Tech-Hounds.com

GeForce 7950GT & 7900GS

PCI Express x16 graphics cards
Chipset GeForce 7950GT 512 MB, 256 bit and 7900GS 256 MB, 256 bit

Those of us who bought the GeForce 7800GTX and still uses it today can be considered lucky. Especially, those who managed to get their hands on a GeForce 7800GTX 512 MB (or two). More than a year after its launch, the GeForce 7800GTX continues to be a benchmark for newer high end graphics cards. If you don't remember, the GeForce 7800GTX is actually the first card in the GeForce 7 series. Excluding the GeForce 7800GTX 512 MB, only the GeForce 7900GTX offers significantly faster performance compared to the GeForce 7800GTX.

Since its launch, there have been several updates for the high end GeForce 7 series, like the GeForce 7800GT, GeForce 7900GTX, GeForce 7900GT and GeForce 7950GX2. The latest updates are the GeForce 7900GS and GeForce 7950GT, launched early last month. The GeForce 7900GS and GeForce 7950GT should mark the final end of the GeForce 7 series (though there are rumors of a GeForce 7950GS).

As we said earlier, only the GeForce 7900GTX and GeForce 7950GX2 offers a significant performance upgrade over the original GeForce 7800GTX. The other updates either offer performance below or on par to the GeForce 7800GTX and the newest updates are no exception.This fact is can be viewed in two ways. You can say that high end performance got much affordable this year, but you can also say 3D graphics is not making much progress in performance. We think it's about time. Games are becoming too heavy for these cards and a new graphics API is just on the horizon. That's why we considered those who bought the GeForce 7800GTX when it came out very lucky indeed - look at just how long it can cope with the demands of new games this year.

Undoubtedly, SLI (Scalable Link Interface) also help the longevity of the GeForce 7 series. If single card performance is not enough for you, you can stick an additional, identical card to the one you're using for a healthy boost of some pixel processing power. There are some caveats with this line of thinking, since proper SLI support are needed for the game to work correctly with SLI.

With next generation graphics card on the horizon, users will less likely be willing to pay more for a graphics card than they want to. Particularly if that graphics card will be obsolete in three to six months time. Users looking to upgrade their high end graphics card will likely wait for the launch of the next generation graphics card instead of buying one right now. However, those who have to buy a PC today don't have that luxury. There's also users who want to upgrade their mainstream graphics card to a high end card to have better performance. They will also most likely be budget conscious as well. The GeForce 7900GS and GeForce 7950GT are targeted for users in this situations - the GeForce 7900GS for its value and GeForce 7950GT for its feature and price / performance.  

Overview

As the name implies, the GeForce 7900GS and GeForce 7950GT are new additions to the GeForce 7900 family of products. The GeForce 7950GT is an upgrade to the GeForce 7900GT, retaining the full specification but with a higher clocked core and memory. Instead of 450 MHz and 660 MHz for core and memory clock, the GeForce 7950GT runs its core and memory at 550 MHz and 700 MHz. You'll undoubtedly notice these clocks are not that much different to higher clocked GeForce 7900GT versions offered by various vendors (520 MHz / 720 MHz). We expect performance to be similar if not the same between the higher clocked GeForce 7900GT and 7950GT. The swing factor in this case may be the HDCP support integrated to all GeForce 7950GT cards. Like many other series before it, manufacturers have also released higher clocked GeForce 7950GT cards, but we weren't able to test one for this article. Hopefully, we'll manage to get some for our GeForce 7950GT round up.

The GeForce 7900GS is another matter. It has the core and memory running at 450 MHz and 660 MHz, still very close to its faster brethren. However, NVIDIA decided to 'snip' one vertex and shader unit, so in total there are only 7 vertex shader unit and 5 pixel shader quads (20 pixel pipelines) on the GeForce 7900GS. In this regard, the GeForce 7900GS closely resembles its earlier brethren, the GeForce 7800GT. But with a price tag of USD 200, you really couldn't complain that much. This is the first GeForce 7900 series card to reach that price point, no doubt a move by NVIDIA to counter ATI's price reduction on their Radeon X1900 series products and the soon to be introduced Radeon X1950 Pro.

We're going to take a closer look at what these cards have to offer in terms of performance. We will be comparing them to the parts they are meant to replace - the GeForce 7900GT. Since many vendors also released higher clocked versions of the GeForce 7900GT, we also included the ASUS EN7900GT TOP which runs its core and memory at 520 MHz and 720 MHz. However, let's look at the some other information before we begin.

