Rage3D GeForce 8-Series Review Rage3D GeForce 8-Series Review
You’ve seen the enormity of the smack-down laid upon the previous generations of graphics hardware in both performance and image quality, and by now some of you lucky folks have even got one or, if you are a particularly devout gamer, two of them sitting in your system right now. Those who do know firsthand the qualities of this chip, those who don’t wish they were those who do.
You know all about its Unified Shader Architecture and its DX10 capability. You know it has some pretty cool features, some very much needed improvements to the GeForce image quality reputation and, most obvious, you also know that it’s fast. Fucking fast.
Yes, I swore. I feel bad. However, to properly project the idea of how fast the G80 is I need to color things a bit. Grab your attention. Take you by the shoulders, as it were, and shake you till you get the enormity of what I’m saying here. Others would tell you that it’s “Blazing Fast” or “Fast as Lighting” or, maybe, in attempt to go colorful themselves, would say something naughty like “It’s Fast as Hell!”. I feel those types of literary turns of wit do not convey the idea properly, however, so bear witness to my words, fine reader, and take them to heart: it is fucking fast. You’ll see plenty of evidence of exactly how fast the thing in throughout this review, I do think.
Don’t tell my Mom I cussed though.
Regardless of your foreknowledge of all things G80 I will restate this simple fact of recent history here: on November the 8th NVIDIA released the G80 GPU, what is arguably the biggest advancement in graphics technology since Ernie Coombs invented the pixel in his tickle trunk during his spare time. It does away with so many of the standard conventions of GPU design that the G80 can almost be considered a whole new approach to graphics hardware. It’s akin to the technological leap from piston powered engines to jet engines. Faster, more efficient, and immediately obsoletes everything before it.
Of course, you don’t “release” a GPU and expect people to buy it (though I’m sure some would!). NVIDIA knows this, which is why they’ve cleverly put the G80 on a PCB and made two new SKU’s that they’ve ingeniously called the GeForce 8800 GTX and GeForce 8800 GTS. Good stuff, Mr. Dressup would approve.
Here’s what NVIDIA’s lineup looks like today:
| NVIDIA's Lineup | ||
| Enthusiast | 8800 GTX | $599 |
| 8800 GTS | $449 | |
| Performance | 7950 GT | $299 |
| Mainstream | 7900 GS | $199 |
| 7600 GT | $159 | |
| Value | 7600 GS | $129 |
| 7300 | <$99 | |
As you can see the 8800 GTX and the 8800 GTS replace the 7950 GX2 and 7900 GTX, respectively, in NVIDIA’s enthusiast class segment. The 7-series cards from the 7950 GT down remain the same for now, but there are rumors that NVIDIA is already hard at work creating GPUs and cards for a complete top-to-bottom 8-series lineup.
The GeForce 8800 GTS, being the more affordable of the two, has less overall performance than the GTX but is no less impressive in the grand scheme of things. It has its G80 GPU clocked at 500MHz, 20 ROPs, and uses an equally upsetting 320bit wide memory bus connected to 640MB of GDDR3 running at 800MHz (1.6GHz).
At $449 the GTS is priced right in line with the other enthusiast class cards these days, but at $599 the 8800 GTX will set you back quite a bit more. Is the extra power worth the extra money? You’ll have to decide that one for yourself (we’ll help by providing you with some nice benchmarks to mull over).
With no response in sight, AMD’s graphics division (which we’ll call “ATI” in this article, for sake of convenience) finds themselves in the disappointingly familiar position of having to play catch-up once again. Something they should be quite accustomed to by now, unfortunately, having being given so much practice in that position these past couple years.Looking at the G80 Architecture diagram below you’ll see no mention of things you’d normally expect to find in abundance on a typical GPU diagram like pixel or vertex shaders. What you see instead are eight clusters of little green and blue squares that represent the innards of the thread processing clusters. Each green square, 16 per cluster for 128 in total, represent what NVIDIA calls a “Stream Processor”, which is a generalized floating-point unit that can operate on pixel, vertex, geometry, or even physics operations.
These stream processors, which unlike most of the rest of the chip run at an astounding 1.35GHz, are at the heart of the first DX10 graphics card on the market and the first to feature a fully unified shader architecture.
mouseover this image for a closer look at the thread processing cluster
As much as it feels like it, the concept of a unified shader isn’t new. In fact, Microsoft’s XBOX 360 gaming consoles has a graphics core that was designed by ATI some years ago which is based on the principals of the unified shader.
