by Alexey Stepin , Yaroslav Lyssenko
08/13/2008 | 02:00 PM
Nvidia ruled the top-end graphics card market for a long period of time but history teaches us that no rule is permanent. The announcement of the Radeon HD 3870 X2 was the first blow at Nvidia’s position showing that two inexpensive GPUs installed on one PCB could challenge one higher-class GPU. The blow didn’t prove fatal due to the drawbacks typical of multi-chip graphics solutions. However, the reduction of the price of the Radeon HD 3870 X2 made it a peril to the GeForce 9800 GTX, the last and the fastest single-chip graphics card based on the G92 core.
On June 25, ATI introduced the new generation of the Radeon HD architecture. That was a heavier blow. The GeForce 9800 GTX was defeated by the modest Radeon HD 4850, a mainstream product costing a mere $199. Nvidia’s countermeasures smelled of despair: the developer cut the price of the GeForce 9800 GTX from $299-349 to $199 and announced the GeForce 9800 GTX+, a card with increased clock rates. The higher clock rates of the “plus” version (738/1836MHz as opposed to 675/1688MHz of the ordinary version) may make the card more competitive against the Radeon HD 4850 but what about the Radeon HD 4870? Its potential is so huge that, according to Diamond Media, it can challenge Nvidia’s flagship GeForce GTX 280 if overclocked well enough!
So the question is if overclocking the GeForce 9800 GTX can ensure a stable advantage over the Radeon HD 4850 and make the card competitive to the Radeon HD 4870? To answer this question in this review we will modify our GeForce 9800 GTX in order to control the GPU voltage and overclock the GPU frequency to the maximum.
Of course, this overclocking method is unreliable and even dangerous for the card but it will help us see the full potential of G92-based cards. This information may be useful for people who have or are going to have a GeForce 9800 GTX, Radeon HD 4850 or Radeon HD 4870.
It is simple to modify the GPU power circuit on the GeForce 9800 GTX. It uses the same controller as the GeForce 8800 GT/GTS, i.e. the Primarion PX3544 chip. What makes it even easier is that the components are placed in the same way. So, the modification is performed by connecting Pin 19 of the PX3544 chip with the ground via a trimming 1,000Ohm resistor.
A multiturn precision resistor is desired for accurate adjustment of the GPU voltage. The PX3544 has very small leads you can hardly solder anything to, so the most convenient point for connecting the resistor is marked with a red dot in the photo below. The best way to read the current voltage is from the row of ceramic capacitors located in the GPU zone on the reverse side of the PCB. You can use any of them. Here, the C666 capacitor, marked with a red dot, is employed.
We did not disable the overvoltage and overcurrent protection of the GPU because this would have increased the risk of damaging the card in the process of overclocking. We didn’t volt-mod the memory circuitry because the Anpec APW7066-based circuit required separate modifications for VDD and VDDQ voltages. It just didn’t seem worth the trouble as we didn’t expect high results from memory overclocking. The goal of this review is to pit the modified GeForce 9800 GTX against ATI’s new solutions but the memory frequency of the Radeon HD 4850 is a modest 2GHz whereas the Radeon HD 4870 with its 3.6GHz GDDR5 memory is obviously unrivalled in this respect. Moreover, the accelerated version of GeForce 9800 GTX (with the plus sign in its name) doesn’t use overclocked memory. It comes with a memory frequency of 1100 (2200) MHz in both versions.
When the driver is loaded, the default GPU frequency is 1.22V in 3D mode. Having increased it to 1.528V we watched the card pass all of our tests at GPU frequencies of 875/2188MHz (879/2214MHz as reported by RivaTuner). This is an achievement: a frequency growth of 200MHz for the main domain of the G92 core and 500MHz for the shader domain The protection system woke up when we tried to increase the GPU voltage further, so we stopped at that. We didn’t want to ruin the card after all – we wanted to benchmark it.
Click to enlarge
For the card not to overheat we set its cooler’s fan at maximum speed. That was not easy for the ears, of course. Surprisingly, the temperature of the overclocked card’s core was only 50°C in idle mode and 85-87°C under load. The developer should be given credit for creating what is perhaps the best reference cooler for modern graphics cards.
