by Alexey Stepin , Yaroslav Lyssenko, Anton Shilov
10/10/2006 | 07:02 PM
The first attempt to build a four-processor consumer-class graphics subsystem undertaken by Nvidia was unsuccessful, to put it mildly. We used a ready-made platform from a system integrator who had been granted permission to sell quad-SLI complexes, yet the tested system was far from stable (for details see our article called Quadtet: Nvidia GeForce 7900 Quad SLI Performance Unveiled). Besides the stability issues, its performance wasn’t as high as we had expected whereas the image quality was horrible in some games due to various visual artifacts. That implementation of the quad SLI concept proved to be almost unusable.
Still, the concept itself definitely has a potential and may achieve what is impossible on dual-processor tandems, let alone single-chip graphics cards. And of course Nvidia didn’t stop to promote the new technology, especially as they had promised that gaming quad-SLI systems would eventually be available for integration by users themselves, not only as ready-made solutions. And now this time has come.
Theoretically, it became possible to build a four-processor graphics subsystem in your home computer about two months ago. It’s when the clumsy and very unreliable GeForce 7900 GX2 graphics card that had been used in the first-generation quad SLI was transformed into a rather compact and simple GeForce 7950 GX2 (for details see our article called Two for One: Nvidia's Dual-Chip GeForce 7950 GX2 Reviewed). Even with certain drawbacks, that card became a sensation among single-PCB solutions, delivering unrivalled performance and image quality. The latter was achieved by offering full-screen antialiasing modes that had previously been unavailable on classic single-chip cards.
However, the option of uniting two GeForce 7950 GX2 cards into a quad SLI subsystem was blocked by Nvidia on the driver level. The company thought it was too difficult for the end user to assemble such a subsystem on his own. There was some logic in this thinking. Until the arrival of the GeForce 7950 GX2 Nvidia could argue that:
But the first point can now be criticized because one GeForce 7950 GX2 consumes about 110-120W which is comparable to the power draw of a single Radeon X1950 XTX. So, two such graphics cards will require about as much power as a Radeon X1950 XTX CrossFire subsystem and will not need a special power supply. Any high-quality 600-650W power supply will do. Still, Nvidia insists on higher numbers. The power supplies it has certified so far all have a wattage of 700W and higher (you can read the full list on the appropriate page of the SLIZone website).
The second point isn’t as urgent as it used to be, too. The GeForce 7950 GX2 is much smaller than the GeForce 7900 GX2 and can be easily installed into almost any standard ATX system case. Of course, the case must be ventilated well, but this rule applies not only to quad-SLI subsystems based on two GeForce 7950 GX2, but also to any high-performance SLI or CrossFire subsystem. The ventilation issues aren’t any news for an experienced home user. What may become a problem when a quad-SLI platform is assembled at home is that the GeForce 7950 GX2 may turn to be incompatible with the mainboard.
Thus, the only argument in favor of purchasing a ready-made quad-SLI system from an authorized system integrator is that you are given a warranty and provided the manufacturer’s tech support. However, this won’t help avoid performance or compatibility issues that must be solved by Nvidia as the developer of SLI technology and writer of the appropriate drivers. It seems that quad SLI wasn’t that difficult to implement on the hardware level, but had a lot of problems with the software as was shown in our preview.
On July 20, 2006, Nvidia finally removed the barrier and released its ForceWare 91.37 driver that would allow using two GeForce 7950 GX2 cards in a quad-SLI subsystem. That version of ForceWare was to come with ready-made quad-SLI systems but for some reason had a beta status and was not officially supported by Nvidia. So, it’s only on August 13 that the limitation on building users’ own quad-SLI subsystems was removed – the company released ForceWare 91.45 which had a lot of corrections to solve the known problems of quad-SLI. You can read the list of changes by downloading the release notes. Right now, ForceWare 91.47 is the latest version of the quad-SLI-supporting driver from Nvidia, and we used it for this review.
We’ve recently obtained a second sample of the GeForce 7950 GX2, so we can now check out what performance and compatibility improvements have been brought by the new driver for quad SLI. Will we indeed be lifted up to an unrivalled level of performance in games? Let’s first recall the basics of Nvidia’s quad SLI technology, though.
Let’s make out the differences first. The flowchart of a GeForce 7950 GX2-based quad-SLI complex is somewhat different from the flowchart of a quad-SLI system based on GeForce 7900 GX2. The discrepancy lurks in the switching circuitry of the MIO interfaces that are used to transfer data necessary for multi-GPU mode. The GeForce 7900 GPU contains two independent MIO interfaces, each of which has a bandwidth of about 1GB/s. Each GeForce 7900 GX2 and 7950 GX2 card has two such GPUs that use one interface to communicate between each other within the same card. On the GeForce 7900 GX2 the second pair of interfaces is wired to the appropriate connectors and is used to connect to a second card in a quad-SLI complex via a standard SLI bridge. Logically and physically it looks like this:
It’s different with the GeForce 7950 GX2 that has only one external MIO connector. Nvidia must have decided that the second connection had a very small effect on performance and abandoned it to simplify the design of the card and, accordingly, the design of a quad-SLI subsystem made out of two GeForce 7950 GX2 cards.