Shader unification is all about efficiency. In a traditional GPU, one with discrete pixel and vertex shader units, there exists a real possibility that for any given scene either the pixel shader engine or the vertex shader engine could go underutilized, leaving hardware idle and affecting potential performance.
In a scene with a lot of pixel shader effects and few vertex shader effects, for example, the pixel shader units are working at or near 100% of capacity while the vertex hardware might be doing very little. In a vertex heavy scene, the opposite would be true. Idle hardware is wasted hardware, which is why chip engineers from all fields work to make sure hardware is working as close to 100% capacity as possible. With discrete pixel and vertex shader units and the way games are designed it’s very difficult to make a GPU work as efficiently as possible.
With an unified shader architecture the GPU can perform vertex or pixel calculations in the same units. If a scene has a lot of pixel shader effects and few vertex shader effects then most of the units would be used for the pixel shader effects while only a few would be used for the vertex shader effects. If there are a lot of vertex effects, then the opposite is true. It is flexibility here that makes a unified architecture efficient, and improvements to efficiency lead to improvements in overall performance. It’s like a spork, and we all know how much a good spork kicks ass.
With NVIDIA’s G80 and ATI’s upcoming R600 GPU, it seems that both companies have decided that the unified shader architecture approach is the way forward.
Texturing Powerhouse
In the cluster diagram to the right you can see both the "stream processors" (scalar arithmetic units, or ALUs) and the texture units. Every cluster has a scheduler which is coupled to 16 stream processors (one ALU block), 16 interpolation units (which will not be further discussed here) and 4 texture address units.
These addressing units each have two texture filtering units at their disposal, unlike in earlier NVIDIA and ATI designs where each addressing unit was coupled to a single filtering unit.
What this means is that simple operations, such as bilinear filtering, operate as if there were 32 TMUs on the chip. More complex filtering however such as trilinear, anisotropic, and FP16 filtering will operate as if there were 64 TMUs.
That's 18.4 billion filtered textured fragments per second for the 8800 GTX with either bilinear or more advanced filtering (unlike the stream processors, the texturing units operate at the core speed).
In practice, when selecting High Quality rendering in the control panel (as all tests throughout this review were conducted), all the pixels on the screen are filtered with at least trilinear filtering, and possibly with some extra anisotropic filtering, so performance is very similar to a graphics card with 64 TMUs.
Overall this is an interesting design choice. Even without its doubled texture filtering ratio (which allows for the nearly free anisotropic filtering), the G80 would still have substantially more texturing power per clock than anything before it.
Also, unlike previous NVIDIA GPUs, the texturing units are decoupled from the arithmetic units. This results in higher efficiency with advanced filtering, as the arithmetic units will not stall while waiting for the texture units to finish. ATI’s R5x0 family also has decoupled texture units, which certainly helped their 16 TMUs compete against NVIDIA's G7x 24 TMUs. Just like on the R580, the texture units on the G80 remain hardwired to specific ALU blocks.
ROPs and Memory Subsystem
The 8800 GTX has six Raster Operation Partitions, or ROPs, and each of those can process 4 pixels per clock for a total of 24 pixels/clock with color and Z processing (the 8800 GTS has one of the ROP partitions disabled, resulting in a total of 20 pixels/clock). For Z-only processing NVIDIA has implemented a new technique that allows for an amazing 192 samples/clock to be processed when a single sample is used per pixel (check the chart below for Z-only results. G80, putting the ZOMG in Z-only).
The G80 ROPs support multisampled, supersampled, and transparency adaptive antialiasing with four new antialiasing mode; 8x, 8xQ, 16x, and 16xQ (more on those later), along with the normal 2x and 4x AA modes we’ve all grown accustomed to. They also support blending of FP16 and FP32 render targets, finally allowing HDR+AA on NVIDIA hardware (maybe now that NVIDIA supports it we’ll actually see it used in games!).