Forceware 91.47

NVIDIA recently updated their reference driver to Forceware 91.47. From our experience, it's definitely an improvement over the earlier, Forceware 91.31. We haven't done a more thorough testing with an SLI setup, but this driver have fixed the bugs we complained about (but not all) with Forceware 91.31. Behavior with FRAPS is much improved, though the blank screen problem still persist with some games (Quake 4 and The Elder Scrolls IV: Oblivion)

.  

This latest driver marks the introduction of the new NVIDIA logo and Control Panel Interface in a reference driver. Hmm, funny. They've updated their logo in the installer but not on the Control Panel. Our experience with the new driver panel interface is, well, interesting. The new panel does make things a little bit simpler and more easy to use for less technically inclined users, but power users will likely be annoyed with the response time. They will also more likely be annoyed that some settings, like SLI, core temperature log options and video enhancements are only available on the new control panel and not the old one. Another concern for those of us who like to tweak the settings are the absence of a 'Reset to Default' button in the new control panel, though they're still present on the old.  Another 'behavior' we don't like with the new control panel is the stubbornness to apply Trilinear optimization the first time it is enabled (though it was disabled in the old control panel).

Some may consider ATI's Catalyst Control Center is the better interface and we agree. However, NVIDIA still find it in their hearts to supply old time users the old, familiar Control Panel without a driver mod or a separate download. Hopefully, they will continue to do so and fix the trivial issues we mentioned.

Heat and Noise

Though we're not completely sure, it would seem the new GeForce 7950GT has a slightly higher temperature than a higher clocked GeForce 7900GT. During our testing, we saw core temperature on the higher clocked GeForce 7900GT goes from 56° Celsius on idle to 62° Celsius on full load. Meanwhile, the GeForce 7950GT goes from 62° Celsius on idle to 68° Celsius on full load. Our version of RivaTuner is not able to monitor the temperature levels of the GeForce 7950GT and 7900GT with the new Forceware 91.47 drivers, so we will be investigating this issue further when we write our GeForce 7950GT round up.

Noise levels are generally the same, though it seems the GeForce 7900GS is a little bit 'noisier'. We could definitely hear the fan at high speed when playing games, even with a headphone attach to our ear. Mind you, our test setup is an open case environment, so the noise we heard shouldn't be much of a problem for users with a closed case.

Performance

As usual, we'll be focusing on performance under default settings and with 4x AA and 16x AF. Anti aliasing and anisotropic filtering are applied from the in-game menu settings (or console) whenever possible. V-sync was disabled on both the driver panel and within the game, if present. The settings we used for each game are pretty much the same as our usual test settings. F.E.A.R was set to 'Maximum' quality (for both graphics and computer settings) but with Soft Shadows turned off to allow performance testing with AA. Quake 4 was set to 'High Quality' and Serious Sam II was set to its 'Maximum' video options preset but we used custom resolutions, AA and AF settings for both of these games.

Due to a misconfiguration in our part, we messed up the results for Splinter Cell: Chaos Theory. As soon as we can, we will update the results. 

We'd like to thank Leadtek for supplying the GeForce 7950GT and 7900GS and Tagan for supplying the additional power supply for this article.

Our test setup
AMD Athlon 64 3500+ socket 939
2 x 1024 MB Kingston KVR 3-3-3 PC3200 DDR-SDRAM
MSI K8N NForce 4 SLI motherboard
Maxtor DiamondMaxPlus9 80 GBs Serial ATA 8 MB buffer
ASUS E-616 DVD-ROM
Tagan TG530-U15 530 watts ATX/BTX power supply

Windows XP Professional with Service Pack 2 installed
NVIDIA Forceware 91.47 reference driver
NVIDIA NForce 4 6.66 reference driver
Creative SoundBlaster Live! 24 bit 5.12.1.512 driver.
DirectX 9.0c

The graphs are pretty self explanatory, but in case you can't see the text, they're arranged in groups of three colors: green for minimum fps, blue for average fps and red for maximum fps. The first group represent test results under default settings, and the second represent results with 4xAA and 16xAF enabled. For High Dynamic Range (HDR) Rendering, the second group represents results with HDR enabled. These results are in frame rate per second.

The results:

Not surprisingly, the performance of both GeForce 7900GS and 7950GT is similar to the reference clocked GeForce 7900GT and higher clocked 7900GT, respectively. So the two sits really well as a replacement for the GeForce 7900GT series. With its slightly higher clock, the GeForce 7950GT offers roughly 2 to 3 fps higher performance than the ASUS EN7900GT TOP we used to represent the higher clocked GeForce 7900GT. As you can see, the difference is not at all noticeable in gameplay or even significant. At 1280 x 960 with AA and AF, the reference clocked GeForce 7950GT can still drop to below 30 fps in some situations, while at 1600 x 1200, both the higher clocked GeForce 7900GT.and 7950GT are practically the same, though the latter comes with more memory.

Variations between runs does complicate things, but it seems the GeForce 7900GS generally has higher minimum frame rates than the standard GeForce 7900GT. From the average fps, the GeForce 7900GT is overall the better card of the two, no doubt due to its number of pixel pipelines. The two are evenly matched, just like the GeForce 7950GT and the higher clocked GeForce 7900GT.

Due to our experience when writing the 'Benchmarking on Trial' article, we decided to add results from gameplay testing sessions for this article. The purpose is simple - to see what these cards have to offer in situations as close to real world situations  as possible. From our experience, results with this kind of testing tend to be lower than results from the F.E.A.R Performance Test. Here are they are - we've separated the first and second part of Afterimage to make the graphs easier to see.


The results from the first part of Afterimage generally falls in line with the results we got from F.E.A.R Performance Test. Generally, because as we can all see, minimum frame rates with AA and AF are much lower than what we're seeing with F.E.A.R Performance Test. From the results, even the fastest card, the GeForce 7950GT, is not enough to push frame rate above 30 fps all the time in F.E.A.R with AA and AF enabled. To do that, looks like you'll need at least a GeForce 7900GTX or equivalent performing card. The trend we saw earlier persists - the GeForce 7900GS is slightly slower than the standard, reference clocked GeForce 7900GT while the GeForce 7950GT is slightly faster than a higher clocked GeForce 7900GT.

In graphics limited situations, the GeForce 7900GS is about 10 percent slower overall than the standard GeForce 7900GT. Not that it mattered, since minimum fps on both card are practically the same give or take some variations between runs (variations between runs are generally higher with gameplay testing sessions than in F.E.A.R Performance Test). The GeForce 7950GT fared slightly better in this regard - minimum frame rates can be significantly higher than the higher clocked GeForce 7900GT (around 10 fps at 1024 x 768). At 1280 x 960, the difference is smaller and it became practically non existent at 1600 x 1200 or when AA and AF is enabled.


Afterimage 2 paints a similar image to Afterimate 1 - both the GeForce 7900GS and standard GeForce 7900GT is not enough to run F.E.A.R at 1024 x 768 with AA and AF. The faster cards like the higher clocked GeForce 7900GT and GeForce 7950GT just barely pass the minimum 30 fps mark, so even with these cards you're not completely out of the woods. Strangely enough, it seems that in both Afterimage 1 and 2, the GeForce 7950GT has slightly lower minimum fps than the higher clocked GeForce 7900GT at 1024 x 768, with AA and AF enabled. At higher resolutions, the GeForce 7950GT is faster, the way it should be.

Not surprisingly, the standard GeForce 7900GS is about 10 to 13 percent slower than the GeForce 7900GT. With AA and AF, the gap between them is smaller, something like 7 to 10 percent. The slightly higher pixel processing power of the GeForce 7900GT is still not enough to allow it to pass the  30 fps minimum mark. If you're going to buy one of these cards, better stick to 1024 x 768 without AA and AF, if you want to run F.E.A.R at nearly full settings. That also seem to apply to both the higher clocked GeForce 7900GT and GeForce 7950GT, though your mileage may vary.

When we begin using F.E.A.R, we concluded that you'll need at least a higher clocked GeForce 7800GTX or equivalent performing card to run this game at close to the highest setting possible. Minimum frame rates should at least be in the mid thirties or near 40 fps, if possible. That's still true today and only the higher clocked GeForce 7900GT, 7950GT and 7900GTX are able to push frame rate fast enough without dipping too much below the 30 fps mark in graphically intensive situations.


Quake 4 is a different beast all together. Running without AA and AF, all these cards are all system limited at 1024 x 768. Once we bumped up the resolution to 1280 x 1024 and 1600 x 1200, we begin to see a definitive difference between the four cards. The difference is small at 1280 x 1024, small enough that you will not notice it but it is there. At 1600 x 1200, the GeForce 7950GT is leading with a significant margin in minimum fps - no doubt helped by the larger memory. However, owners and would be owners of GeForce 7900GT and the GeForce 7900GS have nothing to fear, minimum fps are still well above 50 fps at this point.

With AA and AF enabled, the higher clocked cards quickly pull away from the slower ones. The difference is more noticeable at 1280 x 1024 and at 1600 x 1200. The GeForce 7900GS and standard GeForce 7900GT dropped to below 30 fps at least one time while the higher clocked GeForce 7900GT and GeForce 7950GT can still maintain frame rate slightly above 30 fps. Now, we'll need to see if the same is true in more graphically intensive situations.


Undoubtedly, this timedemo is more system limited than our gameplay testing session results from the Data Processing Terminal level. We don't even see that much of a difference between the GeForce 7900GS and 7950GT at 1600 x 1200. Then why test them at all, you may ask? These results should be closer to what we actually see throughout the entire game. Remember, Quake 4 is mostly a system limited game. It's not until we hit 1280 x 1024 with AA and AF do we see a difference between the four cards. Surprisingly, the jump in average fps from the higher clocked GeForce 7900GT to the GeForce 7950GT is quite high - around 5 fps or 10 percent. We're inclined to once again believe the larger memory is a factor here, making the GeForce 7950GT the only card to pass the 30 fps minimum mark at 1600 x 1200, with AA and AF.

Those of you who are planning to buy a single GeForce 7900GS should be aware that this card is basically a budget card derivative of a high end card. You shouldn't expect much in terms of performance. It's good enough for 4x AA and 16x AF at 1024 x 768, maybe at 1280 x 1024 with Quake 4 and that's about it. The good news is that it should be much cheaper than the card it will be replacing - the GeForce 7900GT. You can pay less to have roughly the same performance.


Strangely enough, the GeForce 7900GS seems slightly faster than both GeForce 7900GT at 1024 x 768 and 1280 x 960. This demo doesn't have that high of variations between runs, so it's very likely something else is at play here. Better memory management perhaps? Overall, the results speak for themselves. You'll get a similar level of performance from all of these cards until you hit 1280 x 960 with AA and AF. Not surprisingly, the GeForce 7950GT takes the lead, leaving the higher clocked GeForce 7900GT by 3 - 5 fps and the standard GeForce 7900GT by 8 - 9 fps or something around 6 to 10 percent faster than the higher clocked GeForce 7900GT.

These results again shows us that the GeForce 7900GS seems to be better suited for 1024 x 768 with AA and AF. At 1280 x 960, it's the slowest of the four cards here, though it still manages a fairly high enough frame rate to allow for comfortable gameplay. At 1600 x 1200, only the GeForce 7950GT has enough pixel processing power to push minimum frame rates slightly above 30 fps in this demo.


Quite a new addition to our benchmark suite, Need for Speed: Most Wanted is quite a burden on these cards. Notice how small the differences are separating these cards. The GeForce 7900GS holds the last position at 1024 x 768 with AA and AF, barely making the minimum 30 fps mark, which is expected at this point. At 1280 x 1024, only the GeForce 7950GT can claim that honor, all other cards have gone below 30 fps at some point in the test. If you want to play at 1600 x 1200, you'll have to get an even faster card or try to overclocking the card by a significant margin. At 1600 x 1200, with or without AA and AF with all four cards - they are just not fast enough to push more than 30 fps all the time.


As the most graphically impressive game this year, The Elder Scrolls IV: Oblivion have made all four cards bend at their knees. Our test settings are pretty close to 'Ultra Quality', with higher Specular Distance, Internal / External Shadows and lower Tree Fade settings. Obviously, that high setting is too much for these cards, you probably want to use two of them in an SLI setup to get higher frame rates or if you want to play with AA and AF. As a single card, only the higher clocked GeForce 7900GT and GeForce 7950GT are able to provide enough pixel processing power to push frame rates higher than 30 fps. Note that this is without AA and AF. With so much textures used for grass and trees, anisotropic filtering comes with a pretty big hit. In fact, the penalty is higher than what we usually see from first person shooter games.


Nothing we can say to make it better or less painful here. Undoubtedly the most dense forest in Oblivion will put even more cramp on your gaming - frame rates offered by these four cards are so low, that only the average fps on the higher performing cards - the higher clocked GeForce 7900GT and the GeForce 7950GT is below 30 fps. That's again without AA and AF. We don't think even a single GeForce 7900GTX is fast enough in this part of Tamriel without dipping to 30 fps at some time, but you should be able to use AA and AF at 1024 x 768.


High Dynamic Range Rendering is certainly becoming a staple of next generation games and graphics engines. Unfortunately, we have to forgo anti aliasing if we want to run HDR with the GeForce 6 / 7 series of cards. HDR performance penalty is roughly the same as 4x anti aliasing, depending on the game and the amount of fillrate the card has. Not surprisingly, both the GeForce 7900GS and standard, reference clocked GeForce 7900GT is slower with HDR than AA than the higher clocked GeForce 7900GT and GeForce 7950GT. Results at 1280 x 960 confirms this, with the two slower cards dipping below the 30 fps minimum mark at one point in the test, while the faster cards are able to maintain minimum Even then, 1600 x 1200 prove too much for the faster cards.