by Tim Tscheblockov
04/24/2004 | 11:55 AM
Every overclocker faces a difficult task when choosing the graphics card for his/her home system today. Instead of well-ordered ranks of graphics cards with sharp separating lines between price niches we see a confusing crowd of devices where high-quality and fast products stand next to value cards, which sometimes have a higher appeal to the user due to their lower price at a tradeoff of a slightly lower performance. Some graphics cards don’t even try to look fast – they just come under loud names, allure the user with cooler highlighting and take him by fright with the futuristic shape of the heatsink.
<%BANNER[article]%>Many graphics cards are overclockable to some extent. There also exist unique graphics cards that offer you excellent overclocking opportunities or permit modification to reach the level of much more expensive products.
In our previous article, we examined top-tier expensive graphics cards and their cheaper versions. Each card underwent overclocking and, if possible, modification, like the RADEON 9800 SE. The frequencies and benchmark results that we achieved don’t pretend to be an etalon. Overclockability is of course an individual trait of each particular graphics card. I hope, though, that our testing the cards in popular games gives you more information about available products and helps you to make your shopping decisions.
In our previous review, graphics cards on the RADEON 9600 XT and GeForce FX 5700 Ultra were the bottom bar. We took them as mainstream products that could compete with “cheapened” top-end cards.
In this article, these graphics cards are on top as we are going to talk about mainstream graphics cards and their cheaper modifications. We also carry on our tradition of including graphics cards of a lower class, which look competitive against the mainstream produce.
By the way, the “lower class” means GeForce FX 5200, ATI RADEON 9200 and NVIDIA GeForce4 Ti4200-8x here. The latter two graphics processors were not originally positioned as “weak and cheap”, but after more advanced DirectX 9-comaptible GPUs appeared, their prices went down into the “value” sector. It doesn’t mean, though, that the RADEON 9200 and the GeForce4 Ti4200-8x have to lose in performance to their new neighbors in the price niche.
Well, you will see the results of the tests shortly. Right now, let’s examine graphics cards participating in today’s tests.
The fastest of ATI’s processors in the mainstream sector is represented in this review by a graphics card from ASUS:
ASUS is true to itself: the wiring of the graphics card differs from the reference design. The card supports VIVO functionality and has hardware monitoring tools (you can keep track of the GPU temperature, voltage and fan rotation speed). Intelligent control over the fan speed is supported. There’s a full set of cables and adapters and a pile of CDs enclosed with the card. There are 128MB of DDR SDRAM memory in BGA chips from Samsung with a cycle time of 3.3ns (Samsung K4D263238E-GC33). The nominal frequencies of the graphics processor and the memory chips on the A9600XT/TVD graphics cards from ASUS are 500/600 (300DDR) MHz.
This device was not willing to overclock, especially its memory. The maximum clock rates at which the A9600XT/TVD worked stably were 560/680MHz. This shouldn’t be a surprise, because the memory chips are rated for working at 600MHz at maximum and the cooler covers only the chips on the face side of the PCB.
The RADEON 9600 PRO GPU appeared earlier than the higher-performing RADEON 9600 XT and headed the mainstream chip series from ATI for a while. After the RADEON 9600 XT came out – it was in fact a slightly improved and “officially overclocked” version of the RADEON 9600/9600 PRO – graphics cards on the RADEON 9600 PRO went down in price.
This graphics processor came to us on a graphics card from Sapphire:
This card looks less imposing than the one from ASUS: an unassuming cooler, no additional cooling for the memory chips, the Rage Theater chip that supports Video-In is missing…
The nominal clock rates fully comply with the standard: 400/600 (300DDR) MHz. The card carries 128MB of graphics memory in relatively fast BGA chips from Samsung with a cycle time of 2.85ns (Samsung K4D26323RA-GC2A) and a rated operational frequency of 700MHz.
Overclocking brought ambiguous results: 425/720MHz (GPU/memory). That is, the graphics processor wouldn’t speed up well, while the memory worked at a higher frequency than on the ASUS card. It is all clear with memory: Sapphire just uses faster chips. As for the GPU overclocking, I may have been unlucky to have a bad specimen. A better cooling system would also have helped to reach higher results. The first supposition is more probable, though, as in our previous review we overclocked a RADEON 9600 XT card from Sapphire (equipped with exactly the same cooler) to 625MHz GPU.
Graphics cards on the ATI RADEON 9600 GPU differ from RADEON 9600 PRO-based ones in frequencies only. However, two cards on these two processors may differ greatly in appearance even if they come from the same manufacturer: the RADEON 9600 is for cheap products with all the ensuing consequences.
