by Alexey Stepin
10/08/2006 | 08:13 PM
The present renaissance of multi-processor graphics hardware commenced back in 2004 when Nvidia announced its SLI technology that would allow using two cards as a single graphics subsystem. In its early days, the SLI platform had all manner of compatibility problems and depended wholly on the driver support for specific games, but all the issues were soon successfully solved and the technology won recognition among enthusiasts who are demanding the maximum possible performance from their computers.
The other leading GPU maker, ATI Technologies, responded to Nvidia with its CrossFire technology that had a number of differences from Nvidia’s SLI. Particularly, ATI’s solution used a hardware frame-compositing unit based on a programmable array Xilinx Spartan. As a result, graphics cards in a CrossFire tandem differed from each other: the Master card was equipped with a Compositing Engine and the Slave card lacked it. The technology was later modernized so that two identical cards could be used.
Still, there was one basic limitation with both ATI CrossFire and Nvidia SLI. You had to put two graphics card into your computer and so you needed a mainboard with two PCI Express x16 slots. Attempts were made to create graphics cards with two GPUs on one PCB that would occupy one PCI Express slot, but none of them, including the ASUS EN7800GT Dual, ever really took off, particularly because such cards, notwithstanding the single-PCB design, still required a SLI-supporting mainboard.
Announced in early June this year, the GeForce 7950 GX2 was Nvidia’s own attempt to create a dual-processor premium-class graphics card with unrivalled performance and without limitations on its use with different chipsets. Our tests showed that the attempt wasn’t entirely a success (for details see our article called Two for One: Nvidia's Dual-Chip GeForce 7950 GX2 Reviewed). The card indeed delivered superb performance in a majority of benchmarks we tried it in, but Nvidia hadn’t got rid of the compatibility issues completely. Despite the use of a special chip responsible for switching the PCI Express lanes, the new card wouldn’t launch on a number of mainboards. Particularly, it didn’t start up on our power consumption measuring testbed based around an Intel Desktop Board D925XCV. At the time of our writing our first review of the GeForce 7950 GX2, the list of compatible mainboards was only 39 names long.
Today, there are 64 entries there, 15 of which are ASUS’ products and 45 of which are mainboards based on Nvidia’s chipsets. In this review we’ll take a look at ASUS’ version of the GeForce 7950 GX2. It is called EN7950GX2/2PHT/1G. Let’s see if one of the major hardware manufacturers has got any surprises in store for us?
Just like any other top-end graphics card from ASUS, this one comes in a really huge box:
It has a carry handle for you to take it home. There is a picture of King Kong on the cover of the box because the software bundle includes the appropriate game. There box still has a flap, but there is no window under it so that you could take a look at the card without taking it out of the package. Inside the colorful wrapping, there is a white cardboard box. The accessories are placed in the compartments and the graphics card itself is firmly fixed in foam rubber to avoid damage during transportation. Here’s the list of accessories supplied with the card:
This doesn’t look gorgeous even in comparison with the ASUS EAX1900XTX, not to mention the EAX1800XT TOP, although this is a premium-class product! Perhaps ASUS tried to reduce the card’s cost, which is of course good, but there’s no trace of luxury associated with the ASUS brand left here. They didn’t even include S-Video and RCA cables for connecting to TV-sets. This doesn’t seem right for a solution of that class, yet some users may think it good that they don’t have to pay extra for games and accessories they will never use.
All senior models in the GeForce 7 family are manufactured by Nvidia’s order at Foxconn and do not differ from the reference samples, so Nvidia’s partners don’t have too many ways to differentiate their products to appeal to the potential customer. They can only experiment with the packaging and accessories or install an original cooling system. The latter is rather rare among top-end cards, and the ASUS EN7950GX2 isn’t going to make an exception. Developing an original cooler that would cope with a GeForce 7950 GX2 with its cramped dual-PCB design is a daunting task and would also make the end product more expensive.
As for compatibility, ASUS’ website is more optimistic than Nvidia’s and lists 21 mainboards from ASUS alone.
