by Alexey Stepin
10/24/2003 | 09:00 PM
Some time ago, our test lab received another specimen of a mass graphics card, Gigabyte RADEON 9600 PRO. The company’s specifications list it as GV-R96P128D. As we promised in our FIC R96P Review, this time we are going to investigate the influence of full-screen anti-aliasing (FSAA) and anisotropic filtering (AF) on the performance of mass-market graphics cards. But first, let us take a closer look at Gigabyte’s product.
The Gigabyte GV-R96P128D came to us in full dress, in its retail package. To my taste, however, the creative potential of the designers was asleep when they made the thing. Well, it follows the company’s “ancient civilizations” theme, but is too insipid. In fact, it resembles the package of Gigabyte Maya II R9200 cards a lot, save for minor details.
I guess they could have presented the ancient Maya theme more esthetically. Anyway, the package enumerates all the main features of the product, including the memory amount – 128MB. Having opened the box, I found the following:
Unlike ordinary “no-name” adapters and cables, the DVI-I-to-D-Sub adapter and the universal TV-out cable from Gigabyte have the company’s name on them:
Besides, the DVI plug is covered with a protective lid. It is a trifle, of course, but nonetheless a pleasing one. The cable is rather unusual, too. Many graphics cards come with an S-Video-to-RCA adapter and two respective cables. Gigabyte accompanies its GV-R96P128D with a cable, which has an S-Video connector on one end and then splits into two tails with S-Video and RCA connectors. This is handy, since you don’t have to worry about adapters. The drawback is that one of the tails is always unused and is just hanging aimlessly. The software bundle is not too huge, but includes a few games. That’s better than nothing.
The drivers CD turned to be a recorded CD-R disk, because we have got a pre-sale version of the product. This is not critical, though. The GV-R96P128D will be coming into retail with a printed CD, while we are going to use the latest official Catalyst driver from ATI.
The deep-blue-colored PCB is nearly an exact copy of ATI’s reference design, which may mean good 2D quality and moderate overclocking potential. The cooler is a solid-looking thing. It resembles the ones ASUS mounts on some of its graphics cards.
Don’t let the color mislead you: the heatsink is made of painted aluminum. The fan’s blades are supposed to shine in ultraviolet light.
The PCB carries eight memory chips from Samsung, marked as K4D263238E-GC2A. In Samsung’s reading, 2A stands for 2.8ns access time. The memory works at 600MHz (300MHz DDR), while the GPU – at 400MHz. In the upper part of the board, you can see a caption with the name of the card, the Gigabyte logo, and the revision number. We have a card featuring revision 1.0.
Compared to FIC R96P, the GV-R96P128D has a noticeably louder cooler. At least, the difference between them is perceivable by the ear. On the other hand, it does its job better than the humble cooler of the FIC R96P. The result is more successful overclocking. After removing the thermal pad (hated by all overclockers, I presume), I managed to speed up the core to 510MHz. Excellent! The memory couldn’t keep the tempo, though, as it allowed speeding it up to 680MHz (340MHz DDR) only. It didn’t even make it to its own specs – 700MHz (350MHz DDR). Any other 10MHz above that resulted into all sorts of hideous artifacts in the onscreen picture. As for 2D quality, it didn’t disappoint. The card produced a good picture in all resolutions up to 1600x1200@85Hz. Thus, the only drawback of the Gigabyte GV-R96P128D I’ve found so far is a relatively noisy, although quite efficient, cooler.
Our today’s tests will show the effect of full-screen anti-aliasing and anisotropic filtering on mainstream graphics cards. Thus, we will be able to estimate the cards efficiency in each work mode. An NVIDIA GeForce FX 5600 Ultra graphics card, working at 400/800MHz core and memory frequencies, will be the reference point for the Gigabyte GV-R96P128D.
The testbed configuration has changed somewhat since our last review. Now it consists of :
We used the following benchmarking software:
The graphics quality was set to maximum in every test. We ran the tests three times each: in “default” mode, with FSAA 4x and with FA 8x. We didn’t adjust the texture quality settings as these provide a too small performance gain with a RADEON GPU. A few extra fps aren’t worth the quality loss.