I took a RADEON 9600 card from Sapphire for my tests:
The card carries the stamp of a “cheapened product”: few details, a passive heatsink, TSOP-packaged memory. There are 128MB of DDR SDRAM onboard in chips from Samsung with a cycle time of 4ns (Samsung K4D261638F-TC40).
The nominal frequencies of the GPU and memory are 325/400 (200DDR) MHz. This card had a splendid frequency growth in my overclocking tests: 425/525MHz.
This graphics processor is the result of an attempt to make the RADEON 9600 still cheaper by cutting its memory bus in two: from 128 bits to 64 bits.
So welcome the RADEON 9600 SE graphics card from Sapphire:
You see that this card has the same PCB design as the previous one (with the original 128-bit bus), while the reduction of the bus width is done by installing twice less memory. Besides that, the GPU has an even humbler heatsink on.
The card uses memory chips from Mosel Vitelic with a cycle time of 5ns (V58C2256164SAT5B). The frequencies of the GPU and memory are 325/366 (183DDR) MHz. The overclocking results were average: 400/400MHz.
The RADEON 9200 is in fact the well-known RADEON 9000 under another name and with support of AGP 8x. Here are the basic characteristics of the chip, in brief: 4 pixel pipelines with one texture-mapping unit per each, support of DirectX 8 version 1.4 pixel and version 1.1 vertex shaders. Notwithstanding the higher index in the name, the RADEON 9200 overall loses in performance to the RADEON 9100 (which is a renamed RADEON 8500 with the same additional AGP 8x support), because the latter has similar functionality, but also has two texture-mapping units per pipeline. Thus the RADEON 9100 is faster at multi-texturing, which is used in a majority of today’s games.
That said, the RADEON 8500/9100, unlike the RADEON 9000/9200, has a status of a “Discontinued Product” – it is obsolete and is not supported by the manufacturing company.
So, the RADEON 9200 comes to us on a graphics card from PowerColor:
You may bet that this card won’t be a leader in our today tests, but it looks much better than the two previous, “cut-down” cards. Once this card belonged to a higher class – the good soldering, relatively complex PCB design, VIVO functionality (thanks to the ATI Rage Theater chip) all confirm this point. Well, we are all growing old as the ex-leaders, RADEON 9800 XT and GeForce FX 5950 Ultra, may say now after being utterly defeated by the newcomer, the NV40.
But back to the RADEON 9200. The card from PowerColor is endowed with 128MB of graphics memory in DDR SDRAM chips from Elixir with a cycle time of 5ns (Elixir N2DS12H16BT-5T). The nominal frequencies of the core and memory are 250/400 (200DDR) MHz. The GPU showed a considerable frequency growth at overclocking, up to 320MHz, but the memory was unstable even at 410MHz. As a result, the maximum stable frequencies were 320/400MHz.
The opposite camp, graphics cards on GPUs from NVIDIA Corp., is represented by several products, too. Leadtek supplied the GeForce FX 5700 Ultra graphics card with GDDR3 memory:
This card follows the reference design and in fact doesn’t differ from any other graphics card on the NVIDIA GeForce FX 5700 Ultra save for the memory type. GDDR3 on GeForce FX 5700 Ultra cards can’t make wonders on the overclocking field. NVIDIA just tries its new memory and get experienced with GDDR3 chips that have numerous advantages over old DDR2 including lower energy consumption and heat dissipation.
Memory chips are only on the face side of the card, but this is no indication of a reduction of the memory bus – they gathered the same amount of graphics memory using few higher-capacity chips (the bus is 128 bits wide; there are four 32-bit chips from Samsung with a cycle time of 2ns, Samsung K4J55323QF-GC20). The nominal frequencies of the graphics processors and memory are 475/950 (475DDR) MHz. Thus, the graphics memory works at a 50MHz higher clock rate than on cards with DDR2. This card did well in our overclocking tests, reaching 560/1100MHz frequencies.
The GeForce FX 5700 Ultra card we tested came from Chaintech:
The graphics card uses scrupulously the reference design from NVIDIA. The cool-looking copper cooler on the face side of the card takes heat off the graphics processors and memory chips. A passive heatsink covers the chips on the backside of the card.
The card carries 128MB of DDR2 memory from Samsung with a cycle time of 2.2ns (Samsung K4N26323AE-GC22). The nominal clock rates 475/900 (450DDR) MHz.
This graphics card was good at overclocking, notching 610/1000MHz – the GPU frequency was slightly higher than with the previous card, while the memory showed a smaller result.
