by Alexey Stepin , Yaroslav Lyssenko
10/14/2010 | 10:19 AM
With the recent announcement of its new GF106 graphics processor and the GeForce GTS 450 card based on it Nvidia has finally got rid of outdated technologies and architectures in all of its gaming solutions. Forcing the good old GeForce 8 architecture into retirement, the new graphics card has proved to be highly successful. As our tests suggest, it is competitive to the Radeon HD 5770 and even occasionally to the Radeon HD 5830. That is an excellent performance for a $129 product and a serious claim to superiority in this price segment.
Like the more advanced Fermi solutions, the GeForce GTS 450 offers full support for Nvidia SLI technology that allows combining several graphics cards into a single multi-GPU subsystem. Such configurations are not very popular unless in extremely fast gaming platforms that accommodate up to three GeForce GTX 480 or four Radeon HD 5870 cards. Negative factors like power consumption, which is very high, and noise, which is often unbearable if you don’t wear headphones, are not important for users of such platforms. They just want to have the highest speed possible whatever the price or other inconveniences may be. In some ways, these are just demo configurations highlighting the benefits of a particular graphics architecture rather than gaming computers for everyday use.
On the other hand, benchmarking multi-GPU solutions built out of mainstream graphics cards yields interesting data that we can analyze to predict the performance of future products from AMD and Nvidia. For example, our recent tests of a GeForce GTX 460 SLI tandem helped us understand what we can expect from the proposed dual-processor card, a successor to the GeForce GTX 295, if it ever comes out. Benchmarking results of inexpensive multi-GPU solutions may also be interesting to users who are planning to upgrade their graphics subsystem in two steps. This is quite easy nowadays since mainboards with two PCIe x16 slots are very common.
So these are the reasons for us to benchmark two GeForce GTS 450 cards in SLI mode. This tandem has the following characteristics compared to higher-class single-GPU solutions:
From a technical standpoint, the GeForce GTS 450 SLI tandem is somewhere in between the GeForce GTX 460 1GB and GeForce GTX 470, being somewhat ahead of both in terms of texturing speed. This tandem looks a well-balanced solution, yet the release of a new GeForce GTX 460 version with 384 active ALUs and 64 texture-mapping processors would make it unnecessary. Nobody would bother to take the trouble of messing with two cards, connecting bridges and driver settings if he can have the same performance from a single graphics card for the same or even less money.
On the other hand, if you already have one GeForce GTS 450 and your mainboard has a second PCIe x16 slot and supports SLI technology, you may indeed consider purchasing a second such card to nearly double your computer's graphics performance at rather low cost.
So, we are going to perform such tests in this review. We will also discuss a new version of the GeForce GTS 450 card produced by ASUSTeK Computer. Being a TOP series product, it claims to be the fastest GeForce GTS 450 available!
The packaging of this card is designed like the box of the ENGTX460 TOP we reviewed recently.
This box is quite informative. You can learn that the card is pre-overclocked to a GPU frequency of 925 MHz. You can then infer that the shader domain frequency, which is twice the main frequency in the GeForce GTS 450, is 1850 MHz. Thus, we’ve got a very fast GF106-based solution here. You can also learn the type and amount of graphics memory from the text on the box. The card supports AIWI technology which means that you can use an iPhone as a Bluetooth controller. This is not a unique feature, though. You can use Android-based smartphones and tablets as a wireless keyboard or joystick just as well. The specific graphics card model has nothing to do with that.
Like with the ENGTX460 TOP, there is some marketing info on the back of the box like “20% cooler, 35% quieter, 50% faster” without any mention of the test conditions.
The packaging is not as gorgeous inside as with top-end cards from ASUS. The graphics card is wrapped into an antistatic pack and lies in an ordinary white cardboard box together with its few accessories.
The card is equipped with all modern interfaces, except for DisplayPort. The only adapter you may need is a power adapter for old PSUs that don’t have a 6-pin connector, and you can find it included into the box. There are no special bonuses among the accessories, but that’s normal for the price category the GeForce GTS 450 belongs to.
We don’t have any complaints about the packaging and accessories of the ASUS ENGTS450 DirectCU TOP. Now let’s take a closer look at the card to find if it has any special features.
This graphics card looks like a scaled-down copy of the ASUS ENGTX460 DirectCU TOP we reviewed earlier. This is quite natural as both belong to the same series and represent the fastest models in their respective product lines.
Of course, being a less advanced solution, the ENGTS450 TOP looks somewhat simpler and more compact. Like on its senior cousin, the cooler casing is larger than the PCB. With the cooling system taken off, we can see the following:
First of all we can see a 4+1 power system which is more powerful than the one of the Zotac GeForce GTS 450 AMP! There is a 4-phase voltage regulator rather than a 3-phase one like on most other GeForce GTS 450 products. Its power transistors are covered with a small heatsink while the components of the memory voltage regulator lack such cooling. Interestingly, ASUS doesn't use its hyped tantalum capacitors everywhere on the ENGTS450 TOP, obviously due to their high cost. You can spot three such capacitors on the reverse side of the PCB whereas the face side carries a row of traditional and cheap polymer capacitors.
The intelligent part of the voltage regulators is represented by two chips from uPI Semiconductor. The uP6206 and uP6101 are responsible for GPU and memory power supply, respectively.
Both controller are located on the reverse side of the PCB, the uP6206 being even covered with a rather heavy aluminum plate. An elastic thermal pad serves as a thermal interface for them.
Like the earlier reviewed card from Zotac, this one has two empty memory chips on each side of the PCB. It indicates that the theoretically possible 192-bit memory access is not implemented here.
There are eight 1Gb GDDR5 memory chips on board. They are manufactured by Samsung Semiconductor and, despite the TOP suffix in the product name and unlike the memory of the Zotac GeForce GTS 450 AMP!, these are K4G10325FE-HC05 chips rated for a frequency of 1000 (4000) MHz. This is also the frequency the memory chips are actually clocked at by this card, delivering a peak memory bandwidth of 64 GBps. As usual, we have two power-saving modes in which the memory frequency is lowered to 324 (1296) MHz or to 135 (540) MHz.
The GPU of our sample of the ASUS card was manufactured on the 26th week of 2010. The marking is not indicative of any culling although the graphics core is pre-overclocked quite greatly compared with the reference card’s parameters. Its frequency is 925 MHz, the shader domain frequency being 1850 MHz, which is an indication of high frequency potential of the GF106 chip. This is in fact a record for the Fermi architecture, even though it is not too impressive considering the relative simplicity of the GF106. As you can learn from our reviews, we managed to overclock a GF104-based card to 900/1800 MHz after software-based volt-modding. As a matter of fact, ASUS used volt-modding, too. The GPU voltage of the ENGTS450 TOP is as high as 1.125 volts in 3D applications. This is going to increase the card's power consumption as we will check out in the next section of this review. The GPU configuration is standard with 192 ALUs, 32 TMUs, and 16 RBEs. The card lowers its GPU frequencies to 405/810 and 51/101 MHz in its two power-saving modes.
The ASUS ENGTS450 TOP is somewhat more conservative than the Zotac GeForce GTS 450 AMP! in terms of interfaces as it is equipped with a 15-pin D-Sub connector. This port is hardly necessary these days since analog connection can be established with an appropriate DVI-I adapter. The card supports SLI and Surround Gaming technologies (the latter in 2D mode only, without stereo glasses). It can output audio over HDMI in all modern formats (including multichannel HD audio).
The cooling system represents a simplified version of the ASUS ENGTX460 TOP’s cooler.
The heatsink is smaller, the fan is 80 millimeters large and has a simpler impeller design, and there are only two instead of three heat pipes (but they are still 8 rather than 6 millimeters in diameter). The direct-touch technology is implemented properly: the pipes are finished well and fitted seamlessly into the grooves of the cooler's aluminum base. The trace left by the thermal grease indicates that the pipes have full contact with the surface of the GF106 die. This is especially important since this chip lacks a heat-spreading cap like the GF100 or the GF104. The overall quality of the cooling system is just as high as that of its senior cousin, and its performance should be sufficient to cool even a greatly overclocked GF106. Let’s check this out right now, though.
As we've found out already, the high GPU frequency of the ASUS ENGTS450 DirectCU TOP is largely due to its increased GPU voltage. Therefore, we are very interested to learn how much power the pre-overclocked card needs. We measured this on our specially configured testbed.
The new testbed for measuring electric characteristics of graphics cards uses a card designed by one of our engineers, Oleg Artamonov, and described in his article called PC Power Consumption: How Many Watts Do We Need?. As usual, we used the following benchmarks to load the graphics accelerators:
Except for the maximum load simulation with OCCT, we measured power consumption in each mode for 60 seconds. We limit the run time of OCCT: GPU to 10 seconds to avoid overloading the graphics card's power circuitry. Here are the obtained results:
Interestingly, the ASUS ENGTS450 TOP has the lowest power consumption in idle mode. Its result of 7.5 watts is only half as high as that of the Radeon HD 5770. The factory volt-modding shows up in 3D mode, though, where the ASUS card's power consumption grows up to 115 watts. This is comparable to the higher-performance Radeon HD 5850 which needs 122 watts. Thus, this GeForce GTS 450 model cannot be counted among the most economical ones, but that's the tradeoff for the TOP suffix in its product name.
The increased GPU voltage couldn’t but affect the card’s heat dissipation. The ASUS ENGTS450 TOP is hotter than some GeForce GTX 460 models which deliver much higher performance. And being incomparably faster in 3D applications, the ATI Radeon HD 5870 is a mere 1°C hotter! On the other hand, the GPU temperature of 74°C is quite a low result for today, so you shouldn’t worry about overheat unless the ENGTS450’s cooler is blocked by a neighboring add-on card.
There is a problem with the card’s noisiness. While its senior cousin ENGTX460 TOP used to stand out among other system components with its louder but constant noise, the ENGTS450 is set up in such a way that its fan is constantly changing its speed under load. As a result, the variations in sound are distracting and more irritating than a constant noise would be.
Our ENGTS450 DirectCU TOP refused to overclock any further because it is already pre-overclocked to 925 MHz with increased voltage. Therefore we will only benchmark it at its factory frequencies. To perform our tests in SLI mode with another GeForce GTS 450 we dropped the ASUS card's frequencies to those of the reference card.
The GeForce driver identified the two cards easily and enabled SLI mode without any problems, so we could proceed to our tests.
We are going to investigate the gaming performance of Nvidia GeForce GTS 450 SLI graphics cards configuration using the following universal testbed:
We used the following ATI Catalyst and Nvidia GeForce drivers:
The ATI Catalyst and Nvidia GeForce graphics card drivers were configured in the following way:
Below is the list of games and test applications we used during this test session:
First-Person 3D Shooters
Third-Person 3D Shooters
Semi-synthetic and synthetic Benchmarks
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 ordinary user doesn’t have to know how to do it. We ran our tests in the following resolutions: 1600x900, 1920x1080 and 2560x1600. Unless stated otherwise, everywhere, where it was possible we added MSAA 4x antialiasing to the standard anisotropic filtering 16x. We enabled antialiasing from the game’s menu. If this was not possible, we forced them using the appropriate driver settings of ATI Catalyst and Nvidia GeForce drivers.
Besides Nvidia GeForce GTS 450 SLI, we also tested the following solutions:
Performance was measured with the games’ own tools and the original demos were recorded if possible. We measured not only the average speed, but also the minimum speed of the cards where possible. Otherwise, the performance was measured manually with Fraps utility version 3.2.3. In the latter case we ran the test three times and took the average of the three for the performance charts.
This game runs in DirectX 11 mode with the highest graphics quality settings.
The GeForce GTS 450 SLI tandem is almost as fast as the single GeForce GTX 470. The tandem is cheaper (it costs twice the price of a single GTS 450, i.e. about $260 whereas a GeForce GTX 470 costs $300 or more) but the small difference in price is hardly worth the trouble of building and maintaining a multi-GPU configuration.
The SLI tandem overtakes the GeForce GTX 470 and is close to matching the Radeon HD 5870. Again, the lower price is the main advantage of the multi-GPU configuration while its lower usability and compatibility are its main downsides. Besides, its performance at 2560x1600 is rather poor.
We don’t see anything new here. The GeForce GTS 450 SLI is quite fast but not as fast as to justify the $40-50 economy compared to the comparable single-chip solutions. Besides, it slows down more than its single-GPU opponents at resolutions higher than 1920x1080.
The advantages of the GeForce GTS 450 SLI are as unclear as in the previous tests: it is comparable to the GeForce GTX 470 in performance at popular resolutions but loses its ground at extremely high display modes. Thus, the only possible reason for building such a tandem is when you have one GeForce GTS 450 and add another one later on as a way to upgrade your graphics subsystem for little money.
The SLI tandem is a mere 10 fps faster than the GeForce GTX 470 at 1600x900, which can hardly matter considering their frame rates of 90-100 fps. You just can't feel the difference. At the higher resolutions the GeForce GTS 450 SLI falls behind the single-chip GeForce GTX 470 in terms of bottom speed. So, the single argument in favor of the SLI tandem is its somewhat lower price.
This game is tested without multisampling antialiasing as it worsens the textures and provokes a performance hit.
The SLI configuration has the lowest bottom speed among the tested solutions, being even inferior to the single GeForce GTS 450 in this respect. So we can see that multi-GPU technologies have not yet got rid of all their problems. But perhaps some of the problems are due to their very architecture and cannot be eliminated.
We enable DirectX 10.1 and 11 modes for graphics cards that support them.
The GeForce GTS 450 SLI is competitive to the GeForce GTX 470 but has a lower bottom speed at high resolutions, e.g. at 1920x1080. This cannot be justified by the price difference.
This game’s integrated benchmark does not report the bottom frame rate, so we use Fraps to get such data.
Like in the previous test, the GeForce GTS 450 SLI stumbles at the resolution of 1920x1080 where it cannot deliver a playable bottom speed, being inferior to the GeForce GTX 470 as well as to the less advanced GeForce GTX 460 1GB. So, using the two entry-level Fermi-based cards in SLI mode makes but little sense here.
Here is another common problem of homogeneous multi-GPU solutions: they depend too heavily on driver support. We couldn’t even launch Lost Planet 2 on our GeForce GTS 450 SLI tandem because the game would crash inevitably. We can only note that the ASUS ENGTS 450 TOP is much faster than the reference card.
We enforced full-screen antialiasing using the method described in our special Mass Effect 2 review.
It’s the first time we see our GeForce GTS 450 SLI tandem enjoy a large lead over the GeForce GTX 470, but it also falls behind the latter at 2560x1600, its bottom speed plummeting below a playable limit. Thus, the more advanced single-processor card proves to be a better choice again.
The GeForce GTS 450 SLI tandem and the single GeForce GTX 470 are roughly equal to each other in this test. The latter is somewhat more expensive but easier to use and free from compatibility issues typical of SLI and CrossFire configurations.
The game’s integrated benchmark cannot report the bottom frame rate. We use DirectX 10 and 10.1 modes here.
We’ve got good results at 1600x900 but the GeForce GTS 450 SLI is only as good as the single GeForce GTX 470 and GeForce GTX 460 1GB at 1920x1080 and higher. So, this SLI tandem doesn't make much sense here as well as in most other tests.
We use DirectX 11 mode for appropriate graphics cards.
The GeForce GTS 450 SLI has a much lower bottom speed than the GeForce GTX 470 and also loses to the latter in terms of average performance at resolutions above 1600x900.
It is in terms of bottom speed that the SLI tandem of two GeForce GTS 450s enjoys an advantage in StarCraft II. The tandem's average frame rate is quite impressive at the two lower resolutions, too. The 2560x1600 results spoil the triumph because the bottom speed is lower than the comfortable 25 fps, but we must admit that even the Radeon HD 5870 fails to reach the comfortable level in this game with FSAA turned on.
We minimize the CPU’s influence by using the Extreme profile (1920x1200, 4x FSAA and anisotropic filtering). We also publish the results of the individual tests across all resolutions.
The GeForce GTS 450 SLI is only second to the Radeon HD 5870 in terms of overall score. It scores over 8000 points, which is quite good for a $250 solution. Judging by these results, a dual-GF106 card would be quite fast, yet we already know that the real games do not agree on this point.
The first test shows that the GeForce GTS 450 SLI feels just as uncomfortable as in most of the games. It is better in the second test, being competitive to the GeForce GTX 470 through the end and to the Radeon HD 5870 at resolutions below 2560x1600.
This benchmark produces an incomprehensible result in points, so we use Fraps to get more understandable numbers. It can only run at 1280x720 and 1920x1080.
We've got nothing new here: the GeForce GTS 450 SLI is comparable to the single GeForce GTX 470 at low resolutions and slower at high resolutions. AMD's RV870-based solutions still have the top places in the Full HD mode.
The SLI tandem doesn’t show anything exceptional in this Fermi-optimized benchmark, being inferior to the GeForce GTX 470 in terms of average speed. Fortunately, it doesn't have a performance hit at 2560x1600.
The GeForce GTS 450 supports Nvidia SLI technology and shows high scalability in multi-GPU mode. The average frame rate of the SLI tandem was an average 70-75% higher than that of the single card depending on the resolution. In some tests we've observed a performance growth of 100% and more. You can see this in the following diagrams which also indicate the competitiveness of that SLI tandem against the GeForce GTX 470 and Radeon HD 5870/5850.
It may seem that the GeForce GTS 450 SLI is quite an appealing solution compared to the more advanced single-chip cards, but we must note a couple of disappointing facts about it. First, this SLI tandem had a lower bottom speed than its single-GPU opponents in some games. This made the gameplay uncomfortable in S.T.A.L.K.E.R.: Call of Pripyat, for example. Second, the GeForce GTS 450 SLI would generally slow down more than its opponents at the resolution of 2560x1600. And third, we had some incompatibility issues typical of multi-GPU solutions. We could not launch Lost Planet 2 at all using the latest version of the Nvidia GeForce driver, and there may be other such games.
All of this makes the GeForce GTS 450 SLI a far less appealing alternative to a single GeForce GTX 470 or Radeon HD 5870, even though it may cost somewhat less money. Such a configuration may be an option for a two-step upgrade when you add a second GeForce GTS 450 to your existing one in order to increase your computer’s graphics performance, yet we still think purchasing a GeForce GTX 470 would be a better investment as it is easier to use and free from compatibility issues.
We also do not expect a dual-processor GF106-based card to be released unless some major graphics card maker takes a fancy to experiment.
Talking about the specific product, the ASUS ENGTX450 DirectCU TOP has left a positive impression on us, just like its senior cousin ENGTX460 DirectCU TOP that we reviewed earlier. The single downside is that the operation algorithm of its fan does not make the card is quiet as it theoretically could be. Otherwise, it is a high-quality GeForce GTS 450 equipped with a high-performance cooler and offering a higher speed than the reference card thanks to factory overclocking. You won’t be disappointed with it if you want a GeForce GTS 450.