As might have been expected, the ASUS EN7950GX2 is an exact copy of the reference sample of the GeForce 7950 GX2 that we have in our labs. The single difference is the ASUS sticker on the top cooler’s fan:
We described the design of the GeForce 7950 GX2 in detail in our earlier review. In brief, the graphics card has two PCBs that are connected with four hex-headed poles and a two-section connector that is responsible for transferring power, MIO and PCI Express signals from one PCB to another.
Each PCB carries one G71 processor and 8 chips of GDDR3 memory from Samsung (K4J52324QC-BC14). Besides that, the bottom PCB is equipped with a PCI Express x16 slot, a MIO connector for connecting two such graphics cards into a quad-SLI system and a PCI Express switch. The top PCB in its turn carries DVI-I and TV-Out/HDTV outputs and an additional power connector.
The graphics cores and the memory chips of the ASUS EN7950GX2 are clocked at 500MHz and 600(1200)MHz, respectively, in full compliance with the official GeForce 7950 GX2 specification (some companies, for example XFX, offer pre-overclocked versions of the GeForce 7950 GX2 that provide increased performance). Although sixteen 512Mb chips of GDDR3 memory give you a total of 1GB, applications can only use 512 megabytes. This is the case with any of today’s multi-GPU consumer-class graphics subsystems.
The cooling system installed on the ASUS EN7950GX2 consists of two standard flat coolers. In spite of their low profile, they do their job well enough thanks to the copper cores that have contact with the GPU dies and to the heat pipes that make sure the heat is distributed uniformly in the heatsinks. The coolers’ task is simplified by the relatively low heat dissipation of the G71 chips clocked at a reduced frequency. There is a layer of dark-gray thermal grease between the coolers and the GPUs; the memory chips have contact with the heatsink through special pads that consist of two pink-colored halves and a layer of white thermal grease in between. The fans of the coolers have a green LED-based highlighting.
The deficiencies of the cooling system installed on the GeForce 7950 GX2 come from the cramped dual-PCB design of the card which makes it hard for at least one of the coolers to work. As opposed to the GeForce 7900 GX2, there are no slits in the PCBs of the GeForce 7950 GX2, so the bottom fan, situated very close to the top PCB, has problems getting enough air. This may eventually lead to overheat of the bottom GPU.
We had to face the problem of overheat closely this time around, unfortunately. Although the testbed was ventilated well, the bottom fan on one sample of the GeForce 7950 GX2 we used for our tests failed suddenly. For some reason, the card’s overheat protection didn’t work and the temperature of the bottom GPU jumped up so high that you could smell the rosin. Rosin melts at about 100-120°C, so that’s how hot the PCB around the graphics processors could have been! Luckily, the card survived through that catastrophe.
We won’t claim that the fans on GeForce 7950 GX2 cards are prone to fail, but this may happen anyway, especially if dust accumulates in the gap between the bottom cooler and the top PCB. It’s difficult to replace the GeForce 7950 GX2’s cooler with something more efficient due to the peculiar design of the card, so we can only recommend that you check out and clean its cooler regularly.
We measured the performance of the ASUS EN7950GX2 in the following testbeds:
We select the highest graphics quality settings in each game, identical for graphics cards from ATI and Nvidia. We do not modify the games’ configuration files and do not use the drivers’ profiles optimized for specific games. The frame rate is measured with the game’s own tools or, if not available, with the Fraps utility. We also measure a minimum frame rate where possible.
Besides the two standard resolutions of 1280x1024 and 1600x1200 pixels, we also used 1920x1200 resolution (with an aspect ratio of 16:10) in games that support widescreen modes. The reviewed card belonging to the top-end product category, we didn’t test it without full-screen antialiasing. We turned on FSAA and anisotropic filtering from the game menu. If such options were unavailable, we forced FSAA and AF through the appropriate options of the ATI Catalyst and Nvidia ForceWare driver.
Besides the ASUS EN7950GX2, the following graphics cards took part in the tests:
ATI’s and Nvidia’s drivers were set up according to our new updates procedure described in detail in the article called Quality vs. Quantity: ATI Radeon X1950 XTX against Nvidia GeForce 7950 GX2 with Maximum Quality:
ATI Catalyst :
Nvidia ForceWare :
The cards were tested in these games and benchmarks:
First-Person 3D Shooters
Third-Person 3D Shooters
Our using new texture filtering settings doesn’t bring any changes into the ranks: the ASUS EN7950GX2 is 15-20% faster than the Radeon X1950 XTX. The latter has one GPU and only 16 texture-mapping units (TMUs), but provides a higher image quality by using a more advanced anisotropic filtering algorithm. These cards both deliver an average performance of over 100fps, so there’s no practical worth in Nvidia’s lead. It is comfortable to play the game on either of the cards.
Call of Duty is a heavier application for the graphics subsystem than the previous test. Top-end single-chip solutions can only approach a comfortable level of performance in resolutions of 1280x1024 and lower if we use FSAA and anisotropic filtering. The dual-chip EN7950GX2, however, feels confident in 1600x1200 even if you enable the highest-quality texture filtering available on the GeForce 7 architecture.
Here’s the first triumph of ATI Technologies. Its single-chip card is rivaling the competitor’s dual-chip product in resolutions up to the yet-exotic 1920x1200.
The EN7950GX2 is ahead of the Radeon X1950 XTX in high resolutions on the Research map, but the gap is a mere 5% or less. Considering that the GeForce 7 architecture suits better for this demo record, which shows a walk along an underground base, than the Radeon X1000 architecture, it looks like ATI wins this round, too.
The EN7950GX2 has no rivals when we enable HDR (FP16), but it’s only because the high dynamic range implementation in the version 1.4 patch isn’t optimized well. Anyway, owners of Radeon X1950 XTX cards can play in the same resolutions as are available to people who have a GeForce 7950 GX2.
Notwithstanding its two graphics processors and 48 pixel processors, the min performance of the EN7950GX2 isn’t far higher than that of the Radeon X1950 XTX. The dual-GPU solution has a bigger advantage in terms of average performance, from 15% to 20% depending on the resolution. Still, both the flagship products are capable of running F.E.A.R. at an acceptable speed with enabled full-screen antialiasing.
It is impossible to use FSAA in this game because it employs the deferred rendering technique. That’s why we can only show you the numbers for the anisotropic filtering only mode.
The ASUS EN7950GX2 is doing well here and meets no competition. It can run the game at a sufficiently high speed in 1600x1200 and even in 1920x1200. The Radeon X1950 XTX is limited to 1280x1024, yet it isn’t far slower than the GeForce 7900 GTX and is superior in terms of image quality thanks to its high-quality anisotropic filtering.
The GeForce 7950 GX2 used to win in all the resolutions, but now the Radeon X1950 XTX somehow beats it in the resolution of 1280x1024 pixels. The EN7950GX2 is superior in the harder modes, but the ATI card allows to play in 1920x1200 with comfort. The user has to face a choice between a higher average frame rate and a near-ideal image quality.
This is an OpenGL application, yet the ATI Radeon X1950 XTX doesn’t give up. It is a little ahead of the GeForce 7900 GTX which is put under the same conditions by disabling the texture filtering optimizations. The single-chip solutions find it hard to challenge the dual-chip ASUS EN7950GX2, though. This graphics card allows playing Prey in high resolutions with the best possible filtering quality and enabled full-screen antialiasing. The GeForce 7900 GTX doesn’t permit that.
The results are similar to those of the previous test, but Quake 4 being a less heavy application, every graphics card included in this review can run it at a high enough speed in high resolutions and with enabled 4x FSAA and 16x anisotropic filtering. The Radeon X1950 XTX is the leader in terms of image quality whereas the ASUS EN7950GX2 is unrivalled in terms of sheer speed.
It is in Serious Sam 2 that our turning on the max-quality texture filtering mode on the GeForce 7 architecture affects the speed of the cards the most. This shooter presents a very high textural load, most of its shaders using at least 4 and up to 7-8 textures. So, the effect from our disabling the optimizations of tri-linear and anisotropic filtering is so strong that the EN7950GX2 can only retain its leadership in the resolution of 1280x1024. In 1600x1200 its performance equals that of the Radeon X1950 XTX, and in 1920x1200 ATI’s solution is ahead! Moreover, the EN7950GX2 doesn’t allow playing the game in 1600x1200 with enabled FSAA and anisotropic filtering because its average speed is less than 50fps. That’s real astonishing for a game the GeForce 7 architecture has always been superior in.
The ASUS EN7950GX2 doesn’t lose much of its speed on switching from the Quality mode (with optimizations) to the High Quality mode in which there are no optimizations. This dual-chip solution wins everywhere, although the Radeon X1950 XTX with its super-fast memory is very close to the leader in 1920x1200. This isn’t very interesting from a practical point of view because 1600x1200 pixels is the highest resolution you can play this game at comfortably.
The ASUS EN7950GX2 holds on to its top place, but not as confidently as in the Quality mode with enabled texture filtering optimizations. In 1920x1200 it is almost overtaken by the Radeon X1950 XTX that has the same number of pixel shader processors and a higher memory performance. The Radeon doesn’t slow down much from switching to the High Quality AF mode.
Our disabling the optimizations affects the speed of the GeForce 7960 GX2 as dramatically in this game as it did in Serious Sam 2 . The card could yield about 60fps in 1280x1024 before, even in the Next Generation Content mode, but now the average frame rate is about 50fps or 15% lower. As opposed to it, the Radeon X1950 XTX doesn’t slow down at all in its high-quality anisotropic filtering mode.
As before, the flagship solutions are both not strong enough to enable the gamer to play in resolutions above 1280x1024, and even in that resolution the speed may bottom out to below 25fps.
The Nvidia GeForce 7 architecture doesn’t allow using HDR (FP16) along with FSAA, so we benchmarked the cards in TES IV: Oblivion with anisotropic filtering only. Without HDR, the game loses much of its visual appeal. As usual, we measured the speed manually with the Fraps utility, so the numbers may be somewhat inaccurate.
It’s all the same, at least in closed game scenes. The Radeon X1950 XTX is still the best of single-chip solutions. The dual-chip ASUS EN7950GX2 is faster still, providing a frame rate of almost 60fps in 1920x1200 resolution.
In spite of its higher scene fill rate, the EN7950GX2 loses its ground in 1600x1200 where the Radeon X1950 XTX provides a higher minimum of speed.
There’s a change of the leader in Titan Quest : the GeForce 7900 GTX now wins in every resolution, including 1920x1200. The EN7950GX2 slows down so much that it even finds itself behind the Radeon X1900 XT. That’s the effect of our disabling the driver optimizations by switching to the High Quality mode. The Radeon X1950 XTX also slows down by 16-17% to 20% on switching to its high-quality anisotropic filtering algorithm.
Radeon X1000 family cards do not support vertex texturing, so they cannot use Shader Model 3.0 to render the water surface with the highest quality in this game. This is only possible on Nvidia’s GeForce 7 cards.
The GeForce 7900 GTX is surprisingly slow if compared with itself in the Quality mode, but the EN7950GX2 is, on the contrary, a little faster, nearly reaching 85fps. This is good news for all the virtual pilots who run this game on a GeForce 7950 GX2.
The GeForce 7950 GX2 used to compete with the Radeon X1950 XTX in 1600x1200 and higher resolutions before, but now it can do so in 1920x1200 only. From a gamer’s point of view, the EN7950GX2 and the Radeon X1950 XTX are equally well suited for space flights in X3 , but the latter provides a bigger reserve of speed in low resolutions.
The ASUS card takes a humble third place and delivers lower minimum performance than the Radeon X1900 XTX in resolutions above 1280x1024. On the other hand, it provides a comfortable frame rate in every resolution with enabled 4x FSAA.
There’s a difference of one GPU and 10-12% of speed between the ASUS EN7950GX2 and the Radeon X1950 XTX. The latter isn’t much slower for using the more advanced anisotropic filtering method. But telling you the truth, the better image quality won’t be conspicuous in this game as well as in other real-time strategies where the player uses a view from above most of the time.
There are almost no changes in comparison with the older texture filtering settings: the ASUS EN7950GX2 is in the lead, but its advantage over the Radeon X1950 XTX is rather small, less than 1000 points, because 3DMark05 defaults to a resolution of 1024x768 pixels without FSAA.
Things are more interesting when we turn on full-screen antialiasing: the Radeons both lose much of their advantage. The Radeon X1900 XT, which was ahead of the GeForce 7900 GTX, now goes abreast to it whereas the lead of the Radeon X1950 XTX shrinks from 20-25% to 5-10%. The GeForce 7 copes with the increased load better. The performance of the EN7950GX2 degenerates by a mere 5%.
It’s a different picture in the second test, and the results are largely the same as at the older texture filtering settings. The only difference is that all the cards have become slower at the new settings.
The third test shows the same picture as the first one: the ASUS EN7950GX2 and the GeForce 7900 GTX suffer a smaller performance hit than the Radeon X1900 XT and X1950 XTX do. The latter remains the leader among the single-chip cards, though.
The performance of the GeForce 7950GX2 plummets down to the level of the GeForce 7900 GTX in 3DMark06. The High Quality mode must have affected some driver optimizations for SLI mode in this benchmark. If it is so, we are going to see the same in individual tests, too.
Our point is confirmed by the results we have in the SM2.0 tests. Although the EN7950GX2 is faster than the Radeon X1950 XTX, it is not as faster as it used to be. It is only a mere 47 points ahead of the GeForce 7900 GTX, which is a negligible difference.
In the SM3.0/HDR tests the results of the EN7950GX2 coincide with the results of the GeForce 7900 GTX with an up to 1 point precision.
SLI mode works here indeed. Otherwise, the results of the EN7950GX2 would be much lower (only 24 TMUs would be active then instead of 48).
It’s different in the second SM2.0 test: the EN7950GX2 wins in 1280x1024 but then slows down to the level of the Radeon X1950 XTX in the higher resolutions. This must be due to the deactivation of certain SLI optimizations in the ForceWare driver because the GeForce 7900 GTX suffers a much smaller performance hit in comparison with its dual-chip relative.
As we wrote earlier, the GeForce 7950 GX2 had used to be the fastest card in most of our benchmarks at the old quality settings, but doesn’t look so confident at the High Quality settings, i.e. without the optimizations. Its performance has fallen to the level of the Radeon X1950 XTX or even lower in such games as Far Cry, Half-Life 2: Episode One, Serious Sam 2, Splinter Cell: Chaos Theory, Tomb Raider: Legend, TES IV: Oblivion, Titan Quest, X3: Reunion, Age of Empires 3 and in 3DMark06. This is 50% of the applications we use to benchmark graphics cards! And although the tri-linear filtering quality provided by Nvidia’s flagship product has grown considerably, the overall image quality (at our settings) is still lower than what the Radeon X1950 XTX provides in its high-quality AF mode.
The ASUS EN7950GX2 doesn’t differ from the ordinary GeForce 7950 GX2 in anything save for the sticker on the cooler. It embodies the design we described in an earlier review. ASUS doesn’t supply luxurious accessories with this graphics card, so the size of its package is perhaps the single point of difference of this version of the GeForce 7950 GX2 from others. It’s sad because ASUS’ top-end products used to have some special feature in the design of the card or its cooler, or in the accessories included with it (you could expect to find a web-camera with an integrated microphone or a very good ASUS XitePad joystick).
On the other hand, additional accessories would have made the graphics card even more expensive, so perhaps the scanty accessories of the ASUS EN7950GX2 aren’t such a great drawback after all.