Rage3D nVIDIA GeForce 7800 GTX Review Rage3D nVIDIA GeForce 7800 GTX Review
Normally the length of time between product announcements from both of these companies can be counted in days or weeks; one would release a new card and, in response, the other would do the same (whether they were ready or not). However, nVidia changed the rules of engagement with the 7800 GTX release last month that I think caught a lot of people by surprise. Instead of the regular paper-launch, NVIDIA had immediate availability of the 7800 GTX on the same day the card was announced. That means that the card was actually on store shelves and available to order online that exact same day.
Clearly ATI wasn't ready for this new tactic. Their R520 wasn't ready to anywhere near that degree, and they obviously couldn't respond to a hard-launch with a paper-launch. So while ATI fights to finish the R520 and get a sufficient supply on hand for a hard-launch of their own, nVidia enjoys zero competition in an industry that thrives on marketing and mindshare.
While NVIDIA would probably like the world to think that the G70 is a completely new chip, it is in reality not much different than the NV40 that came before it. That's not necessarily a bad thing however, the NV40 is a very effective design and already supported SM3.0 so there was really not much need to change things a whole lot, especially when you consider how far ahead of the curve they are with SM3.0 offerings.
The primary difference between the G70 and the NV40 is the pixel pipeline configuration. While the basic layout remains the same, NVIDIA managed to squeeze in 24 "pipelines" as opposed to 16 on the NV40. Along with the pixel pipeline changes, NVIDIA also added 2 more vertex shader units to the G70 for a total of 8.
| Specifications | |||||
| GeForce 7800 GTX | Radeon X850XT-PE | ||||
| Core | G70 | R480 | |||
| Silicon Process | 110nm | 130nm low-k | |||
| Transistor Count (millions) |
302 | 160 | |||
| Core Speed MHz | 430 | 540 | |||
| Memory Speed MHz | 600 GDDR3 (1.2GHz) |
590 GDDR3 (1.18GHz) |
|||
| Bus Standard | PEG x16 | PEG x16 | |||
| Bus Width | 256bit | 256bit | |||
| Pipeline Configuration Textures/Pixels/Z Samples (Per Clock) |
24/16/32 | 16/16/16 | |||
| Peak Memory Bandwidth (GB/s) |
38.4 | 37.8 | |||
| Pixel Fillrate (million pixels/sec) |
6,880 | 8,640 | |||
| Texel Fillrate (million texels/sec) |
10,320 | 8,640 | |||
| API Compliancy | DX 9.0c Shader Model 3.0 |
DX 9.0b Shader Model 2.0 |
|||
| MSRP ($US) | $599 | $549 | |||
NVIDIA® CineFX™ 4.0 Shading Architecture
64-Bit Texture Filtering and Blending
NVIDIA® Intellisample™ 4.0 Technology
NVIDIA® UltraShadow™ II Technology
Composited Desktop Hardware Engine
Advanced Display Functionality
Advanced Engineering
NVIDIA® Digital Vibrance Control™ (DVC) 3.0 Technology
Operating Systems
API Support
Batch files were used when possible for automated benchmarking (the details of the commands used are outlined for each test). When manual benchmarking was necessary FRAPS 2.6.2 was used.
Benchmarking was done with Windows set to the "Adjust for best performance" profile, and all unnecessary Windows services and hardware devices were disabled. The latest drivers for each necessary hardware component were installed prior to testing and kept consistent throughout.
The Onboard sound was disabled for all tests except for Richard Burns Rally which will not load without sound hardware installed and operational.
Test System Specs
Benchmark Software
Since ATI introduced the R300 with its Rotated-Grid Gamma Corrected Multi-Sample AA, it was felt by many that NVIDIA was behind the curve a bit when it came to Anti-Aliasing quality. The Ordered Grid sample pattern and lack of gamma correction made the argument against NVIDIA fairly easy.
With the 7800 GTX however NVIDIA has taken a huge step forward with their Anti-Aliasing techniques and have improved it to the point where it's easily on par with ATI's. NVIDIA now uses a Rotated Grid sample pattern, with gamma correction, and a new option called Transparency AA that is designed to clean up the jaggies unaffected by traditional multi-sample AA. Not only that, but with additional mixed modes and "hidden" settings that can easily be revealed with tweak programs such as RivaTuner and nVHardPage, it could even be said that NVIDIA's Anti-Aliasing quality has passed ATI altogether.
To illustrate the point, I've created a simple JavaScript based page that will let you see, up close, the anti-alias quality of both NVIDIA's and ATI's cards in Half-Life 2. Click the link below to open the page; it'll appear in a new window.
Anti-Aliasing Quality in Half-Life 2
Transparency AAYou might have noticed that in some games there are certain elements that are often rendered using a simple 2D partially transparent texture instead of actual game geometry. Things like fences, grates, vegetation, and other types of architecture that are normally too complex or abundant to render with normal geometry are often rendered using these types of textures. The advantage is developers can use a lot of them to increase the apparent detail of a world without worrying about performance too much. The disadvantage is that these elements are often riddled with jaggies that traditional methods of Multi-Sample Anti-Aliasing can't cope with (Super-Sample Anti-Aliasing can cope with it just fine, but even on today's ultra-fast graphics cards SSAA remains impractical for the most part).
One of the coolest new features the 7800 GTX brings to the table is a form of Anti-Aliasing called "Transparency AA" which effectively solves the problem of aliasing on these types of 2D textures. There are two types of Transparency AA, multi-sampling and super-sampling. Multi-sampling is faster but the image quality isn't as good while super-sampling looks good but runs (slightly) slower.
Clicking the link below will open up a new window which will let you easily see the effect Transparency AA has on a chain-link fence in Half-Life 2 that is made using an alpha-transparent texture.
Transparency Anti-Aliasing in Half-Life 2
Nothing is free, and enabling Transparency AA (it is off by default) will naturally affect performance somewhat. To see how much of an affect TAA has on overall performance I ran my standard Half-Life 2 benchmark test with and without TAA enabled. The benchmark test runs five different levels from Half-Life 2, each representing a fairly significant portion of the game level and each with varying degrees of alpha-transparent texture usage. Each level was run three times and the individual scores were then averaged to get the final score for each resolution and setting.
One technology feature which NVIDIA has been missing these last few years has been Gamma Corrected AA. ATI introduced gamma Corrected Anti-Aliasing technology with their R300 based Radeon 9700 Pro back in the summer of 2002. Since then people have been wondering where NVIDIA was with a similar technology.
Gamma Corrected AA improves the blending of AA edges when using Multi-Sample AA. It can have a positive impact on the overall anti-aliasing quality of the image and is particularly noticeable on very thin architecture such as the power lines you often see in games like Half-Life 2.
Like TAA, Gamma Correction is disabled by default. However, the impact on performance is so small that it's a bit strange why NVIDIA doesn't simply enable it by default.
When looking at pure edge anti-aliasing I think ATI's 6x is still the best available, but both NVIDIA's and ATI's edge anti-aliasing are now very good and pretty much on par with each other when comparing 4xAA to 4xAA. However, overall I would have to give the nod to NVIDIA thanks to Transparency AA, the 8xS mixed mode, and the extra options that can be exposed with various third party tweak tools.
Texture filtering is probably more demanding than Anti-Aliasing when it's done by the book, which is probably why both nVidia and ATI employ adaptive filtering and take shortcuts wherever possible. Anisotropic filtering in particular has a huge impact on performance when it's done without any of those adaptive tricks and shortcuts.
To see how well the 7800 GTX handles anisotropic texture filtering, I loaded up Chronicles of Riddick on both cards and took some screenshots to compare filtering quality. Click the link below to open a webpage which will let you easily compare the filtering quality.
Anisotropic Filtering Comparison in Chronicles of Riddick
Both cards are fairly close when it comes to texture filtering as you can see by looking at the screenshots alone. I would have to say that nVidia has a slight advantage as it appears there are a few areas where the textures are sharper and more detailed however. Ingame though I can definitely see a mip boundary crawling along ahead of my player movements when using the 7800 GTX which is not apparent on the X850XT. This is, of course, known to a lot of people as nVidia's "brilinear" filtering (sort of halfway between bilinear and full-on trilinear filtering).
"Brilinear" filtering is one of those shortcuts I mentioned that nVidia uses to increase performance. It's "On" by default but can be turned off by changing the Image Quality setting from Quality to High Quality in nVidia's control panel (which disables "Trilinear Filter Optimization", "Anisotropic Mip Filter Optimization", and "Anisotropic Optimization"). The link below leads to a page with screenshots of 3DMark05's "Filtering and AA" tool. It should give you an idea of what Brilinear does (each shot was taken with 16x AF forced on via the graphics card control panel).
Anisotropic Filtering Pattern Comparison
All the benchmarking in this review was done with default settings (ie Quality as opposed to High Quality), so to get an idea of the performance impact you could expect to see if you want to get rid of that slight mip boundary creeping along ahead of you in your games, I ran some Quality vs High Quality benchmarks with Doom 3. The results are below.
In straight up screenshot comparisons NVIDIA looks to have better filtering quality in both Quality and High Quality modes when compared to ATI screenshots. However, screenshots certainly aren't a good way to compare filtering quality, and in actual gameplay the Brilinear line which you get with NVIDIA's Quality setting is noticeable as you move around.
With the High Quality setting enabled (optimizations off) I think NVIDIA's texture filtering quality is overall better than that which ATI offers, but because NVIDIA chose Quality as the default mode instead of High Quality, I have to give the nod to ATI in the texture filtering department.
Battlefield 2
Unlike the other tests I've run here, I only tested with and without 4x Anti-Aliasing (I also include Anisotropic results for the other benchmarks). In the future I hope to get more experience with BF2 benchmarking so that I can include AF results as well, but for the limited time I have with the card these results will do (as a side, I believe "Texture Filtering" when set to "High", as it was during my benchmarks, is equivalent to 4x Anisotropic Filtering. I have yet to confirm that however).
Benchmarking was run manually and scores were logged with Fraps. The results are below. Click the text links at the top of the chart to change settings.
Splinter Cell Chaos Theory
The Splinter Cell Chaos theory results are below. Click the text links at the top of the chart to change settings.
Note that with 4xAA at 2048x1536 the X850XT-PE would crash when the game tried to launch which prevented me from getting those results.
Half-Life 2
A batch file was used to automate testing; the command line is below for reference. This batch file was used for each card that was tested. The settings surrounded by < >quick change for each pass:
The Half-Life 2 results are below. Click the text links at the top of the chart to change settings.
Doom 3
Another batch file was used to automate Doom3 testing as well. The command line is below for reference. This batch file was used for each card that was tested. The settings surrounded by < > change for each pass:
The Doom3 results are below. Click the text links at the top of the chart to change settings.
Chronicles of Riddick: Escape from Butchers Bay
Chronicles of Riddick was using another custom batch file that ran each of the 5 built in timedemos that are included with the 1.01 patch. The Shader Model 2.0 path was used for all cards, including the 7800 GTX, and Anti-aliasing and Anisotropy were set in the graphics card control panel.
The Chronicles of Riddick results are below. Click the text links at the top of the chart to change settings.
Far Cry
Again a custom batch file was used to benchmark Far Cry. I used 3 demos included with the newer patches to test performance (from the Cooler, Training, and Volcano levels). Because Far Cry benchmark framerates can vary between each subsequent pass (sometimes fairly significantly), I ran each of the 3 demos 3 times, then averaged the 9 results to get the final score for the detail level.
The command line for the batch file I used to automate Far Cry benchmarking is below. This batch file was used for each card that was tested. The settings surrounded by < & > change for each pass:
The Far Cry results are below. Click the text links at the top of the chart to change settings.
Pacific Fighters
I benchmarked Pacific Fighters by loading the included "N1K1 vs BeauFighter.ntrk" track and logged framerates using Fraps from the beginning of the track for 90 seconds. All the in game details were set to their maximum levels, including "Landscape Detail" which was set to "Perfect", enabling Pixel Shaded water. Video was set to the "NVIDIA GeForce 6800/6600/FX/4/3" profile for the 7800 GTX and the "ATI Radeon X800/9800/9700/9600/9500" profile for the X850XT-PE. Anti-aliasing and Anisotropy were set via the graphics card control panel.
The Pacific Fighters results are below. Click the text links at the top of the chart to change settings.
Richard Burn’s Rally
I used Fraps to benchmark Richard Burn's Rally using two custom demos, one recorded on the Great Britain Chirdonhead II stage with rain and the other on the USA Frazier Wells track. I started Fraps framerate logging as soon as the marshal launched the car and had it automatically stop logging after 90 seconds. I logged each pass 3 times then averaged the results to get the final score for that setting and resolution. The default replay camera was used and sound was enabled. Anti-aliasing and Anisotropy were set via the graphics card control panel.
The Richard Burn's Rally results are below. Click the text links at the top of the chart to change settings.
3DMark03
I benchmarked 3DMark03 using the default settings. I set it up to loop each test three times, it then automatically averaged the results and gave a final score. The results for each test are below.
| 3DMark03 Wings of Fury |
||||||
|
||||||
3DMark05
Like 3DMark03, I set 3DMark05 up to loop three times to get the final score. I tested 3DMark05 using the default detected settings.
| 3DMark05 Return to Proxycon |
||||||
|
||||||
Overclocking the card was pretty easy using NVIDIA's built-in overclocking tool. I let it automatically detect the best overclock, and then manually adjusted it as high as possible. The end result was 522MHz core and 1.34GHz memory, excellent overclocking for such a large chip. Performance results from Doom3 are below.
| [ No AA / No AF ] [ No AA / 16x AF ] [ 4x AA / No AF ] [ 4x AA / 16x AF ] |
| 1024x768 | 1280x960 | 1600x1200 | 1920x1080 | 2048x1536 | |
| Overclock % Faster vs Stock |
0.0% | 0.0% | +3.6% | +4.5% | +11.0% |
Well that's it, the first NVIDIA review on Rage3D and the first time I've ever gone hands-on with a top-end Geforce.
What do I think of it? Clearly the performance is nothing short of amazing. I can't believe that at 2048x1536 with 4xAA the card can still pull along nicely, even with the newest game engines (and in a nod to playability at that level of detail, my CS:Source profile is now set to 2048x1536/4xAA/16xAF details max). And this is with a single card too; I can't imagine how crazy fast things would be with an SLI setup.
The new Transparency AA is pretty cool too. It really does make a significant impact on quite a few games, and the impact it has on performance is fairly negligible (at least with the games I tested). The same goes for the new gamma correction, it too has an insignificant impact on performance... which honestly makes me wonder why NVIDIA has these new features disabled by default. It might be that they cause a few problems in some cases, but I'm of the opinion that it's better to have to turn it off if there are problems than to have to turn it on to enjoy the benefits.
The primary concern now is the price. At $599 it's the most expensive videocard on the market, $50 more than ATI's flagship X850XT Platinum. The price is out of reach for a lot of users obviously, but the real kicker here is that many people have been suggesting that NVIDIA has an even faster part in store which they will reveal depending on how well ATI's yet to be released R520 stacks up. Should the R520 be faster and NVIDIA releases the 7800 Ultra (for lack of a real name), then those early adopters might feel a bit stung knowing that their 2-3 month old top of the line card got replaced so quickly.
Having said that, it looks like the 7800 GTX street price seems to be falling rather quickly. I've seen one or two from some reputable AIBs that are even close to street prices for the X850XT-PE. Certainly it's still out of range for most, but the aforementioned risk of buying a 7800 GTX can be a bit easier to swallow should you find the right deal.
Compared directly to the X850XT Platinum it's easy to be impressed with the 7800 GTX. Not only with the outstanding performance of the part, but also with the features it offers. With games already out to support Shader Model 3.0 and with more sure to come very soon, the oft denounced technology is now a truly viable feature that gamers should be looking for in a new graphics card. Couple that with SLI, which has been welcomed into the marketplace with open arms despite plenty of doubts cast on its practicality, and it's all but impossible to recommend anything but the 7800 GTX when asked for high-end videocard advice.
Before I wrap this up, I feel it necessary to point out the fantastic job NVIDIA did with the launch of the 7800 GTX. Every single graphics card release prior to the 7800 GTX was a "paper release". That more or less means that a product was announced but actual availability was still some ways off (a single month was considered very good). On the very day that the 7800 GTX was announced NVIDIA had cards on store shelves. Not ready to ship, not on the way to stores, actual product right on the shelf. That, right there, is how things should be done. With the 7800 GTX launch NVIDIA set a precedent that other companies must follow.
In short, if you're looking for a high-end videocard right now, the only choice you have is the 7800 GTX.
