately arguing which
anisotropic modes may be collated, in real games the difference in most
scenes can be noticed only through a magnifier. The pictures above show
that perfectly. Even in Morrowind with its vast spaces without grass it
would take quite a lot of time to find scenes where different anisotropy
modes look different.
Anisotropy + AA
Mode |
GeForce FX |
RADEON 9700 |
3DMark03: Game4, Example 1 |
No ANISO/AA |
|
- |
ANISO 8x Application+AA 6xS / 16x Quality+AA 6x |
|
|
3DMark03: Game4, Example 2 |
No ANISO/AA |
|
- |
ANISO 8x Application+AA 6xS / 16x Quality+AA 6x |
|
|
3DMark03: Game4, Example 3 |
No ANISO/AA |
|
- |
ANISO 8x Application+AA 6xS / 16x Quality+AA 6x |
|
|
Elder Scrolls III: Morrowind, Example 1 |
No ANISO/AA |
|
- |
ANISO 8x Application+AA 4xS / 16x Quality+AA 6x |
|
|
Elder Scrolls III: Morrowind, Example 2 |
No ANISO/AA |
|
- |
ANISO 8x Application+AA 4xS / 16x Quality+AA 6x |
|
|
Now let's see what speeds the cards have at the maximum quality:
Even the lighter mode of AA 4xS (instead of 6xS) didn't help GeForce FX.
The Application mode with anisotropic filtering enabled is very GPU-hungry.
And this fact caused the defeat. But is this mode that needed? Our previous
tests showed that the Balanced mode is quite enough.
But although GeForce FX 5800 Ultra is the mightiest accelerator (the
cards on this GPU are already on sale in contrast to RADEON 9800), the
weak points outweigh.
Conclusion
-
GeForce FX 5800 Ultra is a real heater, and in the closed PC case it can
greatly rise the temperature affecting stability of the whole PC unit;
-
FlowFX makes a lot of noise, and its frequent reswitching irritates;
-
All these attributes of a super-accelerator do not help it outsmart RADEON
9700 PRO, except a few cases of victory, but at the maximum AA and anisotropy
levels this card loses the battle;
-
If NV30 were comparable in price to RADEON 9700 PRO, a lot of disadvantages
could be neglected. However, GeForce FX 5800 Ultra sells at, at least,
$425 in those U.S. stores where such cards appeared;
-
All these facts and NVIDIA's striving for NV35 made NV30 an outcast with
the manufacturers. Few companies are going to roll out cards based on NV30,
but it will be done mostly for the sake of press-releases. Such cards won't
be in large supply.
However, NV30 is flexible and programmable, and it is possible that the
drivers will raise its performance. At this moment I'm grateful that NVIDIA
has released v43.45 with errors corrected in some games. There is nothing
much we can say about the speed, but with the drivers v42.82 in RtCW at
1600x1200 it scored 120 fps against 70 with v43.*. You remember that v42.68
was optimized for 3DMark03, - NV30 could work twice faster but in other
games the operating system could hang easily.
If GeForce FX 5800 Ultra approached $370-380, the card could get a chance
to be in demand. It also concerns ASUS V9900 Ultra because this is a copy
of the reference card packed into an attractive box and equipped with old
software. But I heard that such cards are now on their way to Moscow with
the recommended retail price of $399. If it's true, the cards stand a good
chance to succeed!
If the cards are not many in number, the price will go up considerably,
because fans of NVIDIA will pay even $450 for such wonder :-). However,
NV35, an improved version of NV30 and accelerator stronger than even RADEON
9800 PRO, is already behind the corner.
Frankly speaking, NV30 is a card for researchers who are interested
in testing cards with different drivers and analysis of operations of various
3D functions.
Highs:
-
Highest 3D performance (RADEON 9800 PRO is not on sale yet);
-
High build quality (the PCB is very expensive and sophisticated, and the
boards are produced at special factories);
-
High stability and reliability in spite of the heating cooler (especially
from the memory chips);
-
Novations in 3D graphics, some of which can be already used (several modes
of operation, which mostly influences anisotropic filtering).
Lows:
-
Overpriced;
-
The cards are in very small quantities for the end of March.
Write a comment below. No registration needed!
|
|
|
|
|