Just as we had done with the overclocked Radeon HD 3870, we measured the power consumption of the modified GeForce 9800 GTX on a special testbed configured like follows:
The 3D load was created by means of the first SM3.0/HDR test from 3DMark06 running in a loop at 1600x1200 with 4x FSAA and 16x AF. Practice suggests it is a heavier load for the graphics card than the tests of 3DMark Vantage. The Peak 2D mode was emulated by means of the 2D Transparent Windows test from PCMark05. Here are the results:
Click to enlarge
So, the peak power consumption grew by almost 70% under load reaching 182.6W. This is similar to the power draw of the dual-processor GeForce 9800 GX2 and higher than that of the GeForce GTX 280! ATI’s new solutions require much less in comparison. Thanks to the two 6-pin PCI Express 1.0 connectors, the load on each +12V line was always acceptable, never exceeding the permissible limit of 75 watts.
Now we will check out if such extreme methods can help the GeForce 9800 GTX beat the Radeon HD 4850 and 4870 in games.
For our performance tests of the overclocked Nvidia GeForce 9800 GTX we put together the following testbed:
According to our testing methodology, the drivers were set up to provide the highest possible quality of texture filtering and to minimize the effect of software optimizations used by default by both: AMD/ATI and Nvidia. Also, to ensure maximum image quality, we enabled transparent texture filtering. As a result, our ATI and Nvidia driver settings looked as follows:
For our tests we used the following games and synthetic benchmarks:
First-Person 3D Shooters
Third-Person 3D Shooters
We selected the highest possible level of detail in each game using standard tools provided by the game itself from the gaming menu. The games configuration files weren’t modified in any way, because the user doesn’t have to know how to do it. The only exception was Enemy Territory: Quake Wars game where we disabled the built-in fps rate limitation locked at 30fps. Games supporting DirectX 10 were tested in this particular mode.
Besides the modified Nvidia GeForce 9800 GTX, the following graphics accelerators also participated in our test session:
The tests were performed in the following resolutions: 1280x1024/960, 1600x1200 and 1920x1200. If the game didn’t support 16:10 display format, we set the last resolution to 1920x1440. We used “eye candy” mode everywhere, where it was possible without disabling the HDR/Shader Model 3.0/Shader Model 4.0. Namely, we ran the tests with enabled anisotropic filtering 16x as well as MSAA 4x antialiasing. We enabled them from the game’s menu. If this was not possible, we forced them using the appropriate driver settings of ATI Catalyst and Nvidia ForceWare drivers.
Performance was measured with the games’ own tools and the original demos were recorded if possible. Otherwise, the performance was measured manually with Fraps utility version 2.9.1. We measured not only the average speed, but also the minimum speed of the cards where possible.
This game doesn’t support display resolutions of 16:10 format, so we use a resolution of 1920x1440 pixels (4:3 format) instead of 1920x1200 for it.
The modified GeForce 9800 GTX proves to be as fast as the Radeon HD 4870 and Radeon HD 3870 X2, which is good. But you should keep it in mind that the modified card produces too much noise, consumes too much power and generates too much heat. And there is the constant risk of the GPU just burning out.
BioShock doesn’t support FSAA when running in Windows Vista’s DirectX 10 environment. We benchmark graphics cards without FSAA in this game.
The extremely overclocked card from Nvidia is successful in BioShock but the ordinary Radeon HD 4870 at default frequencies is just as fast as it. It is clear that the off-the-self GeForce 9800 GTX+, which has lower GPU frequencies, won’t be competitive to the latter, except being cheaper. Moreover, the speeds are so high here that there is no real need to overclock anything. The Radeon HD 4850 delivers a comfortable frame rate for only $199.
The extremely overclocked GeForce 9800 GTX is barely ahead of the Radeon HD 4850 at 1280x1024 although the performance growth amount to almost 25%.
The more advanced Radeon HD 4870 and Radeon HD 3870 X2 are unrivalled but even they cannot provide really comfortable conditions for the gamer.
It is like in the previous test but the speeds are overall higher. The modified GeForce 9800 GTX equals the Radeon HD 4870 at 1280x1024 but slows down to the level of the junior Radeon at the higher resolutions.
This game is tested at the High level of detail, excepting the Shaders option which is set at Very High This way we try to achieve a compromise between image quality and speed.
It is the first time when our extreme overclocking is rewarding. The modified GeForce 9800 GTX becomes the fastest card in this test. It would be perfectly comfortable to play the game on that card if it were not for the rather low bottom speed, 20fps only. This is a win anyway, even though it comes at a high price. You just can’t use the card for long in such a harsh operation mode.
The frame rate is fixed at 30fps in this game as this is the rate at which the physical model is being updated at the server. Thus, this 30fps speed is the required minimum for playing the game.
The modified GeForce 9800 GTX has a 4% lead over the Radeon HD 4870 at 1280x1024 but the faster GDDR5 memory helps the latter win at the higher resolutions.
The Radeon HD 4850 is outperformed by the overclocked GeForce at 1280x1024 but equals it at the higher resolutions.
This is the second test where the modified GeForce 9800 GTX becomes the leader but the overclocking has no practical value again. In Crysis the card had too low performance. In Episode Two its performance is already high enough without any overclocking. So, the gamer gets no practical benefits from the increased performance of the overclocked card here. You just can’t notice that the GeForce 9800 GTX+ runs this game faster than the Radeon HD 4870.
The game doesn’t support FSAA when you enable the dynamic lighting model, but loses much of its visual appeal with the static model. This is the reason why we benchmarked the cards in S.T.A.L.K.E.R. using anisotropic filtering only.
It is the third time when Nvidia’s solutions are beyond competition but the engine of this game is known to be optimized for Nvidia’s chips. Anyway, the 30% growth in average performance at a resolution of 1280x1200 is quite a nice thing.
The best the modified GeForce 9800 GTX can do is to compete with the Radeon HD 4850 in average speed, being slightly inferior in bottom speed. Not very impressive, especially considering the drawbacks of extreme overclocking.
This is another impractical win for the overclocked card. The off-the-shelf Radeon HD 4870 provides comfortable conditions, too. The Radeon HD 4850 is also good, even at a resolution of 1920x1200, because you don’t need as high a frame rate to play a third-person shooter as you need to play a first-person one.
The overclocked GeForce 9800 GTX can’t do anything about ATI’s new-generation cards. It is only at 1920x1200 that the GeForce outperforms the Radeon HD 4850 by 8-9%. The overclocked GeForce 9800 GTX delivers comfortable performance at the highest resolution but you can’t use the card for long in such a harsh operation mode. As for the GeForce 9800 GTX+, we guess it will be slower.
The game loses much of its visual appeal without HDR. Although some gamers argue that point, we think TES IV looks best with enabled FP HDR and test it in this mode.
The modified GeForce 9800 GTX delivers a frame rate of over 100fps in closed environments and over 80fps in open scenes but its bottom speed is not far different from that of the RV770-based cards and you don’t actually need such a high frame rate for playing the game. In other words, the effect from the volt-modding and overclocking is not worth the trouble. Talking about the GeForce 9800 GTX+, it should feel somewhat more comfortable in open scenes of this game at high resolutions than the ordinary version of the card.
The new add-on to Company of Heroes is tested in DirectX 10 mode only since it provides the highest quality of the visuals.
The extreme overclocking is rewarding here. The overclocked card delivers the best average frame rate at resolutions up to 1600x1200 inclusive. The bottom speed is not high, though. Even at GPU frequencies of 875/2188MHz, the bottom speed of the GeForce 9800 GTX is lower than that of the Radeon HD 4870 and Radeon HD 3870 X2 working at the default clock rates.
The bottom speed of a graphics card is no less important than the average frame rate, so the results of the overclocked GeForce 9800 GTX have little practical worth. We can also say that the GeForce 9800 GTX+ won’t be able to provide comfortable conditions in this game.
The add-on to C&C 3: Tiberium Wars brought no changes into the technical aspect of the game. The game still having a frame rate limiter, you should consider the minimum speed of the cards in the first place.
Command & Conquer 3 enjoys deserved popularity but doesn’t make a good benchmarking tool due to the mentioned peculiarities. As you can see, every graphics card hits the speed limit while their bottom speed is never lower than 24fps even at 1920x1200.
As opposed to Company of Heroes, it is the overclocked GeForce 9800 GTX that is the closest to delivering a comfortable speed in the tech demo as it has a higher bottom speed than the Radeon HD 4870. Alas, the overall performance growth is rather small. And the off-the-shelf GeForce 9800 GTX+ is going to be even slower.
The modified and overclocked GeForce 9800 GTX working at 875/2188MHz can’t overtake the Radeon HD 4870. Nvidia’s solution may only be superior in the SM2.0 tests.
That’s right: the overclocked GeForce 9800 GTX is ahead of the Radeon HD 4870 in the SM2.0 tests but the difference is small even then. ATI’s solution is far better in the SM3.0/HDR tests.
The individual SM2.0 tests prove Nvidia’s victory which is the most conspicuous in the second test that favors neither ATI’s nor Nvidia’s architecture.
The overclocked GeForce managed to outperform the Radeon HD 4870 by 16% in the second SM3.0/HDR test but the latter wins the first test, being only second to the dual-chip Radeon HD 3870 X2. Don’t forget that the Radeon HD 4870 is an off-the-shelf card working at its default frequencies and voltage.
We minimize the CPU’s influence by using the Extreme profile (1920x1200, 4x FSAA and anisotropic filtering).
Increasing the clock rate doesn’t help much in 3DMark Vantage. The modified GeForce 9800 GTX proves that the GeForce 8/9 architecture has passed its prime and should now be viewed as outdated. The best it can do is to close the gap to the Radeon HD 4850. There is no talking about competition with the Radeon HD 4870.
The individual tests agree with the overall scores except that the modified GeForce 9800 GTX overtakes the Radeon HD 4850 in the second test. That’s not the result you would expect from a graphics card working at an increased GPU voltage, producing a lot of noise, and consuming as much power as the GeForce 9800 GX2!
The GeForce 9800 GTX graphics card proved to have high overclocking potential thanks to its 12-layer PCB and thought-through design. The extreme overclocking pushed the performance of the G92 chip up to the next level, enabling it to compete with more advanced solutions. Unfortunately, these performance benefits are still not enough. Let’s check out the summary diagrams.
The graphics card is often not the main performance-limiting component at a display resolution of 1280x1024. Therefore the performance gain from overclocking the core of the GeForce 9800 GTX to 875/2188MHz looks small. The card beats the Radeon HD 4870 in seven out of the 16 tests (the card is tested twice in TES IV: Oblivion – in open and closed scenes). And there is no practical gain in six out of these seven tests. It is only in Crysis that the speed increase is apparent to the naked eye An off-the-shelf GeForce 9800 GTX+ wouldn’t deliver even such performance benefits and wouldn’t be a serious opponent to the Radeon HD 4870. It may even have little advantage over the Radeon HD 4850.
ATI’s new cards are stronger at 1600x1200, the overclocked GeForce 9800 GTX looking even humbler against the Radeon HD 4870 than at 1280x1024. From a practical point of view, only one out of the six wins of the overclocked card is valuable (in S.T.A.L.K.E.R.). Purchasing the junior Radeon seems a wiser choice than overclocking your GeForce 9800 GTX as you have to modify and overclock the latter too hard and too dangerously to achieve tangible benefits.
As the resolution grows higher, the numbers of tests won by the overclocked GeForce 9800 GTX decreases to five. Like in the previous case, the only real win is S.T.A.L.K.E.R. On the other hand, the Radeon HD 4850 can deliver acceptable performance everywhere the modified GeForce 9800 GTX can, excepting the same S.T.A.L.K.E.R., but without any overclocking. The games that prove too heavy for the Radeon, namely Call of Juarez, Crysis and Lost Planet, prove too heavy for the GeForce 9800 GTX, too.
Our experiments didn’t go without consequences for our GeForce 9800 GTX. The card began to hang the system up after a while. The graphics core must have been irrevocably damaged somehow during the tests at the increased voltage. That’s why extreme overclocking is a dangerous sport – for the equipment if not for the sportsman himself (well, liquid nitrogen may be harmful to the overclocker, too).
So, although the volt-modding and overclocking of the GeForce 9800 GTX ensures a considerable performance growth, it brings about but small practical benefits even if you disregard the greatly increased power consumption and heat dissipation. If you could run the overclocked card continuously, the Radeon HD 4870 would still be a better buy. The Radeon HD 4850 is also good across a majority of tests. The GeForce 9800 GTX+ has lower GPU clock rates than the ones we achieved with our 9800 GTX and would have lower performance. Considering that it is more expensive than the Radeon HD 4850, we don’t think it a good buy.
As for the extreme overclocking thing at large, it is still only good for enthusiasts seeking for new records. Graphics cards can’t be used continuously in such a harsh operation mode just as our failed card confirmed. Extreme overclocking is a hobby, and not a cheap one we should guess, but some people appreciate a new performance record more than the money spent to achieve it.