The GeForce FX 5600 Ultra is ahead of Gigabyte GV-R96P128D, which works at lower frequencies. When we amend this by overclocking, the Gigabyte wins in high resolutions.
The GV-R96P128D is most efficient at FSAA. It is running abreast of the GeForce FX 5600 Ultra. When overclocked, the RADEON leaves the GeForce behind.
We see the same picture here. It is a known fact that the anisotropic filtering algorithm implemented by ATI Technologies in its VPUs works faster than the one implemented in NVIDIA GPUs, but at the same time it provides somewhat poorer image quality.
Higher frequencies help GeForce FX 5600 Ultra to win in higher resolutions. The overclocked RADEON 9600 is also not bad, though.
The results look very similar to what we saw in Quake 3. However, GeForce FX 5600 Ultra performs better in this game. The overclocked GV-R96P128D is beyond competition.
With anisotropic filtering enabled, the GeForce FX 5600 Ultra runs faster than RADEON 9600 Pro working at the nominal frequencies, but overclocking pushes the RADEON on top.
Note also that the new version of NVIDIA’s Detonator seems to have got rid of the problems with Return to Castle Wolfenstein. We don’t see the deplorably low results as it was in case of Detonator 44.03.
Again, the RADEON lacks working frequency and is only competitive when overclocked.
The same is true for the FSAA modes, although the GeForce FX 5600 Ultra wins in 1600x1200 here.
With enabled anisotropic filtering, the overclocked GV-R96P128D is again the best, while at its regular frequencies it is noticeably behind the GeForce FX 5600 Ultra.
Serious Sam uses a gaming engine from Croteam, which is more complex and advanced than ID Software’s. The participants performed as follows: the overclocked RADEON 9600 Pro took the first place, when it was working at its nominal frequencies, it arrived the second, and the GeForce FX 5600 Ultra was only the third.
No comments. The above-described picture didn’t change when we enabled FSAA. The non-overclocked GV-R96P128D is behind the GeForce FX 5600 Ultra in 1600x1200, but by 1fps only. This may well be considered a measurement error.
It’s still the same. The Gigabyte GV-R96P128D card goes neck and neck with the GeForce FX 5600 Ultra at its regular clock-rates and breaks ahead at overclocked frequencies. Note also that the RADEON uses a higher anisotropy level!
The Gigabyte card is again a little behind the GeForce FX 5600 Ultra, but strikes back as soon as it gets overclocked.
The GV-R96P128D is better than the GeForce FX 5600 Ultra in 1280x1024, but loses its ground in 1600x1200 even when working at overclocked frequencies. It seems that the higher performance of the graphics memory subsystem of the GeForce FX 5600 Ultra was crucial here.
Putting aside the overclocking results, the two cards are performing similarly.
Flyby Antalus shows an inexplicable favor towards NVIDIA’s GPUs. I wonder if the sly software developers have tweaked anything here, considering this demo is often used by hardware reviewers for benchmarking purposes.
Overclocking helps Gigabyte’s baby this time. Anyway, the 800MHz memory of the GeForce FX Ultra card makes the final blow in 1600x1200: the GeForce is faster than the RADEON by about 30%.
Again, the GV-R96P128D is competitive only when it is overclocked.
We don’t include the results for the FSAA 4x mode because of the well-known problem with GeForce FX GPUs.
There is nothing much to comment upon. The card that processes shaders faster becomes the winner in this test, and the GeForce FX 5600 Ultra GPU is not too good at that.
The GeForce FX shows excellent results in the “default” mode and in case of enabled anisotropic filtering. When FSAA is on, it somewhat loses to the overclocked GV-R96P128D.
The rankings in the second game tests are different. Gigabyte’s card feels confident in the “pure” and FSAA modes, save for 1600x1200 resolution where it loses even when overclocked. As for the anisotropic filtering mode, the RADEON is slower than the GeForce FX 5600 Ultra. At overclocked frequencies, however, it comes ahead, again save for 1600x1200 resolution.