The GeForce FX 5700 differs from the Ultra version in the frequencies of the GPU and memory. There are no differences in functionality or the memory bus width between Ultra and non-Ultra variants of the GeForce 5700. This doesn’t refer to the PCB design, though. The graphics card from Albatron is an example of this different wiring:
An extraordinary thing about the Albatron card is the weirdly-shaped cooler. I wonder whose imagination concocted this construction that looks like a turned-over skull with fans in the eyes. I have no doubt that they could achieve better cooling using these two fans, but in a traditional cooler. For example, they could have removed the combs that block airflow above and below the cooler. Well, some people may find this shape interesting, who knows?
The Albatron card carries 128MB of DDR SDRAM in Samsung chips with a cycle time of 4ns (Samsung K4D261638E-TC40).
The nominal clock rates are 425/500 (250 DDR) MHz. That is, the GPU frequency is 10% lower than in the Ultra version, while the memory frequency dropped by one third and more.
The card has average overclockability: 470/660MHz.
NVIDIA’s attempt to make the GeForce FX 5700 cheaper resulted in 5700 LE graphics cards. They have lower operational frequencies of the GPU and memory compared to products on the regular GeForce FX 5700 chip.
The GeForce FX 5700 LE graphics card comes from Leadtek:
The board carries 128MB of graphics memory in chips from Hynix with a cycle time of 5ns (Hynix HY5DU281622ET-5). The nominal frequencies are 250/400 (200DDR) MHz – in other words, we have a 40% GPU frequency reduction compared to the GeForce FX 5700, while the graphics memory now works only 100MHz slower. The overclockability of our card was really astonishing: 480/600MHz. That is, the GPU clock rate nearly doubled and exceeded the nominal frequency of the GeForce FX 5700 Ultra, while the memory frequency grew by a half.
Well, 600MHz is a really phenomenal case for memory that is rated for 400MHz, according to its model number. 480MHz for the GeForce FX 5700 LE GPU seems to be an expected result because such graphics chips don’t principally differ from the GeForce FX 5700/5700 Ultra that work at 425 and 475MHz. The high overclockability of the GeForce FX 5700 LE is just the consequence of the manufacturer’s having greatly reduced the clock rate beforehand.
GeForce FX 5600/5600 Ultra chips were precursors to GeForce FX 5700/5700 Ultra. They nearly match the newer chips in functionality, but have less powerful vertex and pixel processors. Launching the GeForce FX 5700/5700 Ultra, NVIDIA improved considerably its positions in the mainstream sector where the RADEON 9600 had ruled.
But let’s get back to the GeForce FX 5600. This graphics processor is represented by a no-name graphics card:
The card carries 256MB of DDR SDRAM from Hynix (Hynix HY5DU561622CT-5; 5ns cycle time). The nominal frequencies are 325/400 (200DDR) MHz.
The graphics processor turned to be ready for overclocking, while memory didn’t: 440/410MHz.
NVIDIA uses the “XT” suffix for denoting weakened card models, maybe in spite of ATI’s usage. The GeForce FX 5600 XT is the same GeForce FX 5600, but with a lower GPU frequency and a twice-reduced 64-bit memory bus.
This graphics processor came to us on a card from Palit:
You can see it in the snapshot that they cut down the memory bus installing only half of the memory chips. There are 128MB of graphics memory onboard (Elixir N2DS25616BT-5T; 5ns cycle time). The nominal frequencies of the GPU and memory are 235/400 (200DDR) MHz. The GPU was better at overclocking than the memory: I reached the frequency ceiling at 350/420MHz.
The GeForce FX 5200 is a DirectX 9-compatible processor originally intended for low-end graphics cards. We have an ASUS product on this chip:
The product doesn’t look “cheap” – it comes from ASUS! But anyway you may be sure that this one won’t be a record-setter in our performance tests.
We have 128MB of graphics memory from Samsung here (Samsung K4D261638E-TC50; 5ns cycle time). It normally works at 250/400 (200DDR) MHz and notched 300/525MHz at overclocking. Not bad.
The GeForce4 Ti4200-8x is a representative of the previous generation of graphics processors, a hit of the year 2002. Now it has sunk to the inexpensive products sector and loses in speed to mainstream newcomers. Still, it retains an appealing “price/performance” ratio.
We’ve got a GeForce4 Ti4200-8x card from Axle:
We’ve got small dimensions, an average cooler and a minimum of functionality here – but at a low price!
The card is equipped with 128MB of DDR SDRAM from Samsung with a cycle time of 4ns (Samsung K4D261638F-TC40). The nominal clock rates of the GPU and memory are 250/512 (256DDR) MHz. Curiously, this memory is rated for 500MHz at maximum, while the manufacturer of the card makes the chips work at 512MHz instead of using faster, but more expensive memory.
The card did well at overclocking, for its class: 300/550MHz.
Now we have shuffled our cards – let’s play?
The testbed was configured like follows:
Software:
We tested all graphics cards in two operational modes. In both cases, the graphics quality was set to the maximum in the driver (the texture quality setting in Catalyst and the “speed-performance” slider in the ForceWare panel), but we also enabled 8x anisotropic filtering and 4x full-screen anti-aliasing in the second mode. We tried to avoid other factors interfering with the results of the cards proper, so we ran all games without sound (save for Deus Ex 2 that refused to launch without sound). Vsync was disabled.
We had high speeds and high resolutions in our previous article. Now that we deal with slower products we limit ourselves with 1024x768 and 1280x1024 without FSAA and AF and 1024x768 for FSAA+AF.
We keep the same set of gaming tests, only now we use IL-2 Sturmovik: Aces and the P40_vs_Zero.ntrk track instead of IL-2: Sturmovik: Forgotten Battles. Besides that, we run IL-2 with the maximum quality settings and enabled pixel shaders. Interested to see what happened to our cards? There are some surprises…
We also use the more “real” F15-Low track instead of Aerobatic in Lock On. In Max Payne 2 we use a different scene, where the performance is not so badly affected by the CPU any more. Other tests and settings remained unchanged: we use the maximum graphics quality settings available.
Far Cry is probably the most advanced game among all included into this review. Far Cry uses vertex and pixel DirectX 9 shaders and switches to DirectX 8 shaders when version 2.0 ones are not supported. The image quality doesn’t degenerate because of that. That’s why the results of the graphics cards on the RADEON 9200 and GeForce4 Ti4200 should be regarded outside of others because they don’t support DirectX 9 and work in an “easy” mode.
Well, there are no results of the RADEON 9200 in Far Cry Demo as this graphics cards behaves incorrectly in this test. You can see it in the screenshots (the RADEON 9200 is top, the GeForce4 Ti4200 is bottom).
ATI RADEON 9200
NVIDIA GeForce 4 Ti4200
A portion of triangles in cliff models is not reproduced by the RADEON 9200 and the water surface has a terrific color that changes as the player moves.
So, the results follow:
Far Cry Demo uses extensively DirectX 9 pixel shaders, so the overall leadership of graphics processors from ATI shouldn’t be a surprise.
The results of the graphics card in this test are mostly determined by the speed of executing pixel shaders. That is, by the efficiency of the pixel processors and the clock rate of the chips. For example, the two cards on the NVIDIA GeForce FX 5700 Ultra chip with different memory types and frequencies show practically the same results, while the GeForce FX 5700 Ultra with ordinary DDR2 is faster than the board with GDDR3, because it has a higher GPU frequency and lower memory clock rate. Overclocking is most rewarding for the GeForce FX 5700 LE thanks to its doubled graphics processor clock rate.
The RADEON 9600 SE with its twice reduced memory bus and a lower memory frequency loses to the results of the RADEON 9600 very little, while the “cut-down” GeForce FX 5600 XT is slower than the “full-fledged” GeForce FX 5200.
The GeForce4 Ti4200-8x is exempt from executing DirectX 9 shaders and shows results on the same level with the GeForce FX 5700. Overclocking helps it to get even with the GeForce FX 5700 Ultra or RADEON 9600 PRO.
We didn’t test the graphics cards in Far Cry Demo enabling such heavy features as full-screen antialiasing and anisotropic filtering because the cards would work incorrectly.
Halo uses pixel shaders of versions 1.1, 1.4 and 2.0. This game doesn’t support full-screen anti-aliasing, so again we didn’t turn it on as well as anisotropic filtering.
The fastest cards on chips from both ATI and NVIDIA show similar results, but the GeForce FX 5700 Ultra based ones are slightly faster than the RADEON 9600 XT at overclocking due to their higher frequencies.
The RADEON 9600 PRO-based card added little frequency and thus profits little from overclocking, while the overclocked RADEON 9600 reaches its level.
The GeForce FX 5700 LE behaves distinctly from other cards on NVIDIA GPUs. At its regular frequencies this card is even slower than the GeForce FX 5600, but at overclocking it outperforms the GeForce FX 5700 and closely follows the GeForce FX 5700 Ultra.
The oldies RADEON 9200 and GeForce4 Ti4200 behave differently: the former doesn’t execute version 2.0 shaders, but only outperforms the GeForce FX 5600 XT and GeForce FX 5200, while the GeForce4 Ti4200-8x offers you the performance of a GeForce FX 5700/RADEON 9600 PRO.
