Tech-Hounds.com

Deeper with the GeForce 7900GTX

PCI Express x16 graphics cards
Chipset GeForce 7900GTX 512 MB, 256 bit

The wait is finally over. Last month, NVIDIA launched their newest generation of graphics chips - the GeForce 7900. Just like the previous generation, there are two SKUs based on this chip: the high end GeForce 7900GTX and the more affordable GeForce 7900GT. Despite rampant speculation of a 32 pipeline graphics chip, the GeForce 7900GTX remain a 24 pipeline card, much like the GeForce 7800GTX. In fact, most of the specification between the two is pretty much the same, except for core and memory clocks. NVIDIA seems to be taking a more conservative approach this time around, no doubt not wanting to repeat what happen to the GeForceFX series.

The most significant changes on the GeForce 7900 series seem to be cost / production related. Not only did NVIDIA managed to shrink the die size of the chip (thus keeping production cost down), they were also able to throw out some transistors from the die and save more space in the process. It's no wonder that cards using this chip are selling at much lower prices than what they're charging for the GeForce 7800GTX at launch. No doubt, this is a calculated move to maintain and expand market share. Archrival ATI's X1900 based chips have a much larger die, so to counter this move, ATI have to cut their profit margins to keep their parts competitive.

Of course, while price is an important factor in any hardware purchase decision, most gamers still put performance first before everything else. However, unlike a few years ago, gamers nowadays are more aware of graphical features such antialiasing and anisotropic filtering. If the cards are fast enough (or priced high enough), it must deliver both performance and image quality. This is one area that NVIDIA seem to be 'lacking' - antialiasing usually comes at a significant drop in frame rates on NVIDIA cards. So, imagine how our ears perked up when NVIDIA claimed that the GeForce 7900GTX have much less performance penalty than the GeForce 7800GTX when running with antialiasing and anisotropic filtering enabled.

While moving to a smaller process (0.11 to 0.09 micron for the GeForce 7900) allowed manufacturers to reduce cost, there's no point of claiming the resulting product a new generation if there are no performance increases. Thankfully, clock increase is a by product of moving to a smaller process. With the same design and architecture, higher clocks (usually) means higher performance. Unlike the previous generation, the GeForce 7900GTX is clocked much higher than its successor. The reference specification is 650 MHz for core and 800 MHz (1600 MHz effective) for memory. In comparison, the GeForce 7800GTX ran at 430 MHz core and 600 MHz (1200 MHz effective). If you're paying attention, you'll probably notice that the GeForce 7900GTX memory clock is actually lower than what's on the monster that was the GeForce 7800GTX 512 - 1700 MHz. While its memory clock is slower, the GeForce 7900GTX no doubt has a much higher core clock - a 100 MHz difference to the 550 MHz of the GeForce 7800GTX 512. With that high a core clock, it seems odd NVIDIA didn't push the memory as high as they can go. Some have cited cost as the reason, but NVIDIA did mention they improve antialiasing performance on the GeForce 7900GTX. So, did they (finally) made some changes to the GeForce 7800GTX design? Maybe the GeForce 7900GTX don't really need that high memory clock. it will be interesting indeed to see just what NVIDIA tweaked on the GeForce 7900GTX.

Image Quality Comparison

For the launch of the GeForce 7900GTX, NVIDIA have updated their Forceware driver to version 84.21. In addition to several bug fixes and additional support for new cards, these drivers also provide some performance updates, particularly for F.E.A.R. Normally, any performance updates are good, but we're a little skeptic because of what NVIDIA have been caught doing in the past. Concerns about optimizations may not be that unfounded, as 3DCenter and Elite Bastards found out in their articles (which you can read here and here). So, before moving on to the performance evaluation, we took some time examining the screenshots and performance from these drivers on a GeForce 7800GTX.

After testing these new drivers, we found there's no reason to worry about questionable optimizations. Performance under High Quality and Quality settings are where they should be and turning off performance optimizations related to anisotropic filtering work as intended. Visually there's hardly any difference between these drivers and the previous versions, but let's take a look at the screenshots just to make sure.

Below you'll see several screenshots from the GeForce 7800GTX and 7900GTX. To keep things short, we skipped the non AA and AF screenshots, so it's all 4x AA with 16x AF.

Call of Duty

GeForce 7800GTX



GeForce 7900GTX



There are no differences between the screenshots, so the increase in performance seems valid. Call of Duty is one game that also experience an increase in performance with the new drivers - a 10 percent increase running without AA and AF. With AA and AF enabled, the increase is negligible, mostly 3 - 5 fps from a total of above 150 fps (1 percent). So, from the looks of it, NVIDIA managed to squeeze more performance from system limited functions, rather than just memory or rendering tweaks.

Nascar 2003

GeForce 7800GTX



GeForce 7900GTX



Same thing here, no difference between the two. We didn't test this game for performance, since it's very system limited, but from experience, you can expect a frame rate of around 50 - 60 fps in this game with an Athlon 3500+.

Homeworld 2

GeForce 7800GTX



GeForce 7900GTX



These static screenshots are pretty much the same. Performance wise, the 84.21's results are faster, but the differences are negligible. After all, we're already hitting very high frame rates. This game is even more system limited than Call of Duty, so even AA and AF scores got a slight boost. We skipped the aliasing in motion captures, since we're seeing the same thing on btoh cards.

SW: KOTOR

GeForce 7800GTX



GeForce 7900GTX



First of, sorry for not getting the angle perfectly right, but as you can see the image quality (aliasing and filtering wise) is the same from both cards. So, it looks like there's nothing new from the newer drivers. The performance increase is valid, at least as far as we can tell from with the games in our benchmark suite.

Transparency Antialiasing

GeForce 7800GTX



GeForce 7900GTX



Hmm, what do we have here? There are definite differences between the 7800GTX and 7900GTX screenshots, with the 7900GTX not as effective at removing aliasing as the 7800GTX. Ouch, could this be true? Quality wise, transparency antialiasing have actually gone down with the GeForce 7900GTX. If you look at the screenshots from the X1900XTX (which you can see here), these GeForce 7900GTX screenshots actually have more aliasing than ATI's adaptive antialiasing at 4x. Hopefully this is only a driver bug, but this is still very dissapointing. Let's take a look at Serious Sam II, just to make sure its not application specific.

GeForce 7800GTX



GeForce 7900GTX



Ah, looks like it is application specific. The screenshots here are pretty much the same, transparency antialiasing wise. It's pretty odd, but we think it's probably related to how each application handle transparent textures. It certainly will be interesting to see what kind of performance the GeForce 7900GTX can offer when running with 4x transparency antialiasing enabled.

High Dynamic Range Rendering

GeForce 7800GTX



GeForce 7900GTX



Last but not least, HDR rendering. As you can see, very little has changed. There is a difference between the screenshots (the light reflecting of Sam's left shoulder), but that's probably nothing and you certainly won't notice it in gameplay. What about Serious Sam II?

GeForce 7800GTX



GeForce 7900GTX



There is a small difference between the two (look at the waterfall), but again that's negligible. There seems to be a slight difference in HDR between the GeForce 7800GTX and 7900GTX, so maybe NVIDIA did 'cut some corners' in this regard. However, the differences are minimal and you most likely won't notice the differences in actual gameplay.

With these new drivers (and the new card), you can expect pretty much the same image quality you can expect from NVIDIA's cards. There are some issues, but hopefully NVIDIA will be able to iron them out. Just goes to show you to always be on the look out for image quality changes each time a driver comes out.

Performance

We'll be focusing on performance under default and 4x AA with 16x AF. After that, we'll take a quick look at transparency anti aliasing and HDR performance. We didn't test 8xS performance, since this mode of antialiasing doesn't always work well with all games in our benchmark (for example with soft shadows in SW: KOTOR). To put these scores into perspective, we've included results from both a standard clocked and a higher clocked GeForce 7800GTX (the ASUS EN7800GTX TOP), running at the same settings.

We'd like to thank both Tagan and Kingston for supplying the additional power supply and 1 GB memory modules for this article. We'd also like to thank Leadtek for providing the GeForce 7900GTX sample card.

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
ATI Catalyst 6.2 reference driver
NVIDIA Forceware 84.21reference 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.These results are in frame rate per second.

Standard settings

The results:

If you look at the numbers, you'll see that the GeForce 7900GTX is generally slower than the GeForce 7800GTX when running without AA and AF. It's quite odd. If we were system limited, the results should be at least very close, not about 10 percent slower on the higher clocked card. We believe this is a driver problem, since other games are affected as well. It's not until we hit 1280 x 1024 and above with AA and AF do we see the GeForce 7900GTX taking a small lead. It's actually more effective than the GeForce 7800GTX, with 'only' a 40 percent frame drop compared to the 50 percent of the GeForce 7800GTX. We'll be looking at efficiency later on, but for now this looks 'interesting'.

The core clock speed difference also helps the GeForce 7900GTX maintain an 80 fps minimum in this game, whereas the GeForce 7800GTX dipped slightly below that. In real life, that means you probably be seeing 80 fps all throughout gameplay - Call of Duty is capped at 80 fps by default.

	GeForce 7800GTX		GeForce 7900GTX
	Default	4x AA 16x AF	Default	4x AA 16x AF
<30 fps	0	0	0	0
30-45 fps	0	0	0	0
45-60 fps	0	0	0	0
60-90 fps	5	35	5	5
90-120 fps	4	39	8	43
>120 fps	67	2	63	28
Total	76	76	76	76

results are in seconds

In case you're wondering, the table above is actually a brokedown of the benchmark results, displaying the time spent throughout the replay organized into several different frame rates ranges. Because this is a replay and not a timedemo, the replay runtime is the same between all cards. With a timedemo, the total number of seconds will vary according to settings, ie a heavier setting will have a higher number of seconds because the benchmarks takes longer to finish.

Now look at the 60 - 90 fps range, where we most likely will experience the minimum fps. You can see that you'll spend quite some time here with the GeForce 7800GTX, but not so with the GeForce 7900GTX. Remember, this is at 1600 x 1200. So, you'll spend most of your time above 90 fps with the GeForce 7900GTX. That's extremely fast. As you can see, the GeForce 7900GTX is only worth it if you want to play at 1600 x 1200 with AA and AF, any less you could 'make do' with the GeForce 7800GTX.


You might wonder what's the point in benchmarking these old games - just look at the frame rates - who needs 400 fps? However, since these games are texture dependent, they're perfect for testing AA and AF performance. No shaders to complicate things. It's only at 1280 x 960 and above do we again see the GeForce 7900GTX leaving the GeForce 7800GTX. Just like in Call of Duty, AA and AF penalty is smaller on the GeForce 7900GTX - around 25 percent at 1600 x 1200, compared to 40 - 50 percent on the GeForce 7800GTX. With a minimum fps of 47 fps, you can play at 1600 x 1200 with AA and AF and experience no lag whatsoever. You might even be able to still have fairly fluid game at an even higher resolution, but Homeworld tops out at 1600 x 1200.

The game looks just great at that resolution with AA and AF. If only Relic fix the color banding issue and add bump mapped textures. Oh well, one can dream.

	GeForce 7800GTX		GeForce 7900GTX
	Default	4x AA 16x AF	Default	4x AA 16x AF
<30 fps	0	0	0	0
30-45 fps	0	0	0	0
45-60 fps	0	0	0	0
60-90 fps	0	1	0	0
90-120 fps	1	11	1	1
>120 fps	69	58	69	56
Total	70	70	70	70

results are in seconds

We could all see we're system limited here - frame rates between the cards under different settings are not all that different. Of course, that's not the whole story. The table can't really capture what we're seeing on the graph. If you're wondering, that's because the performance difference between the GeForce 7800GTX and 7900GTX can only be 'seen' above 120 fps. The GeForce 7900GTX is so fast it's ridiculous - not many people can tell if the difference between 120 and 150 fps or more. And that's at 1600 x 1200, mind you.


There's a quite interesting fact about this game - on NVIDIA's cards, choosing a camera angle with the car in view seems very detrimental to frame rate. We always chose the front bumper view for replay and it's not until we ran this test on the Radeon X1900 XTX did we see the difference. Back to the evaluation: the GeForce 7800GTX is (obviously) slower that its successor - the GeForce 7900GTX is hardly breaking a sweat at 1600 x 1200 with 4x AA and 16x AF enabled. Of course that may change once we enable transparency antialiasing.

If you look at the numbers, you'll see the GeForce 7800GTX losing about 30 percent of its performance with AA and AF on higher resolution. In contrast, the GeForce 7900GTX only lost about 15 percent or about half of it's sibling, Even the minimum frame rates are absurdly high - 90 fps. Running at 1600 x 1200 with AA and AF, you'll most likely never notice the difference in minimum frame rates from 1024 x 768 - a mere 10 fps drop.

	GeForce 7800GTX		GeForce 7900GTX
	Default	4x AA 16x AF	Default	4x AA 16x AF
<30 fps	0	0	0	0
30-45 fps	0	0	0	0
45-60 fps	0	0	0	0
60-90 fps	0	141	0	0
90-120 fps	21	117	41	112
>120 fps	240	3	220	149
Total	261	261	261	259

results are in seconds

A 90 fps minimum at 1600 x 1200 with AA and AF. The GeForce 7800GTX can't really put up a fight at this resolution. We can see that for the most part, the GeForce 7900GTX stays fairly close to the frame rates under default settings. That means we're system limited most of the time. The first split of Harwood Forest is still quite heavy though. Again,it's obvious we only see the true worth of the GeForce 7900GTX at high resolutions, at least in old games like these.


Even the mighty must fall. Smoke and explosions can still bring down the mighty GeForce 7900GTX to its knees. The GeForce 7900GTX is still not fast enough for you to run this game above 1280 x 1024 with AA and AF all the time. Hell, look at the minimum frame rates without AA and AF, it's already below 30 fps. So, the tendency to lose frame rate with explosions and debris we've seen on NVIDIA's cards is still present on the GeForce 7900GTX. Although there are improvements in performance, its mostly come from the clock speed difference alone.

We wouldn't be aware of this fact if we were only looking at average frame rates. Instead of just relying on subjective evaluation, you can see for yourself the minimum frame rates are just too low. Thankfully, smoke and explosion don't happen all the time throughout gameplay. Funny, it's not until now did we notice the drop between average and minimum frame rates between AA and AF results with non AA and AF is similar.

	GeForce 7800GTX		GeForce 7900GTX
	Default	4x AA 16x AF	Default	4x AA 16x AF
<30 fps	1	3	0	2
30-45 fps	3	4	3	4
45-60 fps	3	11	2	4
60-90 fps	10	8	13	9
90-120 fps	6	0	5	2
>120 fps	0	0	0	0
Total	23	26	23	21

results are in seconds

Like we said earlier, 1600 x 1200 with AA and AF in this game is still too heavy, even for the GeForce 7900GTX. Look at the sharp drop - that's when the screen is literally filled with smoke.Even the GeForce 7900GTX took quite a hit here. It's worse for the GeForce 7800GTX. It doesn't stand a change and quickly falls to below 30 fps each time there's an explosion and smoke.


We originally thought the problem with Call of Duty is limited to OpenGL games, but it looks like we're wrong. It's not until we hit 1280 x 1024 and above do we see the GeForce 7900GTX flexing its muscles. With a GeForce 7800GTX, you probably don't want to run this game above 1280 x 1024 with AA and AF. Even a slightly higher clocked GeForce 7800GTX is barely supplying high enough fps. That's not true with the GeForce 7900GTX. Once again it prove it's worth - you can definitely enjoy lag free gameplay at the highest resolution with AA and AF (just don't expect the same at even higher resolutions).

The GeForce 7800GTX lost around 10 - 30 percent performance across the board with AA and AF enabled, Meanwhile, the GeForce 7900GTX is slightly better: losing about 10 - 25 percent, but that hit isn't really noticeable below 1280 x 1024. This is different from what we've seen so far from texture dependent games. Brothers in Arms is quite new and it's not a purely texture dependent game, it make some use of shaders as well.

	GeForce 7800GTX		GeForce 7900GTX
	Default	4x AA 16x AF	Default	4x AA 16x AF
<30 fps	0	0	0	0
30-45 fps	0	0	0	0
45-60 fps	0	4	0	0
60-90 fps	24	77	14	46
90-120 fps	61	23	69	36
>120 fps	49	30	51	52
Total	134	134	134	134

results are in seconds

Although the table may look similar, we can tell from the graph that the GeForce 7800GTX can still drop quite a lot when explosions fills the screen (the two sharp drops). If you look at the GeForce 7900GTX graph, you'll see no such drops. That means frame rates stay consistent throughout the rest of the gameplay. At higher resolutions, the picture changed a bit, but the GeForce 7900GTX can still manage just fine at 1600 x 1200. Remember though, that sharp drop may not be noticeable at all - on both occasion the frame rates is still fairly high at around 45 fps at 1024 x 768. So, at 1600 x 1200, it should be around 35 - 40 fps (the minimum fps is 37 fps for the GeForce 7900GTX).