There is only one external MIO connection, between the GPUs on the PCB with the PCI Express x16 connector and switch. In theory, this may have a negative effect on the synchronization of the four GPUs and reduce the efficiency of quad SLI technology, but we can only check this out in practical tests.
We described the rendering modes available on a quad-SLI system in detail in our preview of this technology, yet we want to remind you about them, in brief.
So, the first method, Alternate Frame Rendering, means that the driver gives out four frames to the GPUs and each GPU processes its own frame. It looks like this:
So, this method is a variation of the ordinary dual-processor AFR and works in the same way. The difference is in the number of GPUs working together in the same graphics subsystem. Nvidia claims this method ensures the highest efficiency because it involves no balancing overhead. Each GPU is given a fair share of total work. With two GPUs, the performance growth is nearly two-fold and may amount to 90%. A quad-GPU system is likely to have worse scalability, yet it should anyway provide a considerable performance boost.
As we wrote in our earlier reviews, the AFR method has certain peculiarities when it comes to render-to-texture operations (such rendering techniques as environmental mapping, shadow mapping and others belong here). In this case the information about all the changes in the render targets must be sent from one GPU to another. This greatly increases the load on the MIO interface and, accordingly, the GPU synchronization overhead. In order to avoid sending large amounts of data through the MIO interface, Nvidia recommends the developer to perform a Clear() command to clear the color values in the render targets for each frame. But sometimes this is not possible, particularly when the results of the rendering of the previous frame are necessary to render the current frame. So, Nvidia recommends to create two render targets, one target for render-to-texture operations in even-numbered frames and the other in odd-numbered frames. Well, we don’t know how many programmers use this technique to optimize their code for dual-processor graphics subsystems, not to mention quad-processor configurations. Moreover, it is reported that quad-GPU solutions cannot yet use this rendering mode for Direct3D applications.
The Split Frame Rendering method splits one frame into 2 or 4 parts depending on the number of GPUs in the system. The total load is shared dynamically among the GPUs. If the driver notices that one GPU is loaded more than another, it can change the size of the frame parts to restore the balance.
This method is less efficient than AFR due to a bigger synchronization overhead and doubling of certain rendering operations.
The AFR on SFR method is a combination of the described modes. It works like this: the first frame is rendered by the first and second GPUs in SFR mode while the second frame is rendered by the third and fourth GPUs. The results of the rendering are output on the screen one by one, as in AFR mode:
This method is compatible with all applications that correctly support Alternate Frame Rendering and seems to be the optimal mode for a quad-SLI configuration.
SLI Antialiasing (SLI AA) mode is meant to improve the antialiasing quality. It combines the antialiasing performed on the separate GPUs into one final frame.
A quad-SLI system allows to use a unique 32xs SLI AA mode when each GPU performs 8xs antialiasing with a shift relative to the results of the other GPUs. After that, the results are combined into the final frame.
This provides the best antialiasing quality available in the 3D industry, but at the expense of a huge performance hit even the four GPUs sometimes cannot make up for. A quad-SLI system uses SLI AA patterns different from what are used on an ordinary dual-GPU SLI complex.
In the compatibility mode only one GPU out of the quad-GPU configuration is active. This mode doesn’t give you any performance gains.
The second GeForce 7950 GX2 we are going to use in this review is Gigabyte’s GV-3D1-7950-RH card, so let’s have a closer look at it. Note that the symbols 3D1 denote a dual-processor design by analogy with GV-3D1-68GT.
The package of the card should be familiar to you if you’ve read our previous reviews of Gigabyte’s products. It is oriented vertically:
The box is decorated alike to the package of the Gigabyte GV-NX79X512DB except for the colors. Red is traditionally associated with ATI Technologies, which might be confusing if it were not for the large caption “GeForce 7950 GX2”. Among the logotypes denoting various capabilities of the GV-3D1-7950-RH there is a white rectangle with the words “Windows Vista Ready”. The GeForce 7950 GX2 is based on ordinary G71 chips, so this label probably means that the card comes with appropriate WGF-supporting drivers.
Besides the GV-3D1-7950-RH card itself, there are the following things in the box:
This is a surprisingly scanty set of accessories for a premium-class solution. We hadn’t expected Gigabyte to be as gorgeous as ASUS is with some of its products, yet we did hope to find more and newer games or some useful trifle like a gamepad.
Well, not all users are actually interested in the accessories they can get with a graphics card. Most are quite satisfied with the standard set including everything necessary just to use the purchased product. This standard set is indeed enclosed with the Gigabyte GV-GV-3D1-7950-RH and the user manual is written well, telling you all the info you may need, particularly how to enable SLI mode and update the graphics card’s BIOS. The manual also provides a list of compatible mainboards. A curious fact: the illustration in the manual’s SLI-related section shows a pair of GV-3D1-7950-RH cards connected with a standard MIO bridge, which is in fact a quad-SLI configuration. This is quite a clear indication that it’s now possible to build quad-SLI systems with your own hands!
Nvidia permits no experiments with PCBs of its premium-class graphics cards. That’s why the Gigabyte GV-3D1-7950-RH is nothing else but an ordinary GeForce 7950 GX2 manufactured by Foxconn for Nvidia and then shipped to Gigabyte.
Gigabyte’s stickers on the cooler’s casing and fan are the only things that differentiate this product from a reference GeForce 7950 GX2. Otherwise, this is the same two-storied design we described in our earlier review called Two for One: Nvidia's Dual-Chip GeForce 7950 GX2 Reviewed.
The graphics card consists of two PCBs combined into a single whole with four hex-headed poles. The PCBs each carries one Nvidia G71 chip, but they differ in other components. The bottom PCB is equipped with PCI Express x16 and MIO connectors and also carries a PCI Express x48 switch that distributes the available PCI Express lanes among the GPUs and makes the GeForce 7950 GX2 compatible with mainboards that do not officially support SLI technology. But in spite of that switch, there are only 64 mainboards on the compatibility list as yet. This is not much considering the total number of available mainboards with a PCI Express x16 slot.
The top PCB is simpler: besides the standard kit of one GPU and eight memory chips, it has a power circuit and a standard communication suite (two DVI-I connectors and one universal TV-Out/HDTV connector). It’s on the top PCB that the single (as opposed to the GeForce 7900 GX2) additional power connector is located. The PCBs are united into a single whole with a special two-section connector: one section transfers power and MIO signals and another, the longer one, transfers PCI Express x16 signals.
Each PCB carries eight 512Mb chips of GDDR2 (Samsung K4J52324QC-BC14), which give you a total of 1GB, but applications can only make use of 512MB, as on ordinary single-GPU top-end graphics cards. The memory and GPU are clocked at 600 (1200) MHz and 500MHz, respectively, in full compliance with the official GeForce 7950 GX2 specification. There have already appeared pre-overclocked versions of the GeForce 7950 GX2. Particularly, the XFX GeForce 7950 GX2 XXX Edition is clocked at 570MHz GPU and 775 (1550) MHz memory.
The Gigabyte GV-3D1-7950-RH is cooled with a couple of standard coolers the GeForce 7950 GX2 usually comes with. Using a copper core that contacts with the GPU die and a heat pipe built into the aluminum base for a uniform distribution of heat, the cooler ensures high cooling efficiency in spite of its small size and the small diameter of the fan. The G71 chip clocked at a rather low frequency of 500MHz doesn’t generate too much heat, and this helps the cooler, too.
There’s a layer of dark-gray thermal grease between the GPU and the cooler. The memory chips contact the cooler through elastic pads that consist of two halves with a layer of white thermal grease in between. The tight contact ensures good heat transfer.
Two 2.16W blowers, about 45mm in diameter, cool the heatsinks. They have a green LED highlighting, but it is not very bright and will hardly be visible in a system case with a side window.
The main problem with the standard cooling system of the GeForce 7950 GX2 is that the bottom cooler nearly touches the top PCB and obviously has problems getting enough air. The bottom GPU may overheat as a consequence, especially in a poorly ventilated system case. Installing a second GeForce 7950 GX2 to enable quad-SLI mode may aggravate the problem, so if you are going to build such a subsystem, you should purchase a large system case that permits to install additional fans.
Unfortunately, Gigabyte’s stickers don’t hold fast on the fans’ rotors. During our tests the sticker on the bottom fan of our card peeled off and stuck in the fan’s blades. With the two-storied card design, we couldn’t see the bottom cooler and didn’t spot the problem right away. The overheat protection system didn’t work for some reason, and the temperature of the bottom GPU grew up to a critical value. The system hung up, but the card survived, fortunately. We guess the PCB could have been as hot as 100-120°C near the GPU.
This is an example of what troubles a failure of the bottom cooler of your GeForce 7950 GX2 may provoke. We advise you to remove any stickers from the fans so that the described situation never occurred with your card. You should also check out and clean the card’s fans regularly to avoid problems and make the card’s life longer.
CrossFire technology at its fastest still requires a Master card, so a Radeon X1950 CrossFire Edition has been released. From a technical point of view, this Master is no different from the Radeon X1800 XT CrossFire Edition or the Radeon X1900 XT CrossFire Edition.
The parts of the frame are still being combined on a programmable Xilinx XC3S400 array from the Spartan-3 family. It is coupled with two Silicon Image SiI163В TMDS transmitters that can output in dual-link DVI mode at a max resolution of 2560x1600@60Hz. The 3-channel 10-bit ADV7123 DAC working at a frequency of 330MHz allows to connect monitors with an analog interface and use resolutions up to 2048x1536@75Hz. Devices with composite (RCA) or S-Video connectors can’t be attached to the card because the compositing engine and the DAC are located behind the TV encoder integrated into the GPU. To support TV output, the Radeon X1950 CrossFire Edition would require an additional chip to encode TV signal. This would make the Compositing Engine even more complex.
Other than the Compositing Engine, the design of the Radeon X1950 CrossFire Edition is analogous to the Radeon X1950 XTX. Their main parameters like GPU and memory frequencies coincide, too. The card carries eight GDDR4 chips from Samsung (K4U52324QE-BC09) clocked at 1000 (2000) MHz. The GPU works at 650MHz. So, there can’t be a problem of frequency misbalance between the Master and Slave cards that make up a CrossFire tandem.
The Master and Slave are linked with a Y-shaped cable with two DVI connectors and one dual-channel DMS connector. The male-type DVI is attached to the Slave card, and the DMS plugs into the appropriate connector on the Master. The remaining female-type DVI is for connecting the monitor. The CrossFire concept can be illustrated like follows:
As before, the CrossFire system can work in four different modes: Scissor, SuperTiling, Alternate Frame Rendering and Super AA. The Scissor and AFR modes are alike to the SFR and AFR modes in Nvidia’s SLI: the former splits the frame in two parts, the relative size of which is dynamically adjusted by the driver depending on the load on the GPUs, and the latter alternately outputs frames each of which is wholly rendered by one of the GPUs.
The AFR mode ensures better performance scalability in comparison with the Scissor mode because it doesn’t involve any load balancing overhead – each chip processes a whole frame. But it has the same problems as Nvidia’s SLI with render-to-texture operations. This mode has a specific feature: it works only if there is an appropriate game profile in the Catalyst driver. It can be forced for games that do not have such support (set the Catalyst A.I. option at Advanced).
CrossFire’s SuperTiling mode has no analog in Nvidia’s SLI technology. In this mode the frame is tessellated into 32x32 tiles. Each tile is processed by one GPU. This mode works only with Direct3D applications.
The Super AA mode serves the same purpose as SLI AA. That is, it improves the image quality by using advanced full-screen antialiasing techniques. The user is offered 8x, 10x, 12x and 14x Super AA. The number of samples is doubled in 8x and 12x modes: each card in the CrossFire tandem performs 4x or 6x FSAA, respectively, with a shift of the samples. Then the results are combined into the final frame. The 10x and 14x levels of FSAA bring in additional antialiasing, equivalent to 2x super-sampling, to improve the overall antialiasing quality even more. The double number of texture samples in Super AA mode gives you anisotropic filtering equivalent to 32x. Combined with the High-Quality AF option, this ensures an unsurpassed image quality.
We compared the Nvidia GeForce 7950 GX2 quad-SLI and the ATI Radeon X1950 XTX CrossFire subsystems on platforms configured like follows:
The drivers were set up in such a way as to provide a comparable, yet highest possible texture filtering quality.
We selected the highest possible graphics quality level in each game. We didn’t edit the games’ configuration files. The speed was measured using the game’s integrated tools or, if not available, by means of the Fraps program. We also measured the min speed of the cards where possible.
Besides the two standard resolutions of 1280x1024 and 1600x1200 pixels, we also used a resolution of 1920x1200 pixels with an aspect ratio of 16:10. We enabled FSAA and anisotropic filtering from the game’s menu. If this was not possible, we forced them using the appropriate driver settings. Besides the standard full-screen antialiasing modes, we used 8x/14x Super AA and 8x/16x SLI AA, enabling them from the drivers. This provides a most comprehensive picture of the performance and capabilities of today’s fastest multi-GPU graphics subsystem from ATI and Nvidia available for ordinary users.
We did not enable FSAA for those games that do not support FSAA due to the specifics of their engine or make use of HDR (FP16) because the GeForce 7 family cannot perform FSAA along with floating-point HDR. The following games and applications were used in this test session:
First-person 3D shooters:
Third-person 3D shooters:
This game not supporting extreme full-screen antialiasing modes, we had to limit our tests to the regular 4x FSAA plus 16x AF mode.
The first of our benchmarks shows that quad-SLI technology still needs improvements: it is slower than the single GeForce 7950 GX2. We can’t pinpoint the reason for that, but it is most likely the driver’s fault.
Meanwhile, the Radeon X1950 XTX CrossFire platform works without problems, providing a 30% performance boost over the single Radeon X1950 XTX card.
It’s the same as in the previous test: the efficiency of the CrossFire system is approaching the theoretical maximum whereas the GeForce 7950 quad-SLI, on the contrary, has a zero speed gain.
We should acknowledge that the CrossFire tandem isn’t far faster than the single GeForce 7950 GX2. One might wonder where is the promised might of the Radeon X1950 XTX CrossFire for $900 and what is the worth of the GeForce 7950 quad-SLI that costs over $1100?
The quad-SLI platform shows itself well in 8x SLI AA mode since it has four GPUs against the Radeon X1950 XTX CrossFire’s two. The speed is not high enough in resolutions above 1280x1024, but you can use 8x SLI AA.
The CrossFire platform regains the first place in the extreme mode because 16x SLI AA mode on the quad-SLI system seems to be very resource-consuming. ATI’s solution can yield only 45fps on average in its Super AA 14x mode, though. This is below the comfortable level for a first-person shooter.
The GeForce 7950 quad-SLI platform provides a performance gain in the resolution of 1920x1200 pixels, but it cannot reach the 100fps mark as the Radeon X1950 XTX CrossFire does.
As opposed to Call of Duty, the Quad-SLI doesn’t win the SLI AA 8x/Super AA 8x mode in Far Cry. It is even a little slower than the Radeon X1950 XTX CrossFire here.
Although Far Cry is not a hard trial for modern graphics cards, especially working in multi-GPU mode, it is arguable if you can enable SLI AA 16x mode on the quad-SLI system in 1920x1200. The reserve of speed is on the brink of comfortable even in 1600x1200. The Radeon X1950 XTX CrossFire is free from such problems, ensuring a frame rate of 80fps even in 1920x1200.
The results on the Research map can be commented in the same way as the results on the Pier level, although the performance is generally higher here.
The implementation of HDR in the current version of Far Cry works best on the GeForce 7 architecture, yet it is the ATI Radeon X1950 XTX CrossFire tandem that wins the race of the top-end multi-GPU platforms. The quad-SLI system isn’t any better than the single GeForce 7950 GX2.
The GeForce 7950 quad-SLI seems to wake up on the Research map. Beginning with 1600x1200 resolution, it delivers a considerable performance gain, from 10% to 20%, over the single card. This helps it compete with the Radeon X1950 XTX CrossFire in this test.
It’s for the first time in this test session that we see the quad-SLI system getting on top. At last we’ve got some advantage from having four GPUs in one graphics subsystem. The quad-SLI enjoys a 30% advantage over the single card and a 20% advantage over the Radeon X1950 XTX CrossFire. The latter yields a comfortable frame rate, too, and also provides a higher image quality with its advanced anisotropic filtering algorithm.
The quad-SLI system performs successfully in 8x FSAA mode, too. It gives you over 60fps in 1600x1200 which is quite a comfortable speed.
The multi-GPU systems are both not fast enough for you to play the game comfortably at the hardest FSAA settings. You can note, however, that the GeForce 7950 quad-SLI system is ahead in 1280x1024 resolution, but the Radeon X1950 XTX CrossFire gains the lead in the higher display modes thanks to its faster memory as well as to its less resource-consuming 14x Super AA.
The deferred rendering technique makes it impossible to enable FSAA in this game, so we can only give you the numbers for the anisotropic filtering mode.
The GeForce 7950 quad-SLI is again as fast as the single GeForce 7950 GX2 whereas the Radeon X1950 XTX CrossFire feels at ease even in 1920x1200 resolution. ATI’s solution is nearly 80% efficient, a very high percent, especially considering the zero performance growth you have with the GeForce 7950 quad-SLI here.
The GeForce 7950 quad-SLI doesn’t bring any performance improvements in Half-Life 2: Episode One, either. Yes, the performance of a single GeForce 7950 GX2 is enough here to play in 1920x1200 with enabled 4x FSAA, but you don’t need a second card for that! Moreover, the GeForce 7950 quad-SLI is about 20% slower than the Radeon X1950 XTX CrossFire in 1920x1200.
The quad-SLI platform does better than the Radeon X1950 XTX CrossFire in SLI AA 8x/Super AA 8x mode, yet this is not crucial because both the platforms provide a comfortable frame rate in every resolution, including 1920x1200.
The ATI platform still makes all the resolutions playable, but Nvidia’s GeForce 7950 GX2 quad-SLI system cannot deliver a comfortable frame rate in 1920x1200 with enabled 16x SLI AA.
Prey is one of the few games we’ve checked out so far where the GeForce 7950 quad-SLI can show its worth. It delivers a performance boost of 70% in 1920x1200! That’s an achievement for a quad-GPU graphics subsystem whose scalability is a priori worse than that of a dual-GPU one. However, the Radeon X1950 XTX CrossFire is slower than the leader by only 15-17% in that resolution and almost rivals Nvidia’s subsystem in the lower resolutions while being cheaper and providing a better-looking image.
The GeForce 7950 GX quad-SLI looks better than the competitor in SLI AA 8x/Super AA 8x mode because it provides a comfortable frame rate in 1600x1200 and, with accidental slowdowns, in 1920x1200. ATI’s tandem delivers an acceptable speed in 1280x1024 only.
The rivals exchange places, but their speeds are too low here for you to play the game normally.
It’s only in the resolution of 1920x1200 pixels that the GeForce 7950 quad-SLI enjoys a considerable lead over its opponents. In the lower resolutions the speed gain is a mere 3-5fps. The Radeon X1950 XTX CrossFire is but very little behind the quad-SLI platform but gives you a better anisotropic filtering.
The multi-GPU systems both allow to play in every resolution, including 1920x1200, in SLI AA8x/Super AA 8x mode, but the GeForce 7950 quad-SLI boasts a bigger reserve of speed. It is everywhere ahead of its competitor by 10-20%.
The GeForce 7950 quad-SLI platform cannot satisfy a demanding user in SLI AA 16X/Super AA 14x mode because its average speed is below 60fps even in 1280x1024. The Radeon X1950 XTX CrossFire offers a higher frame rate even in 1600x1200.
Nvidia’s solutions aren’t superior in Serious Sam 2 anymore. The new Radeon X1950 XTX CrossFire tandem wins every resolution and its average speed is high enough for practical use even in 1920x1200. The GeForce 7950 quad-SLI system does worse here, being limited to 1600x1200 resolution. You can note that the efficiency of Nvidia’s quad-SLI solution is low, like in a majority of previous tests.
The Radeon X1950 XTX CrossFire allows playing this game comfortably in 1280x1024 with enabled 8x Super AA. In other cases the performance is too low for you to be able to use the 8x SLI AA or 8x Super AA settings in practice.
Neither ATI’s CrossFire nor Nvidia’s quad-SLI can offer a performance gain over the corresponding single graphics cards. Moreover, the quad-SLI system has a lower minimum of speed in comparison with the single card. The game obviously has some compatibility problems with multi-GPU graphics hardware.
Super AA works correctly on the CrossFire tandem here. There is no performance hit in comparison with 4x FSAA mode as we switch to the higher antialiasing quality. We even get a minor performance increase (which may be due to the “manual” way of testing, though). It means that each chip in the tandem does its job, and the hardware frame compositing unit helps avoid performance drops. Unfortunately, this is not the case with the GeForce 7950 quad-SLI system.
The Radeon X1950 XTX CrossFire slows down in 14x Super AA mode. You can play the game in 1280x1024 with some comfort, but the higher resolutions are almost unplayable. The results of the GeForce 7950 quad SLI are very low. It gives you less than 30fps on average and even less than 20fps in high resolutions. Nvidia has still got a lot of things to improve and correct in the current implementation of its quad SLI technology.
Our attempt to run this game on the quad-SLI platform leads to a considerable performance hit whereas the efficiency of the CrossFire system is near to the theoretical maximum.
The quad-SLI platform is in the lead, but the gap shrinks almost to zero in the resolution of 1600x1200 pixels. Both these multi-GPU platforms, Radeon X1950 XTX CrossFire and GeForce 7960 GX2 quad SLI, ensure a very comfortable speed in SLI AA 8x/Super AA 8x mode.
The quad-SLI platform looks good in this test. It is a rare case when 16x SLI AA mode can be used in practice, even though in 1280x1024 resolution only. The average performance of Nvidia’s solution is already below 40fps in 1600x1200. Meanwhile, the Radeon X1950 XTX CrossFire isn’t much slower than in 8x Super AA mode and easily gives out a frame rate of over 60fps in every resolution.
Tomb Raider: Legend proved to be incompatible with ATI CrossFire technology: when you try to enable multi-GPU mode, the performance plummets heavily. The quad-SLI system is free from that problem, yet is not very efficient, either. It ensures a performance growth of 15-20%. Owners of a GeForce 7950 GX2 can play the game in the Next Generation Content mode, but not in resolutions above 1280x1024. It is in this mode that the game gives you the highest image quality, so you should better lower the resolution than lose in the level of detail and special effects.
It’s virtually impossible to play the game with enabled 8x SLI AA or 8x Super AA, yet we can note that the GeForce 7950 quad-SLI and the Radeon X1950 XTX CrossFire are roughly equal to each other in this test.
The same is true for 16x SLI AA/14x Super AA mode. The combined power of two of today’s fastest graphics cards is still not enough to run Tomb Raider: Legend with a level of antialiasing higher than 4x.
The GeForce 7 architecture doesn’t permit to use HDR (FP16) and FSAA at the same time, so we only benchmarked the game with anisotropic filtering. Without HDR, the game loses heavily in its visual appeal. As usual, we measured the speed manually with the Fraps utility, so the numbers may be somewhat inaccurate.
The Radeon X1950 XTX CrossFire platform delivers superb performance in all resolutions. Nvidia’s GeForce 7950 GX2 is slower, yet allows playing the game comfortably at 1920x1200. The results of the GeForce 7950 quad-SLI subsystem are unsatisfactory: it is in two resolutions slower than and in one resolution equal to the single GeForce 7950 GX2.
To remind you, this test is performed in Oblivion’s closed environments like houses, dungeons, etc. The game becomes much more demanding as soon as you go out.
The Radeon X1950 XTX CrossFire is the only solution here that provides a really comfortable speed in the open scenes of TES IV. The GeForce 7950 GX2, working in the max-quality texture filtering mode, has rather good results, too.
The results of the quad-SLI configuration suggest that this technology needs to be improved more.
The GeForce 7950 quad-SLI system offers a performance gain of over 100% in high resolutions. This helps it achieve a 50% lead over the Radeon X1950 XTX CrossFire. Unfortunately, it is only a second case (Prey was the first) we’ve seen so far that quad-SLI technology works so perfectly.
The quad-SLI subsystem is successful in 8x SLI AA mode, too. Titan Quest isn’t a very demanding game, though. It’s different to see the difference between 40fps and 60fps in this game due to the specifics of its genre. The basic requirement here is that the frame rate was never lower than 25fps.
Owners of premium-class multi-GPU solutions can try to enable the highest available full-screen antialiasing modes. The 1920x1200 resolution is an exception, but it is yet not very popular on gaming computers.
Graphics cards from the Radeon X1000 family do not support vertex texturing and cannot use the appropriate shaders to render the water surface with the highest possible quality in this game. This quality is only achievable on Nvidia’s GeForce 7 cards.
Although the game engine is optimized for the GeForce 7 architecture and uses OpenGL by default, the GeForce 7950 quad-SLI has problems here, too. Like in many other cases, it provides a zero performance growth. Note, however, that the CrossFire system does the same, so this is probably the game’s problem rather than ATI’s or Nvidia’s.
The GeForce 7950 quad-SLI is sufficiently fast in 1600x1200, but the single GeForce 7950 GX2 offers the same speed (being in fact a SLI tandem, it allows to use SLI AA).
It’s important to have as high a frame rate as in first-person shooters to play a flight sim normally. An average speed of 60fps and higher is what you need. That’s why if you use the hardest FSAA modes on your Force 7950 quad SLI or Radeon X1950 XTX CrossFire, you should limit yourself to 1280x1024.
Quad-SLI technology works correctly, but its performance scalability is poor, as opposed to that of the Radeon X1950 XTX CrossFire. The latter outperforms the single Radeon X1950 XTX by 40% in 1920x1200, which is a good, although not perfect, result.
ATI’s and Nvidia’s solutions are equals when we enable SLI AA and Super AA. ATI wins the low resolutions, but the GeForce 7950 quad-SLI system is superior in 1920x1200.
The GeForce 7950 quad SLI slows down after we turn on 16x SLI AA. The game speed is below comfortable even in 1280x1024. And on the contrary, the Radeon X1950 XTX CrossFire switches from 8x to 14x Super AA easily enough and only lacks speed in 1920x1200 resolution.
The game doesn’t support non-standard FSAA modes, so we can only publish the data for 4x FSAA mode.
Quad-SLI technology works incorrectly again. Installing a second GeForce 7950 GX2 into the system and enabling the appropriate mode leads to a performance hit of 50-100% depending on the resolutions. The Radeon X1950 XTX CrossFire is free from such problems, although its minimum speed is lower than that of the single card.
CrossFire technology provides an up to 50% performance growth and the graphics subsystem is not the bottleneck anymore because the speed doesn’t grow up above 45fps. The GeForce 7950 quad-SLI brings about a performance increase, too, but smaller than 50%.
Unfortunately, ATI’s tandem cannot repeat its successful performance in Super AA mode. The quad-SLI system is no better, though. These multi-GPU configurations cannot provide enough speed even for this rather undemanding (in comparison with first-person shooters) real-time strategy.
The top-end multi-GPU subsystems look even worse in 16x SLI AA/14x Super AA mode. The numbers have no practical value for a gamer.
The Radeon X1950 XTX CrossFire tandem is in the lead, outperforming the GeForce 7950 quad-SLI complex by as many as 1765 points. This is quite a discomfiture considering that Nvidia’s solution boasts four GPUs, a total of 96 pixel processors, 96 texture-mapping units and 64 ROPs.
The first test confirms that the total scores are correct. The quad-SLI is hardly any better than the single GeForce 7950 GX2. It becomes a leader in 8x Super AA mode, but this has no practical value since 3DMark is not a real game. The Radeon X1950 XTX CrossFire regains the leading position after switching to the higher level of antialiasing.
It’s all exactly the same in the second test. We can’t say quad-SLI technology doesn’t work. It provides a certain performance gain over the single GeForce 7950 GX2, yet this gain is too small to justify the purchase of a second card for over $500.
The small advantage of the quad-SLI system over the single GeForce 7950 GX2 diminishes almost to zero in the third test whereas the CrossFire platform shows its best. Perhaps the quad-SLI system will have a higher speed at the default driver settings, but as we wrote in our previous review, the default settings produce an awful texture filtering that is far inferior to what the competitor offers.
Although the single Radeon X1950 XTX is a little slower than the single GeForce 7950 GX2, the multi-GPU systems based on these graphics cards behave differently: the GeForce 7950 quad-SLI is again worse than the Radeon X1950 XTX CrossFire by an impressive 1189 points.
The quad-SLI gets closer to the CrossFire platform in the SM2.0 tests – the gap is only 383 points. This agrees with the results of the single cards and is indicative of the fact that the pixel processor performance is not a decisive factor for the SM2.0 tests.
The SM3.0/HDR tests are quite a different story: the Radeon X1950 XTX CrossFire is suddenly ahead here, increasing the lead over the quad-SLI to 1167 points. This is a gap of about 25%!
Once again the quad-SLI system works incorrectly, performing much slower than the single card. There are no problems in SLI AA mode and the quad-SLI platform, as usual, outperforms the Radeon X1950 XTX CrossFire at 8x FSAA but loses to it at 16x FSAA. It must be due to full-screen antialiasing that the data do not agree with the overall scores for the SM2.0 tests.
The same is true for the second test. The only difference is that the single Radeon X1950 XTX equals the performance of the single GeForce 7950 GX2 because the second SM2.0 test doesn’t require a high fill rate as the first test does.
So, it’s now possible for you to assemble a GeForce 7950 GX2 quad-SLI system with your own hands. You only need an appropriate mainboard and a couple of GeForce 7950 GX2 cards. About half a year has passed since we were introduced to quad-SLI technology, and Nvidia took this time to do a colossal but successful job on solving the driver-related issues. We don’t see a corrupted image, spontaneous system reboots and other unpleasant things. However, there are still a number of problems with the Control Panel. The current version of ForceWare doesn’t always enable the full-screen antialiasing mode that you select in the Control Panel and you have to enable/disable the appropriate option there to make it work. The same option accessible through the tray icon doesn’t work at all.
Despite the small deficiencies, we can now evaluate the GeForce 7950 quad SLI basing only on its performance rather than on our negative impressions as in the previous review of that technology.
Our tests have shown that the GeForce 7950 quad-SLI system is hardly worth its cost. It’s only in a few applications that we’ve observed a considerable performance boost in the standard modes (1280x1024, 1600x1200 and 1920x1200 resolutions with 4x FSAA and 16x AF), namely in F.E.A.R., Prey, Titan Quest, and 3DMark06. This is only four applications out of twenty! A minor increase of speed was observed in five more games: Far Cry, Quake 4, Serious Sam 2, Tomb Raider: Legend and X3: Reunion. Quad-SLI technology looks even worse than that if you compare it with the Radeon X1950 XTX CrossFire tandem which improves performance in 13 tests out of 20 and only refused to work in Hitman: Blood Money and Tomb Raider: Legend.
So, under typical conditions the GeForce 7950 quad-SLI system wins in 5 applications out of 20 (F.E.A.R., Prey, Tomb Raider: Legend, Titan Quest and IL-2 Pacific Fighters) whereas the Radeon X1950 XTX CrossFire is better in 12 applications out of 20 (Battlefield 2, Call of Duty 2, Ghost Recon: Advanced Warfighter, Half-Life 2: Episode One, Serious Sam 2, Splinter Cell: Chaos Theory, Elder Scrolls: Oblivion, X3: Reunion, Age of Empires 3, Rise of Nations, 3DMark05, and 3DMark06). In the remaining three games (Far Cry, Hitman Blood Money, Quake 4) the two competing graphics subsystem deliver similar performance.
As for extreme antialiasing modes like 8x, 14x and 16x in different resolutions, the Nvidia GeForce 7950 quad-SLI system isn’t unrivalled, either. Here’s a small table that lists the maximum quality settings in different games for the GeForce 7950 quad-SLI and the Radeon X1950 XTX CrossFire.
As for the minimum level of comfort, we base ourselves on the subjective impressions that may vary depending on the person. We guess the following speeds are comfortable:
click to enlarge
You can see that even in the highest resolution and with extreme levels of antialiasing, the quad-GPU system from Nvidia wins in fewer cases than the dual-processor solution from ATI.
Summing everything up, we should note that although the subsystem with four Nvidia GeForce 7900 chips doesn’t spontaneously reboots and shows no rendering artifacts, quad SLI technology still needs improvements in terms of driver support. There are problems with the control panel, there are performance hits as you add a second card into the system, there is poor performance scalability and a lot of other drawbacks that clearly show that quad SLI hasn’t yet matured really.
It’s not yet clear if it ever does. A few days ago there appeared photographs of the next-generation graphics card Nvidia GeForce 8800. Judging by them, the card is going to have two +12V power connector, so the power consumption of the GeForce 8800 may be as high as 150W. Thus, even if dual-processor cards clock their GPUs at a reduced frequency, the power consumption of a graphics subsystem with four GeForce 8800 will hardly be lower than 500W (it may rather be 600W or more). The market offers 1000W and higher power supply even today, so such a high level of power consumption is not a problem. The problem is how to dissipate those 500 watts from the graphics processors only. It means that the promotion of quad SLI technology will stop or at least slow down due to the practical issues involved in building systems with four graphics processors.
The Gigabyte GV-3D1-7950-RH has the same highs and lows as the reference GeForce 7950 GX2 because it is a copy of the reference card but with Gigabyte’s stickers, accessories and packaging. That’s actually the fate of every premium-class graphics card today.