Tomb Raider: Angel of Darkness uses pixel shaders 2.0 very broadly, but when the graphics card didn’t support DirectX 9 on the hardware level, we used the settings for shaders 1.1. At the same time, we lost some of the special effects, such as blurring of the objects, which are beyond the “camera” focus. Although the scene should now be drawn faster theoretically due to the fact that graphics processors have less work to do.
The results are another proof to the statement. The old buddies RADEON 9200 and GeForce 4 Ti4200-8x perform pretty nicely. For instance, GeForce 4 Ti4200 managed to outperform the entire family of NVIDIA based graphics cards. RADEON 9200 in 1280x1024 doesn’t work correctly in Tomb Raider game “losing” a part of the game scene, that is why we do not show the results for this graphics card in 1280x1024 resolution.
The situation between other testing participants is more than evident: the ATi based solutions are ahead, because they perform 2.0 shaders much more efficiently.
Deus Ex 2 is based on the modified Unreal Tournament engine using DirectX 8 vertex and pixel shaders that is why all graphics cards appear in equal testing conditions here.
All in all, ATI based graphics cards outperform their competitors on NVIDIA chips. The results obtained in Deus Ex are mostly determined by the fillrate and pixel shaders processing speed, that is why overclocking the graphics chip provides a more significant performance boost than overclocking the graphics memory. The biggest performance increase was obtained on NVIDIA GeForce FX 5700 LE and ATI RADEON 9600, which graphics cores overclocked best of all.
It is remarkable that GeForce 4 Ti4200-8x performs just brilliantly here running as fast as NVIDIA GeForce FX 5700 Ultra. In fact, this is not at all surprising: although GeForce 4 Ti4200-8x works at lower clock frequency than GeForce FX 5700 Ultra, it boasts two TMUs per pipeline while the latter features only one texturing unit per pipeline. Therefore, since they have 4 pipelines each, the maximum theoretical texturing speed of GeForce 4 Ti4200-8x turns out higher.
Unreal Tournament 2004 Demo doesn’t use shaders 2.0 that is why all graphics cards are in equal testing conditions. There are no problems with forced anisotropic filtering and full-screen anti-aliasing that is why we would like to offer you not only the “pure” performance results, but also the results with forced 4x anti-aliasing and 8x anisotropic filtering in 1024x768 resolution.
In “pure” performance modes the leadership belongs to RADEON 9600 XT and GeForce FX 5700 Ultra, which run almost equally fast. The overclocked RADEON 9600 PRO falls a little bit behind RADEON 9600 XT working at its nominal frequencies, and RADEON 9600 is just almost as fast as RADEON 9600 PRO. The RADEON 9600 SE with a narrow memory bus is defeated by RADEON 9200 even despite the faster graphics processor.
The overclocked NVIDIA GeForce FX 5700 and GeForce FX 5700 LE perform almost equally fast and lack a tiny bit to be able to catch up with NVIDIA GeForce FX 5700 Ultra. GeForce FX 5600 doesn’t intend to win even after we overclock it, while its analog with a narrower memory bus, GeForce FX 5200 XT, yields to the card based on a less powerful graphics processor, GeForce FX 5200, but featuring a fully-fledged 128bit memory bus.
NVIDIA GeForce 4 Ti4200-8x outperforms all graphics cards based on NVIDIA chips except the fastest ones: GeForce FX 5700 Ultra.
When we enable fully-fledged full-screen anti-aliasing and anisotropic filtering the laurels at once go to NVIDIA based solutions.
The good old RADEON 9200 and GeForce 4 Ti4200-8x give up their positions: RADEON 9200 doesn’t support multi-sampling suing a more resource hungry super-sampling instead. Ti4200-8x doesn’t have any specific architectural multi-sampling optimizations, which are typical of the more up-to-date graphics processors, and at the same time features slower anisotropic filtering algorithm. The oldies share their position with the GeForce FX 5200 and RADEON 9600 SE and GeForce FX 5600 XT with “cut-down” memory bus, because FSAA requires much more from the memory bus, so that narrower memory bus simply kills the performance in this case.
Max Payne 2 graphics engine uses pixel shaders 1.1, 1.4 and vertex shaders 1.1. All DirectX9 compatible graphics cards together with RADEON 9200 will process shader version 1.4, and GeForce 4 Ti4200-8x only shader version 1.1, because it doesn’t support shaders version 1.4 at all.
Here ATI based graphics cards appear a bit faster than NVIDIA based solutions. The biggest performance gain during overclocking is as usual observed by NVIDIA GeForce FX 5700 LE and ATI RADEON 9600.
The behavior of GeForce 4 Ti4200-8x is really remarkable: while RADEON 9200 runs fast though not outstanding, Ti4200-8x defeats all other graphics cards completely! It probably owes these results to the absence of pixel shaders version 1.4 support, which allows this solution to follow a simpler way.