The GV-R96P128D runs the third game test better with FSAA rather than with anisotropic filtering.
This time the RADEON 9600 Pro is again good in the FSAA 4x mode, but falls behind the GeForce FX 5600 Ultra in the anisotropic filtering mode.
Overall, the Gigabyte GV-R96P128D card handles full-screen anti-aliasing better than anisotropic filtering.
The results of the first game test prove the efficiency of the RADEON 9600 PRO based graphics card at full-screen anti-aliasing. In other cases, the GV-R96P128D suffers from lower working frequency. You can see it from the performance gain provided by overclocking.
The GeForce FX 5600 Ultra wins the second game test, although the overclocked GV-R96P128D catches up with it in the “default” mode.
In the third game test, the GV-R96P128D manages to perform worse with enabled FSAA, although goes abreast of the GeForce FX 5600 Ultra in the “default” mode. With enabled anisotropic filtering, the overclocked RADEON 9600 Pro is slightly better than the competitor.
This is the same situation as we saw in Splinter Cell. The graphics processor that handles shaders better is the leader. In our case, it is the RADEON 9600 Pro represented by the Gigabyte GV-R96P128D graphics card.
The results of the GeForce FX are in between those of the GV-R96P128D working at nominal and overclocked frequencies.
The GeForce FX 5600 Ultra wins in the FSAA mode, except 1280x1024 resolution.
After turning on anisotropic filtering, the GV-R96P128D easily outperforms the competitor even at regular frequencies.
This new benchmark is a sophisticated one as far as graphics complexity is concerned. It features scenes with a high overdraw parameter and uses pixel shaders a lot. Overall, this is a next-generation benchmark for graphics cards.
The next-generation nature of the test is evident from the low results we have in the “default” mode. Only the high frequencies of the GeForce FX 5600 Ultra help it here.
When FSSA is on, the gap between the cards is wider, while Gigabyte’s solution retains the leadership. However, it becomes small again in 1600x1200.
The same is true for the modes with enabled anisotropic filtering . Gigabyte GV-R96P128D defeats GeForce FX 5600 Ultra in no time.
What do we learn from our today’s testing? If we were to consider the “default speed” mode only, the GeForce FX 5600 Ultra GPU might have been called the winner. This is quite natural, since this graphics card has memory working at 800MHz (400MHz DDR). Still, modern games like Splinter Cell or Aquamark 3 need a more efficient GPU architecture, namely shaders processing speed. GPU and memory frequencies alone are not very crucial in such applications. Here, the Gigabyte GV-R96P128D is showing simply excellent performance. Engineers from ATI Technologies have accomplished what NVIDIA developers haven’t.
The R350 architecture (as represented by Gigabyte GV-R96P128D) looks even more advantageous when full-screen anti-aliasing or anisotropic filtering are enabled. This card often catches up with the higher-frequency competitor or even outperforms it in these modes. However, the Gigabyte card lacks higher working frequencies really dramatically, especially for the graphics memory. You have seen what makes us think so: overclocking boosts the performance of the card significantly. I guess that the 0.13micron manufacturing technology allows the frequencies above 400MHz, while faster memory would help RADEON 9600 Pro based cards outperform the competitor (GeForce FX 5600 Ultra) in all the tests and resolutions.
As for this particular card, I liked it a little less than the FIC R96P mostly because of the louder cooler. On the other hand, this cooler is more appropriate for overclocking than the one we saw on FIC’s card. Otherwise, Gigabyte GV-R96P128D is an ordinary RADEON 9600 PRO based graphics card with all the bonuses and shortcomings of this VPU. It works well in current applications and is quite ready for the new generation of computer games, which we see approaching. The accessories coming with the product are not bad at all. Gigabyte made sure that the user does not get too bored and included a few games and a software DVD-player. The assembly quality of the card is certainly up to the mark; at least, we didn’t reveal any problems when we were testing it.
This product would be a good buy for those who are looking for an inexpensive graphics card that would provide acceptable performance in modern shader-hungry computer games.
Highs:
Lows: