by Alexey Stepin
04/19/2008 | 01:41 PM
Supported by the entire graphics card range, from cheap $59 products to luxurious $800 ones, multi-GPU technologies have only been really popular when applied to top-end solutions. It is easy to explain. Many people who buy premium-class solutions do not actually care about the price. They can just as easily buy two or even three advanced graphics cards to get highest performance possible in particular games. It is this user category that ensures a stable market niche for ATI CrossFire and Nvidia SLI systems. As we learned in our recent test session, these systems can indeed provide unprecedented performance in many applications notwithstanding some drawbacks.
It’s different with mainstream graphics cards costing up to $200-250 and accounting for a large share of 3D graphics hardware sales. All of them can work in dual-processor subsystems, this feature being touted by ATI and Nvidia as a means to increase your graphics subsystem’s performance at low cost. This is true, yet our tests have showed that purchasing a single card from a higher category is often a preferable alternative as it ensures similar or higher performance without any compatibility issues typical of multi-GPU systems. Unfortunately, ATI users do not have this opportunity today: AMD’s high-end graphics solution, ATI Radeon HD 3870 X2, is already a dual-processor tandem, carrying two RV670 chips on a single PCB.
Nvidia, on the contrary, offers a number of rather advanced single-chip graphics cards, but the company doesn’t stop to promote its multi-GPU technology on the mass market. The recently released GeForce 9600 GT supports 2-way SLI, being able to work in pair with another such card. Based on the G94 core, this card belongs to the $169-189 category where it is competitive against the Radeon HD 3850 and 3870. Although endowed with only 64 ALUs, it feels all right in modern games, often being about as fast as the more expensive GeForce 8800 GT 512MB.
Considering the price and performance of the GeForce 9600 GT, we are going to see if there is a reason to build a SLI tandem out of two such cards. Will this configuration be a good alternative to Nvidia’s GeForce 8800 GTS 512MB and GeForce 9800 GTX and to the Radeon HD 3870 X2? As usual, we’ll also consider such aspects as noise, power consumption, and ease-of-use.
We’ll perform our tests using the Gigabyte GV-NX96T512H-B card which is a precise copy of Nvidia’s reference sample.
The GV-NX96T series includes three different graphics cards but all of them come to retail in the same box of ordinary size. The wrapper is painted lilac colors and looks quite appealing. Inside it there is the main box made from thick white cardboard. The graphics card is fixed securely in a foam-rubber tray. Besides it, the box contains:
We’ve got a bare minimum of accessories here without any free software. That’s normal for the price category the GeForce 9600 GT belongs with.
However, the lack of software for HD video playback is a problem because all G92- and G94-based graphics cards have a full-featured hardware HD video decoder and can thus be used not only in gaming computers but also in home multimedia systems. This is especially true for the GeForce 9600 GT with its rather low power consumption. In fact, the lack of a HD video player prevents the user from utilizing one of the most touted features of the product unless you invest additional money into it. That’s somewhat unfair on the card manufacturer’s side as well as on the GPU developer’s (who only allows using the hardware decoding and post-processing features to the developers of proprietary software players).
The installation guide is a folding poster; the user manual is a classic brochure. We don’t have any complaints about them.
Overall, the packaging of the Gigabyte GV-NX96T512H-B is pretty and secure while the accessories are up to the product class. The single downside is the lack of HD video playback software – you’ll have to buy it separately for as much as $50-100.
Nvidia’s reference design differs considerably from what Gainward used in its Bliss 9600 GT 512MB GS we reviewed earlier.
The power circuit is simpler: the GPU voltage regulator has two phases as opposed to three on the Gainward card. A dual-phase ADP3208A PWM-controller from Analog Devices is responsible for GPU power supply; we haven’t seen this chip before. Interestingly, the manufacturer’s website says that shipments of this chip were discontinued in 2008. As the reserves of such chips are exhausted, other power controllers may be installed on GeForce 9600 GT. The card receives external power through a standard 6-pin PCI Express 1.0 slot. The GeForce 9600 GT consuming about 60W, the card might do even with no external power at all, but the developer didn’t want to put all the load on the PCI Express slot.
The PCB carries eight GDDR 3 memory chips (K4J52324QE-BJ1A). The chips have a capacity of 512Mb (16Mb x 32), a voltage of 1.9V and a rated frequency of 1000 (2000) MHz. The Gainward Bliss 9600 GT 512MB GS comes with this memory, too. The card’s memory frequency is standard at 900 (1800) MHz. Accessed across a 256-bit memory bus, this ensures a bandwidth of 57.6GB/s. As we already know, the GeForce 9600 GT doesn’t suffer from insufficient memory bandwidth. The card has a total of 512 megabytes of memory, which is a standard capacity even for inexpensive mainstream solutions.
The GPU is revision A1. The card’s GPU frequencies are standard, too: 650MHz for the main domain and 1625MHz for the shader domain. The GPU configuration is standard with 64 universal scalar ALUs, 16 (32) TMUs, and 16 ROPs. The texture processors have the same architecture as in all latest solutions from Nvidia, i.e. with two filter units per each two address units. Theoretically, it means 32 TMUs, but from a practical point of view the GPU has rather 16 TMUs as there is no talking about “free” tri-linear and anisotropic filtering with this architecture. Practice suggests that the TMUs and ROPs are not a bottleneck in the G94 design, though. There is no protective frame on the GPU package, but that’s not a problem as the cooling system is small and light.
As opposed to the Gainward Bliss 9600 GT 512MB GS, the Gigabyte card has standard interfaces: two dual-link DVI ports and a universal connector for analog video output. Although the G94 incorporates an integrated DisplayPort controller, the card doesn’t allow for the installation of an appropriate connector. Besides the power PCI Express 1.0 connector, the PCB carries a standard MIO plug for SLI connection and a 2-pin S/PDIF plug for connection to the sound card’s digital output to enable the audio-over-HDMI feature.
The Gigabyte GV-NX96T512H-B comes with the reference cooler we are familiar with by the GeForce 8800 GT 256MB. It has a copper sole contacting with the graphics core and a heatsink considering of thin aluminum plates. Flat heat pipes in the cooler’s base help distribute the heat more uniformly. The cooler has a rather large fan. The heatsink being placed at an angle, the hot air is exhausted sideways, towards the side panel of the system case. The cooler is secured firmly with 12 spring-loaded screws to prevent any misalignment that might damage the GPU.
This cooler is not very efficient and its first version didn’t always cope with the job on the GeForce 8800 GT 512MB. The GeForce 9600 GT consumes less power and generates less heat, so the compact single-slot cooler is quite okay for it.
As opposed to the Gainward Bliss 9600 GT 512MB GS we reviewed earlier, the Gigabyte GV-NX96T512H-B is a copy of the reference card. So, we measured its power consumption again using the following testbed:
The card is loaded in 3D mode by the first SM3.0/HDR test from 3DMark06 launched in a loop at 1600x1200 with 4x FSAA and 16x AF. In the Peak 2D mode the card performed the 2D Transparent Windows test from PCMark05. Here are the results:
Despite the considerable discrepancies in the power circuit design, the cards from Gigabyte and Gainward have almost identical results. The biggest load is on the external power connector, yet it is not higher than 32W. The total power draw is not higher than 60-61W. That’s an excellent result for the performance the GeForce 9600 GT delivers in games.
Comparing the theoretical peak power draw of two SLI-linked GeForce 9600 GT cards, which is about 120W, they consume somewhat more than the GeForce 9800 GTX and GeForce 8800 GTS 512MB (about 100W) but less than cards like ATI Radeon HD 3870 X2 or Nvidia GeForce 9800 GX2 (consuming 170W and 180W, respectively).
We measured the level of noise produced by the card with a digital sound-level meter Velleman DVM1326 using A-curve weighing. The level of ambient noise in our lab was 36dBA and the level of noise at a distance of 1 meter from the working testbed with a passively cooled graphics card inside was 43dBA.
The reference single-slot cooler proved to be no louder than the advanced dual-slot system from Gainward. Its efficiency is lower, though. The GPU temperature is 45-46°C when idle and grows to 62°C under load. For comparison, the Gainward Bliss 9600 GT 512MB GS has a GPU temperature of 42°C and 58°C when idle and under load although the card is pre-overclocked. The difference is not big, however. And 62°C is not really high for a modern GPU – we are accustomed to much higher numbers. The G94 is indeed a very lucky chip in terms of electrical and thermal characteristics.
The GeForce 9600 GT doesn’t change the fan speed when switching from 3D to 2D mode, so two such cards are going to produce more noise than such quiet cards as GeForce 8800 GTS, 9800 GTX and Radeon HD 3870 X2.
Our overclocking attempt proved to be a success despite the compact cooler. The highest GPU frequency the card was stable at was 770/1925MHz. That’s far better than we achieved with the Gainward card. The memory could overclock to 1100 (2200) MHz. That’s quite a good result for non-extreme overclocking.
We didn’t have any compatibility problems with the Gigabyte GV-NX96T512H-B card. It started up successfully on every PCI Express 1.0a and 1.1 mainboard we tried it with and passed the tests normally.
We used a testbed configured like follows:
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, Adaptive Anti-Aliasing/Multi-sampling in ATI Catalyst and Antialiasing – Transparency: Multisampling in Nvidia ForceWare. As a result, our ATI and Nvidia driver settings were as follows
ATI Catalyst settings:
For our tests we used the following games and benchmarks:
First-Person 3D Shooters:
- Battlefield 2142
- Call of Juarez
- Call of Duty 4
- Enemy Territory: Quake Wars
- Half-Life 2: Episode Two
- S.T.A.L.K.E.R.: Shadow of Chernobyl
Third-Person 3D Shooters:
- Lost Planet: Extreme Condition
- Tomb Raider: Legend
- Hellgate: London
- The Elder Scrolls IV: Oblivion
- Company of Heroes: Opposing Fronts
- Command & Conquer 3: Tiberium Wars
- World in Conflict
- Futuremark 3DMark05
- Futuremark 3DMark06
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. 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 DirectX 10 mode.
The tests were performed at resolutions of 1280x1024/960, 1600x1200 and 1920x1200 pixels. 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.
For the sake of more illustrative analysis we included the following graphics cards to participate in this test session:
We took an ASUS EN9600GT TOP for a SLI partner to the Gigabyte GV-NX96T512H-B. This card is based on the reference PCB design as well. Being a pre-overclocked product, we lowered its clock rates to the standard values.
This game does not support display resolutions of 16:10 format, so we use a resolution of 1920x1440 pixels (4:3 format) instead of 1920x1200 for it.
Every graphics card tested can ensure comfortable gaming conditions here. You don’t have to buy a second card to have more speed in Battlefield 2142. However, the SLI configuration with two GeForce 9600 GT cards delivers the highest performance: it is ahead of the GeForce 9800 GX2 at 1920x1440 and rivals the 3-way and 4-way CrossFireX configurations based on Radeon HD 3870.
BioShock does not support FSAA when running in Windows Vista’s DirectX 10 environment. That’s why we benchmarked the cards without FSAA.
As we learned in our GeForce 9800 GX2 review, the game is compatible with dual-chip configurations but we can’t achieve a noticeable performance growth using a pair of GeForce 9600 GT.
That is a bad news for affordable SLI even it’s just a single failure. Purchasing a GeForce 8800 GTS 512MB or 9800 GX2 will free from you potential problems with software support for SLI technology.
If you don’t have the necessary sum of money, you can still play the game on one GeForce 9600 GT as the frame rate is high enough even at 1920x1200.
The GeForce 9600 GT SLI configuration provides a nearly-comfortable level of performance at 1280x1024. On the other hand, the ATI Radeon HD 3870 X2 has an even higher result while occupying only one PCI Express x16 slot and not requiring a CrossFire or SLI compatible mainboard.
The GeForce 9600 GT SLI is slower than the GeForce 8800 GTS or GeForce 9800 GTX at higher resolutions. The SLI tandem doesn’t look like an appealing buy.
In this case, the addition of a second GeForce 9600 GT into the system leads to an impressive outcome: at 1920x1200 this subsystem easily beats the GeForce 9800 GTX, let alone the GeForce 8800 GTS 512MB. However, the latter card makes the game playable at 1920x1200 too, but is free from the drawbacks typical of multi-GPU solutions. It’s up to you to decide if the difference in performance is really worth it.
This game is tested at the High level of detail, excepting the Shader option which is set at Very High. This way we try to achieve a compromise between image quality and speed.
The GeForce 9600 GT SLI subsystem is unrivalled here, but its average speed is hardly above 30fps, which is far from comfortable. Alas, you can’t avoid slowdowns in especially heavy scenes when playing Crysis at high graphics quality settings.
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 performance growth is amazing in Quake Wars, being over 100% in some modes. On the other hand, this doesn’t affect your playing comfort in comparison with the GeForce 8800 GTS 512MB or the 9800 GTX due to the frame rate limiter.
We can see performance benefits from having two GeForce 9600 starting from 1600x1200. This system is 14-15% ahead of the Radeon HD 3870 X2 and GeForce 9800 GTX at 1920x1200, reaching the desired speed of 60fps. So, SLI technology is a good means to increase performance in this game if you’ve got one GeForce 9600 GT and an appropriate mainboard. Other HL2 fans should think about buying a more advanced single card such as GeForce 9800 GTX.
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.
The GeForce 9600 GT SLI configuration is only inferior to the recently released dual-processor monster GeForce 9800 GX2. That’s a huge potential indeed. We can note, however, that the single high-end cards, including the dual-chip Radeon HD 3870 X2, feel more confident at 1920x1200 ensuring a higher minimum of speed.
The GeForce 9600 GT SLI has no alternatives in Lost Planet if you don’t count in the expensive dual-chip GeForce 9800 GX.
There is a tip for lowering this game’s system requirements without losing much in terms of visual quality – just switch some settings from High to Medium. The difference can hardly be spotted with a naked eye.
The dual-card SLI configuration that would cost you about $318-398 is faster than the latest single-chip flagship card from Nvidia by 11-24% depending on resolution. The gaming experience is going to be the same in any case, but the GeForce 9800 GTX and the GeForce 8800 GTS 512MB are surely easier to handle.
Hellgate Longon is a yet another in which the GeForce 9600 GT SLI meets no real opponent that would allow playing at 1920x1200 with 4x multisampling. The ATI Radeon HD 3870 X2 might make one, but its driver is not optimized well for this game.
The game loses much of its visual appeal without HDR (and without FSAA due to technical reasons). Although some gamers may argue that point, we believe that TES IV the best with enabled FP HDR and test it in this mode.
The GeForce 9600 GT SLI platform is competitive against the Radeon HD 3870 X2 in closed environments but has a very low minimum of speed in open scenes. Its frame rate can occasionally plummet to below 25fps even at 1600x1200. When working as a single card, the GeForce 9600 GT keeps the frame rate above 26fps. Anyway, that’s yet another test where the two GPUs are obviously better than one.
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 GeForce 9800 GTX is the only card to match the performance of the GeForce 9600 GT SLI, but the ATI Radeon HD 3870 X2 remains unrivalled when it comes to minimum speed. Excepting the latter, none of the cards can provide a minimum speed of higher than 15-16fps, which means inaccurate control over the game process in action-heavy scenes.
The game has a frame rate limiter, which is why one should consider the minimum speed of a graphics card on the first place.
Although the GeForce 9600 GT SLI successfully hits the frame rate limit in every display mode, its minimum speed is somewhat lower than with the other tested solutions.
Every graphics card is slow in this game. At a resolution of 1280x1024 the single GeForce 9800 GTX and GeForce 8800 GTS 512MB have a certain advantage as they deliver a somewhat higher minimum speed. On the other hand, the GeForce 9600 GT SLI can be called a leader in terms of average frame rate.
The GeForce 9600 GT SLI subsystem is about as fast as the GeForce 9800 GTX and GeForce 8800 GTS 512MB. All of them are considerably slower than the ATI Radeon HD 3870 X2.
The SLI configuration has good results in all the three tests, but it is only a definite leader in the third test, beating the Radeon HD 3870 X2 even.
The 3DMark06 results are more interesting: the system with two GeForce 9600 GT cards working in SLI mode scores over 12,000 points. This is over 1000 less than the total score of the Radeon HD 3870 X2 that features far more advanced computing capabilities.
The described SLI configuration isn’t impressive in the SM2.0 tests being somewhat faster than the GeForce 8800 GTS 512MB and almost as fast as the GeForce 9800 GTX.
The GeForce 9600 GT SLI is far more confident in the SM3.0/HDR tests, outperforming every single-chip card and being only inferior to the dual-chip Radeon HD 3870 X2 with its 640 shader processors.
It’s all very different when we enable 4x FSAA. The GeForce 9600 GT SLI doesn’t show anything extraordinary in the total result for the SM2.0 tests, it is now ahead of every single-chip solution from Nvidia, including the GeForce 9800 GTX. The Radeon HD 3870 X2 is left behind in the first test, too. ATI’s solution is dragged down by its texture processor architecture with two filter unit per each two address units but it wins the second test due to its impressive computing capabilities.
The ATI Radeon HD 3870 X2 is in the lead in both SM3.0/HDR tests but the GeForce 9600 GT SLI tandem is second with a large lead over the single-chip solutions. The alternate frame rendering proves to be better than the double amount of ALUs and TMUs in the GeForce 9800 GTX and 8800 GTS 512MB.
The GeForce 9600 GT SLI configuration has done very well in our tests, showing high performance and proving that the G94 chip has optimal architecture. Delivering highest performance across a number of applications, this pair of mainstream graphics cards shows a strong muscle, but we need to examine the picture in more detail to recommend or not to recommend this graphics subsystem to you.
The GeForce 9600 GT SLI had higher performance than one such card in nearly all of our tests.The exceptions were BioShock where the GeForce 9800 GX2 and ATI Radeon HD 3870 X2 worked well, but the GeForce 9600 GT SLI had no performance growth at all and Call of Juarez where the SLI subsystem was hamstringed by inefficient graphics memory management, the old-time problem of Nvidia’s solutions.
But in most of the tests the GeForce 9600 GT SLI subsystem had the same or higher performance as such cards as ATI Radeon HD 3870 X2, Nvidia GeForce 9800 GTX and GeForce 8800 GTS 512MB.
Summing it up, the GeForce 9600 GT SLI is currently a high-quality and fast graphics subsystem (at least for the games we tested it in). If Nvidia goes on optimizing its drivers for 2-way SLI, this configuration may be future-proof as well.
There are a number of points that can lower the appeal of this graphics subsystem:
Thus, affordable SLI proves to have a limited range of applications. It may only be interesting for people who have a rather expensive SLI-supporting mainboard but also a mainstream graphics card. For them, adding a second GeForce 9600 GT into the system and enabling SLI mode is going to be a nice means of increasing 3D performance. But the old method – replacing your existing GeForce 9600 GT with a single top-class graphics card – will still work better for other users.
Talking about the Gigabyte GV-NX96T512H-B, it is an ordinary GeForce 9600 GT, with Nvidia’s reference PCB design and cooler. It is a good choice for every gamer with a limited budget. A well-balanced solution, the card shows superb results (for its class) at resolutions of 1280x1024 and 1600x1200/1680x1050. It has the ATI Radeon HD 3870 as its market opponent.
Gigabyte’s product doesn’t have specific features such as direct support of DisplayPort or S/PDIF TOSLINK, but it supports HDMI including audio-over-HDMI. The single-slot cooler makes the Gigabyte card suitable for multimedia system, yet it comes without a DVI-I → HDMI adapter and HD content player. You’ll have to buy them separately, which is not good. With its accessories, the GV-NX96T512H-B is targeted at gamers rather than at video fans. By the way, the GV-NX96T series includes three models: the described 512H-B, the 512H with a Zalman cooler, and the 512HP with a passive cooling system.