You can also see it from the screenshots below. Max’ leather jacket as displayed by a GeForce 4 Ti4200-8x based system lost sheen and glare, and the faces and skin of the characters lost something like a side lightning, which made them look mode realistic. The two screenshots below are obtained on RADEON 9200 (top) and GeForce 4 Ti4200-8x (bottom). Take a look:
ATI RADEON 9200
NVIDIA GeForce 4 Ti4200
When we enable FSAA and AF, the situation remains generally the same, though the oldies give up their positions immediately. GeForce 4 Ti4200-8x runs as fast as GeForce FX 5700, while RADEON 9200 falls even behind GeForce FX 5200 because of the extremely resource-hungry super-sampling algorithm it uses in case of enabled FSAA.
Prince of Persia uses DirectX8 shaders that is why all testing participants are in equal testing conditions.
In this game NVIDIA based graphics cards outperform the competitors. The “cut-down” RADEON 9600 SE and GeForce FX 5600 XT appear at least 1.5 times slower than their fully-fledged analogues, so that they yield even to GeForce FX 5200.
When we enable full-screen anti-aliasing, the cut-down graphics cards fall even farther behind their counterparts, and RADEON 9200 and GeForce 4 Ti4200-8x lose their nice positions.
The gaming engine of Star Wars uses OpenGL, involving vertex and pixel programs corresponding to DirectX8 shaders via special extensions. ATI RADEON 9200 works not like the other testing participants in this game: there is no halo effects around the bright light sources, and the armor of the main characters looks poor. If you need an example, please take a look at the screenshots from RADEON 9200 (top) and NVIDIA GeForce FX 5600 (bottom):
ATI RADEON 9200
NVIDIA GeForce FX 5600
Now let’s check the actual performance results:
Here the NVIDIA based graphics cards are significantly ahead of the rivals. The biggest performance boost during overclocking was demonstrated by NVIDIA GeForce FX 5700 LE, GeForce FX 5700 Ultra and ATI RADEON 9600.
When we enable full-screen anti-aliasing and anisotropic filtering the overall picture remains almost the same. The “cut-down” RADEON 9600 SE is still falling behind RADEON 9600, but the results for GeForce FX 5600 XT appear more than twice as low as those of GeForce FX 5600.
RADEON 9200 uses super-sampling for FSAA implementation falls almost down to GeForce FX 5600 XT, while Ti4200-8x loses much less of its speed due to multi-sampling support.
During the tests in an updated IL-2 game version we used maximum graphics quality settings, including “Advanced” landscape level of detail, when pixel shaders are used for water surfaces.
The graphics cards participating in our today’s test session differ by the implementation of pixel shaders support, so that IL-2 determines the graphics cards functionality and hence uses different shader sets for different cards. It turned out that there are three sets like that: the first set for GeForce 4 Ti4200, which supports maximum version 1.3 shaders; the second set is for RADEON 9200 with version 1.4 shaders support, and the third set for graphics cards supporting versions 2.0 hardware shaders. The differences in image quality are visible with a naked eye in this case. Below you can see three screenshots taken from an actual benchmark track: P40_vs_Zero.ntrk. The top one belongs to NVIDIA GeForce 4 Ti4200-8x, the middle one – to ATI RADEON 9200, and the bottom one – to NVIDIA GeForce FX 5600:
NVIDIA GeForce 4 Ti4200-8x
ATI RADEON 9200
NVIDIA GeForce FX 5600
It seems to me that the worst sea surface and coast line quality was provided by NVIDIA GeForce 4 Ti4200-8x, and the best – by DirectX9 compatible graphics cards. But again, this is my subjective opinion.
Of course, the results of the graphics cards using different shaders for sea and coastline cannot be compared directly, therefore we will make our verdict about the testing participants performance separately for DirectX9 compatible cards, NVIDIA GeForce 4 Ti4200-8x and ATI RADEON 9200:
NVIDIA GeForce 4 Ti4200-8x uses its own set of pixel shaders and outperforms all other graphics cards with a huge gap, until anisotropic filtering and full-screen anti-aliasing are enabled. In this case Ti4200-8x lets NVIDIA GeForce FX 5700 Ultra and ATI RADEON 9600 XT/9600 PRO take the lead.
ATI RADEON 9200 uses a different shader set and appears between RADEON 9600 and RADEON 9600 PRO, but when FSAA and AF are enabled, it falls behind all the testing participants because of the super-sampling technology it uses. It only manages to beat GeForce FX 5600 XT and GeForce FX 5200.
Among DirectX9 compatible graphics cards ATI based solutions appear slightly ahead. The results of this benchmark are mostly determined by the pixel shader performance and fillrate, that is why the biggest performance boost was demonstrated by those cards, which graphics processors overclocked best of all. They are NVIDIA GeForce FX 5700 LE and ATI RADEON 9600. With enabled full-screen anti-aliasing and anisotropic filtering the memory bus workload grows up a lot that is why the graphics cards with narrower memory bus fell significantly behind their fully-fledged analogs.
Lock On uses DirectX8 shaders that is why all graphics cards were tested in equal conditions.
There are no surprised in Lock On results. All in all, ATI based graphics solutions managed to get slightly ahead of the NVIDIA based ones. The biggest performance boost is as usual demonstrated by GeForce FX 5700 LE and ATI RADEON 9600.
ATI RADEON 9200 and NVIDIA GeForce 4 Ti4200-8x run not bad at all, about the same as ATI RADEON 9600 and NVIDIA GeForce FX 5700 LE, but as soon as FSAA and AF are enabled, they lose their positions. ATI RADEON 9200 manages to outperform only the cut-down version of NVIDIA GeForce FX 5600 XT, while GeForce 4 Ti4200-8x falls below GeForce FX 5600 even.
The graphics engine of this strategy game loads the graphics cards in real time quite heavily, and uses DirectX8 pixel and vertex shaders.
The results of the fastest cards in this test in 1024x768 appeared limited by the CPU speed, although ATI based cards still turned out a little bit faster: well, it looks like ATI drivers send fewer calculations to the CPU.
When we moved to 1280x1024 resolution, the results appeared less CPU dependent. And the performance difference between the fastest cards again balanced towards ATI. However, as soon as we enabled full-screen anti-aliasing and anisotropic filtering, the leadership got overtaken by NVIDIA, and GeForce FX 5700 Ultra based cards successfully outperformed ATI RADEON 9600 XT, while GeForce FX 5700 defeated RADEON 9600.
The old buddy RADEON 9200 with a fully-fledged 128bit memory bus outperforms a more up-to-date RADEON 9600 SE featuring a 64bit memory bus only. But when FSAA and AF are enabled, RADEON 9200 rolls back to the very last position in the list.
NVIDIA GeForce 4 Ti4200-8x runs as fast as GeForce FX 5700, but in heavy work modes it also falls back to the level of GeForce FX 5700 LE.
Well, just like in the previous article called Winter-Spring 2004: Graphics Cards on Top Graphics Processors and Their Low-Cost Versions – Overclocker’s Best Choice, let’s say a few words about each of our today’s testing participants.
ATI RADEON 9600 XT. With disabled full-screen anti-aliasing and anisotropic filtering it outperforms its competitors on NVIDIA GeForce FX 5700 Ultra in the majority of benchmarks due to faster DirectX9 pixel shaders performance and higher VPU frequency. But in heavy modes, with enabled FSAA and AF NVIDIA GeForce FX 5700 Ultra manages to take revenge very often due to higher memory bus bandwidth and aggressive anisotropic filtering optimizations.
The overclocking potential of RADEON 9600 XT based graphics cards may be pretty high, but unfortunately, it is impossible to guarantee that. For instance, a RADEON 9600 XT based graphics cards, which we tested in our previous session worked fine at 625MHz core and 740MHz memory frequency, while our today’s piece managed to show the maximum of 560MHz chip and 680MHz memory.
The most dangerous competitors to RADEON 9600 XT are the cards on NVIDIA GeForce FX 5900 XT chips. They are based on fully-fledged NV35 GPUs, use 256bit memory bus and often turn out faster than RADEON 9600 XT even without overclocking. Note that RADEON 9600 XT and GeForce FX 5900 XT cost about the same.
ATI RADEON 9600 PRO. This solution works at a lower chip and memory frequency yielding about 15% of their performance to RADEON 9600 XT. At the same time, they cost about 20%-25% less. If you overclock your RADEON 9600 PRO, you may not be able to reach the working frequencies of the RADEON 9600 XT solution, but it will still remain an attractive purchase in its price range.
ATI RADEON 9600. This solution works at even lower frequencies and falls about 20-25% behind RADEON 9600 PRO. Due to lower clock frequency of the VPU, the RADEON 9600 based graphics cards usually boast very attractive overclocking potential. We managed to overclock our today’s RADEON 9600 sample so that its core frequency exceeded that of RADEON 9600 PRO, though the memory frequency failed to reach that of the PRO version. All in all the overclocked card pe4rformed really close to what the RADEON 9600 PRO is capable of.
So, it might make sense to buy a RADEON 9600, but only if you intend to overclock it right away. When it works at its nominal frequencies, NVIDIA GeForce FX 5700 appears faster in most cases, while the price difference between the two is not that big at all.
ATI RADEON 9600 SE. This is a cut-down version of the RADEON 9600, featuring 64bit memory bus. It is about 30-40% slower than RADEON 9600, but also about 30-4-% cheaper. It defeats its direct competitors, NVIDIA GeForce FX 5600 XT and GeForce FX 5200, in no time, but manages to outperform RADEON 9200 only in case FSAA and AF are enabled. Nevertheless, RADEON 9600 SE seems to be the most optimal solution of all low-cost DirectX9 compatible solutions.
NVIDIA GeForce FX 5700 Ultra. This is a flagman of the NVIDIA series targeted for the mainstream market. The use of GDDR3 memory doesn’t provide any advantages to the GeForce FX 5700 Ultra solution compared with the regular DDR-II, but can theoretically affect the price point. GeForce FX 5700 Ultra based graphics cards cost about the same as RADEON 9600 XT based ones. Although they lose to the latter in advanced games, they can compete with the RADEON 9600 XT pretty successfully in all other applications and games, especially with enabled full-screen anti-aliasing and anisotropic filtering. The most dangerous competitors for NVIDIA GeForce FX 5700 Ultra are GeForce FX 5900 XT based graphics cards, which are just a little bit more expensive but considerably faster at the same time.
NVIDIA GeForce FX 5700. It differs from the Ultra version by lower working frequencies of the GPU and graphics memory. You will not be able to reach the working frequency of the GeForce FX 5700 Ultra memory even if you overclock your card, because the 5700 is equipped with a much slower and cheaper memory chip. But you will anyway be able to get the GPU close to 475MHz clock. The overclocked GeForce FX 5700 was just a tiny bit slower than GeForce FX 5700 Ultra, and it could be a great argument in favor of these graphics cards, actually, if it were not for such powerful competitors as RADEON 9600 PRO and GeForce FX 5700 LE.
NVIDIA GeForce FX 5700 LE. This is an even slower solution differing from GeForce FX 5700 by much lower GPU frequency and slightly lower graphics memory frequency. Due to low nominal frequencies, GeForce FX 5700 LE demonstrates outstanding overclocking potential: the GPU frequency almost doubled as a result of overclocking. Unfortunately, I do not have any statistics at hand about the NVIDIA GeForce FX 5700 LE overclocking, but everything indicates that these cards will have excellent overclockability. For instance, the card, which took part in our today’s review after overclocking appeared just a little bit slower than NVIDIA GeForce FX 5700 Ultra.
Keeping in mind low cost of GeForce FX 5700 LE based graphics cards and their theoretically (unfortunately, there is no statistics yet) high overclocking potential, we would recommend these cards as an excellent choice in the appropriate price category.
NVIDIA GeForce FX 5600. This previous generation graphics chip yields a lot in performance to NVIDIA GeForce FX 5700/RADEON 9600 and runs as fast or slightly faster than NVIDIA GeForce FX 5700 LE. Since there are competitors like faster RADEON 9600 solutions or potentially highly overclockable GeForce FX 5700 LE, there is hardly anything positive we could say about FX 5600 today.
NVIDIA GeForce FX 5600 XT. This is a modification of the just mentioned GeForce FX 5600 with a lower GPU frequency and 64bit memory bus. The performance of this solution is very often even slower than that of GeForce FX 5200, not to mention RADEON 9600 SE. This solution is priced as high or even higher than RADEON 9600 SE and GeForce FX 5200, that is why I wouldn’t consider GeForce FX 5600 XT a good choice today.
NVIDIA GeForce FX 5200. This graphics processor is intended for the value DirectX9 compatible graphics cards. It almost always outpaces GeForce FX 5600 XT, which makes it even more attractive, bearing in mind the price, especially when we are talking about products from the “much less than $100” price range. The No.1 competitor of this solution is RADEON 9600 SE: the cards based on this chip are a little bit more expensive, but usually run much faster than the 5200.
Well, I have just expressed my opinion about the mainstream graphics solutions available in the today’s market. But it doesn’t at all mean that it is the only acceptable point of view. You have the benchmarks results in front of you. Evaluate, compare, draw your own conclusions.
I would only like to say that I didn’t express my opinion about ATI RADEON 9200 and NVIDIA GeForce 4 Ti4200-8x on purpose. You see everything in the diagrams and no comment is actually necessary. You are lucky if you have a GeForce 4 Ti4200-8x and it managed to last you until spring 2004. It means you will be able to wait until fall 2004 when NV40 and R420 based solutions for the mainstream market appear.