by Alexey Stepin
07/22/2003 | 03:17 AM
As is known, the launch of ATI RADEON 9700 graphics processor and R300 architecture caught NVIDIA unawares. The company simply couldn’t offer any adequate response to ATI Technologies’ attack for 8 months. However, in fact, the NV30 chip, which gave birth to the today’s product family from the Santa Clara manufacturer appeared not very successful.
The cards based on GeForce FX 5800 used very expensive 12-layer PCB and were equipped with very expensive DDR II memory working at 1000MHz frequency but featuring very narrow 128bit bus. And the cooling solution of these graphics cards aka FlowFX, which was specifically developed for this graphics card family and was probably intended for deaf people became almost a legend. The worst thing for NVIDIA was the fact that their new GeForce FX 5800 Ultra solution was very often slower than RADEON 9700 PRO or demonstrated very low advantage over the competitor, which was an unbelievable situation for NVIDIA, really. All this resulted into inadequate reaction of the market to this product. In the long run, NVIDIA stopped the production of their new solution having manufactured only around 100 thousand pieces of this graphics processor. However, NVIDIA learned its lesson and took all mistakes of the failed NV30 into consideration. Very soon, in January 2003 a new NV35 chip came out. Compared with NV30, the new NV35 boasts higher performance, and uses 256bit memory bus for the regular DDR SDRAM. NV35 appeared a pretty successful product: having borrowed all the best from the predecessor, it was deprived of the predecessor’s drawbacks.
It is a paradoxical fact, but 3dfx engineers and developers used to say about the failure of GeForce256 as a product and its success as an implementation of the new architecture. Now that these people have actively participated in the development of the almost failed GeForce FX 5800, the same thing is true for NV30.
Here is what the new NVIDIA’s GPU can do (for more details on the NV35 performance and features, please, see our article called NVIDIA GeForce FX 5900 Ultra (NV35) Review):
NVIDIA GeForce FX 5900 Ultra
Number of transistors
Graphics memory controller
Graphics memory frequency
Peak memory bus bandwidth
Max graphics memory size
AGP 3.0 4x/8x
Pixel pipelines, pixel shaders
Texturing unites per pipeline
Max number of textures during multi-texturing
Texture filtering types
Max anisotropy level
Pixel shaders version
Vertex pipelines, vertex shaders,
Vertex shaders version
Full Screen Anti-Aliasing
Number of samples
2 (OGSS, OGMS),
Technologies aimed at higher memory bandwidth efficiency
Hidden Surfaces Removal (HSR)
As you see, the features list of the new NVIDIA solution deserves all our respect. By the way, even though the reference NVIDIA cards have got rid of the super-noisy FlowFX cooling system, they acquired an even larger solution instead, which still occupies two PCI slots inside the case and is still pretty noisy. Nevertheless, some graphics card makers took a different way: they developed and intruded their own cooling solutions for NV35 based graphics cards. Today we are going to take a closer look at a product like that from ASUSTeK Computer.
The new solution from ASUS Company based on GeForce FX 5900 looks very attractive: classical dark-green PCB with a stylish-looking cooling solution, which fans glimmer in the UV light, by the way. The printed circuit board of this graphics card is shorter than the reference GeForce FX 5900 Ultra PCB from NVIDIA: they removed mostly the power supply components located along the very right edge of the card. On the left, you can see a TMDS-transmitter from Silicon Image implementing the DVI-I Out and a spot for the VIVO chip, which will be installed onto the Ultra version of this product.
When we took the card, we were surprised to feel how unusually heavy it was, which was definitely caused by a massive copper heatsink covering the graphics chip and the memory chips. The heatsink foot was of some very complex shape with special lugs for the memory chips, which ensures very tight contact with their upper surface. Throughout the entire test session we didn’t have any complaints about the cooling. Moreover, this cooling solution works almost noiselessly, at east we couldn’t hear it working against the background of the CPU cooler and HDD fan. We can only thank ASUS engineers for this touching care about the user’s ears.
The only comment about the construction of this graphics accelerator has to do with the power supply connector: it is installed vertically, which is not very convenient, and holds to the PCB only by the soldering. Just for you to get a better idea of what I am talking about: the same connector on a few other graphics cards is installed horizontally and is fastened to the PCB not only by the soldering but also by two special plastic hook-lugs.
The reverse side of the PSB hides nothing remarkable: you can only see the spots planned for additional 128MB of graphics memory used in the Ultra version. Unfortunately, the card is pretty long even though it has a shorter PCB than the reference, and when installed it may block 1 or 2 DIMM slots. All in all, ASUS V9950 can surely be called a high-class graphics accelerator, boasting very high-quality design, brand name cooling solution and very rich bundle.
As for the items shipping together with the graphics card, the list is impressively rich. The manufacturer proved highly generous and included 6 CD-disks with various games and software into the package. Besides, when we opened the box we found an irreplaceable TV-cable with and RCA and S-Video connectors. Here is the detailed list of items bundled with the ASUS V9950 graphics card:
The last disk contai9n demo versions of such games as WarcrAnisotropic Filteringt III, Splinter Cell, Big Mutha Truckers, BREED, Colin McRae Rally 3 and TOCA Race Driver. Well, the happy owner of this graphics accelerator, won’t ever be bored, that’s for sure. Let’s take a closer look at the CD-disk with utilities. Our experience suggests that there might be something really “yummy” over there. And our expectations proved absolutely justified: besides the standard drivers set we found ASUS SmartDoctor, ASUS Digital VCR, ASUS GameFace and ASUS Video Security utilities. I would like to dwell on each of these programs a little bit more now.
This utility is intended for monitoring of such graphics card status parameters as temperature and fan rotation speed. We have already seen soothing like that by TYAN TGM. However, since ASUS V9950 is not equipped with any hardware monitoring systems, ASUS SmartDoctor grew into a common overclocking utility with very limited opportunities. The maximum frequencies you can set with the help of this program make 420Mhz for the chip and 890MHz for the memory, although the potential of this graphics solution is much bigger and it can work fine at 480MHz chip and 915MHz memory, which can be set using NVIDIA CoolBits.
ASUS Video Security appeared a pretty simple utility for video observation, which can also detect any movement within the view window. This software can also e-mail the necessary pictures to the preliminarily set electronic address.
Since we didn’t have any video-signal source and the VIVO chip on the graphics card, we couldn’t actually play with this software. However, I intuitively assume that this software is very unlikely to suit for serious use in security systems.
As for ASU Digital VCR, the name suggests that this is none other but a digital video recorder.
Note that this program can record a video stream only in DVD (MPEG-2) or VCD (MPEG-1) formats, while iuVCR can record this video stream in any format up to DivX, compressing it on the fly. However, ASUS Digital VCR looks very nice and can win the hearts of the large HDDs owners. Of course, this software is absolutely not necessary for the graphics card we are reviewing today.
I suppose that the most interesting program discovered on the disks is ASUS GameFace, intended to establish video communication in team games with multiple players, which will help the players to coordinate their actions much better. Unfortunately, this program didn’t get along well with the drivers of our video input device (Hauppauge WinTV), generating the “blue screen of death” every now and then. I didn’t manage to figure out what the matter was, I assume that this utility requires VIVO from the graphics card. Also I would like to stress that you have to install ASUS’ own drivers if you want to have the ASUS GameFace utility work correctly.
In general this program seems pretty handy, however, in my opinion, it will make sense only if you have a fast channel (starting from 256Kbit and up). I don’t think that model owners will manage to play and e-mail videos at an acceptable speed and with an acceptable latency.
Since ASUS V9950 graphics adapter belongs to the upper price group, it will have to compete with ATI RADEON 9800 Pro. Note that we will test both graphics cards in the nominal and in the overclocker mode. As you may have noticed, our testbed configuration remained the same:
The list of benchmarks use has also remained the same:
Every gaming benchmark was run with the highest graphics quality settings. We used 1024x768, 1280x1024 and 1600x1200 resolutions. This time besides two usual tests in the “raw speed” mode and FSAA 4x + Anisotropic Filtering 8x, we also included a so-called “stress-test”. Each graphics card was tested with the maximum Full-Scene Anti-Aliasing and Anisotropic Filtering settings. The hardest mode for ASUS V9950 on NVIDIA GeForce FX 5900 appeared FSAA 8x + Anisotropic Filtering 8x, and the hardest mode for ATI RADEON 9800 PRO was FSAA 6x + Anisotropic Filtering 16x. This way we tried to find out if the today’s leaders can provide acceptable performance in the heaviest modes and if it makes sense at all to use these modes in contemporary gaming applications.
We are still using the same games here, namely Return to Castle Wolfenstein, Jedi Knight 2: Jedi Outcast, Quake III Arena and Serious Sam: The Second Encounter. However, we are about to remove Jedi Knight 2 from this list and to include something more objective, where we will see the actual graphics accelerator performance, not limited by the system CPU.
Due to higher core and memory frequencies and to traditionally excellent performance of NVIDIA solutions in Quake3, the graphics card from ASUS wins an indisputable victory in modes without anti-aliasing and anisotropic filtering, and proves very successful in 4x FSAA + 8x Anisotropic Filtering mode. The only exception appeared 1024x768 resolution. However, in the heavier testing conditions the situation changes. Unfortunately, V9950 doesn’t support 6x anti-aliasing that is why it had to fight against the competitor using much more resource-hungry 8x FSAA, which definitely told on the results. Moreover, in 1600x1200 the graphics accelerator seemed to lack memory and it simply didn’t enable FSAA at all.
Although RTCW is based on Quake3 engine, ASUS V9950 behaves in an absolutely different way here. Our hero is defeated by ATI RADEON 9800 PRO in all modes, and even overclocking doesn’t improve the situation. In the hardest mode, its gaming performance doesn’t even reach an acceptable level.
It is no secret for anyone for a log time now, that Jedi Knight II doesn’t care that much about the graphics accelerator used in a system. However, higher working frequencies of ASUS V99500 manage to give this card a little advantage over the competitor in this game. Of course, the only exception is FSAA 8x + Anisotropic Filtering 4x, where neither chip, nor graphics memory can cope with the tremendous workload. Nevertheless, you still can play safely in this mode.
In Serious Sam: The Second Encounter ASUS V9950 shows its real best defeating completely even the overclocked ATI RADEON 9800 PRO. However, in case we have FSAA 8x + Anisotropic Filtering 8x the situation turns very familiar: the performance of GeForce FX 5900 drops down and RADEON 9800 PRO, which is in a slightly more advantageous situation here, dashes forward.
In Unreal Tournament 2003 the picture is really interesting: all in all the tested graphics cards perform more or less equally, with a small advantage on ASUS’ part. However, more advanced RADEON 9800 PRO architecture, namely, excellent support of shader operations and highly efficient compression technology, help this solution not to fall behind the rival and in some cases, even to outpace GeForce FX 5900. This is especially typical of those tests where FSAA and anisotropic filtering are both enabled.
Splinter Cell seems to be the hardest game in our test session. However, it showed very well that the efficient internal architecture of the graphics processor is extremely important: ATI RADEON 9800 PRO won in all testing modes. Moreover, the average gaming performance appeared very low, so that even with the high-end graphics card you will be able to play this game comfortably without any FSAA.
Now we will discuss some synthetic benchmarks results. As usual, we would like to start with an old, but very popular benchmark set – 3DMark2001 SE. We decided to skip the hardest test modes here, because they are not very interesting for us, to tell the truth. Besides, we came across some complaints on the web about several IHVs cheating in heavy 3DMark modes.
The indisputable winner in the first gaming test is ATI RADEON 9800 PRO. Overclocking of ASUS V9950 improved the situation a little bit, but still is not enough to help ASUS graphics card outpace the rival.
The situation is similar in the second gaming test. Overclocking again doesn’t save ASUS V9950 from failure.
The third gaming test also brings no news. ATI RADEON 9800 PRO hold tight to the leading position.
In the fourth gaming test, which is the most interesting one in the entire 3DMark2001 SE benchmark set, the situation is very exciting. Due to powerful shader units, ATI RADEON 9800 PRO managed to get ahead, but only in case no Anti-Aliasing was enabled. Moreover, the overclocked ASUS V9950 manages to catch up with the ATI graphics card working at the nominal frequencies. With FSAA 4x + Anisotropic Filtering 8x enabled, the situation changes to just the opposite and V9950 leaves its rival pretty far behind due to higher working frequencies. In the overclocked mode this graphics accelerator is simply beyond any competition. Although some on-line sources claim that there are a lot of cheats in the 3DMark drivers of both independent hardware vendors, the results obtained do comply with the theoretical conclusions and hence can be regarded as adequate.
There have been a lot of arguments about NVIDIA’s cheats in 3DMark03, but we were not going to join the choir. We simply tested our ASUS V99500 in the 3DMark03 with 330 patch, which is “protected” against any optimizations. Here are the results we got:
The Game 1 test from this test package is the easiest one, it doesn’t involve DirectX 8 and 9 features. ATI RADEON 9800 PRO won in this test, while GeForce FX 5900 refused to work with FSAA 8x + Anisotropic Filtering 8x, probably because its graphics memory was not enough to pass this test successfully. However, we also can’t exclude the possibility of some problems with the drivers.
In the Battle of Proxycon test the victory was in the hands of ASUS V9950, although it is not an indisputable victory this time. In the nominal mode RADEON 9800 PRO is faster, and ASUS V9950 manages to catch up with the rival only in 1600x1200. However, as soon as the workload increases, ASUS graphics card becomes a definite leader in all resolutions.
In Troll’s Lair test the situation is a little bit different: in Performance mode ASUS V9950 falls behind RADEON 9800 PRO in all resolutions, but under tangible workloads it manages to defeat the rival in all resolutions.
In Mother Nature test the situation looks really sad: the weak ALU unit NV35 inherited from the unsuccessful predecessor determined the outcome of this benchmark. This unit simply drags ASUS V9950 to the bottom and we see ATI RADEON 9800 PRO show almost double advantage over the competitor.
Of course, if we didn’t use patch 330 the results of V9950 could be completely different, but they could never be called honest and fair. I think that Mother Nature appeared a pretty illustrative benchmark revealing all weak spots of GeForce FX 5900.
This is highly complicated test from the geometrical point of view (700,000 polygons in a single frame). So, it can bring almost any contemporary graphics adapter down to its marrowbones.
Under regular workload ASUS V9950 working at the nominal frequencies goes a little behind ATI RADEON 9800 PRO, but as soon as we overclock it, our hero takes the lead. With FSAA 4x and Anisotropic Filtering 8x GeForce FX 5900 based graphics card shows all the power of the latest NVIDIA graphics solution leaving the competitor behind in all cases. Even in the hardest test ASUS V9950 performs almost as fast as RADEON 9800 PRO, despite higher level of anti-aliasing and much higher workload on the chip and memory subsystem. Since GeForce FX 5800/5900 processes pixel shaders 1.1 not too much slower than ATI competitors, and since GeForce FX 5900 also boasts pretty high 1st generation vertex shaders processing speed, the results obtained in this test look quite logical to us although we cannot disregard the fact that the drivers for this benchmark could be fine0tuned by the independent hardware vendors.
We have just tested ASUS V9950 graphics card based on the new NVIDIA GeForce FX 5900 graphics processor. We liked this product first of all for its excellent design and quality and for the smart cooling system. The 2D image quality appeared simply brilliant: the card provided crystal-clear image in all resolutions up to 1800x1440x75Hz. The overclocking potential of this product also appeared very good. We were also very pleased with the rich package: ASUS’ bundle looks very impressive especially against the background of some other manufacturers who include just a CD-disk with the drivers or stuff the package with some useless items like screwdrivers or posters.
As for the performance, the situation here is not very simple. Of course, the performance was quite high, however, the products of the kind can’t boast any stable advantage over the ATI RADEON 9800 PRO competitor. Moreover, GeForce FX 5900 can even turn out slower than the rival in some games, which use pixel and vertex shaders very actively. Nevertheless, ASUS V9950 performs very fast in the today’s games and makes the impression of a very good product overall. This graphics adapter supports all the latest features and is an excellent choice for those of you who are looking for high-performance and high-quality solution for contemporary games and applications.