Each of the six memory partitions on the 8800 GTX GPU provides a 64bit interface to memory giving us a combined 384bit interface width. On the 8800 GTS one of the memory partitions have been removed resulting in an 320bit interface width. A high-speed crossbar, similar to the one on NVIDIA’s previous 7-series GPUs, supports DDR1, DDR2, DDR3, GDDR3, and GDDR4 memory types on both SKUs. The 8800 GTX uses GDDR3 clocked at 900MHz (1.8GHz), providing 86.4 GB/s of raw memory bandwidth. The 8800 GTS also uses GDDR3, but has it clocked at 800MHz (1.6GHz) providing 64.0 GB/s of bandwidth (the exact same bandwidth as ATI’s X1950 XTX, incidentally).The 8800 GTX is one massive piece of hardware. At 10.5 inches long it’s the longest graphics card ever produced by either IHV (ATI’s AIW X1800 XL was the previous record holder at 10 inches). So long in fact that on a standard ATX motherboard it extends over the edge by about an inch. This might obviously cause some problems in quite a few PC cases out there, so to help minimize potential issues NVIDIA moved the dual-PEG connectors from the back of the card to the top edge. The 8800 GTS in contrast has just the one PEG connector in the traditional location, and is about the same length of the X1950 XTX which shouldn’t cause any problems in most all enclosures.
 BFG 8800 GTX and XFX 8800 GTS |
 BFG 8800 GTX and ATI X1950 XTX |
 BFG 8800 GTX and BFG 7900 GTX |
 BFG 8800 GTX and NVIDIA 7950 GX2 |
 XFX 8800 GTS and ATI X1950 XTX |
 XFX 8800 GTS and BFG 7900 GTX |
 XFX 8800 GTS and NVIDIA 7950 GX2 |
 8800 GTX, 8800 GTS, 7950 GX2, and 7900 GTX |
 The 8800 GTX extends over the back of ATX motherboards |
 BFG 8800 GTX Dual PEG Connectors on the top edge of the card |
 XFX 8800 GTS single PEG Connector on the back end of the card |
The GTX and the GTS share the same cooling unit, but unlike the 7900 GTX cooler that pushes air in both directions from a fan mounted at the center of cooler, the 8800 cooler is closer in design to the one ATI uses on their X1950 XTX. A blower style fan located towards one end of the cooler forces air through a plastic housing over an array of cooling fins, and then most of the heated air is sent on out the back of the case. Strangely enough there are openings in the plastic shroud just before the end of the card that lets some air escape into the case. I’m not sure why NVIDIA designed it this way, but there they are in the second pic below. The fan itself is extremely quiet, even spun up under extended load.
 BFG 8800 GTX Cooling Unit. The XFX 8800 GTS cooler is the same design |
 Auxilliary air vents on the cooler? We're not sure why these are there. |
Unlike the all copper X1950 XTX cooler though the 8800 cooler only uses copper on the bit of the cooler that contacts the GPU. The base of the cooler and the fins are aluminum. This has the advantage of lowering the overall weight of the card (both the 8800 GTX and the X1950 XTX are about the same weight, incidentally) at the expense of overall cooling capability. To compensate, I guess, there are two heatpipes to help move heat around. The biggest transfers heat from the GPU area to the far end of the aluminum fin array, with a smaller “U” shaped one laid flat beneath the fin array (you can see it poking out in the third photo of the third row below). The cooler also cools the GDDR3 memory modules, the NVIO chip (mentioned below), and some power regulators on back of the card.
 Cooler base and profile |
 Cooler base |
 Match it up |
 Blower style fan |
 Top view of the cooler with the shroud removed |
 Side view showing the longer of the two heatpipes |
 You can just see the second heatpipe poking out |
 Closer shot of the second heatpipe |
 The array of narrow cooling fins |
 Straight shot out the back of your case |
Overall the 8800 cooler is very well designed, very quiet, and does a remarkable job at keeping the works cool.
With the cooler removed the first thing you’ll probably notice is that the G80 is actually hidden beneath a large heatspreader. Normally I wouldn’t think twice about taking something apart and giving you a good look at what would normally be hidden, but in this case my nerves aren’t quite strong enough.
Around the GPU are arrayed the Samsung GDDR3 memory modules rated for 1.1ns (900MHz), 12 of them on the GTX and 10 on the GTS. To the left of the GPU is a curious little chip that caused a lot of speculation leading up to the launch of the G80. This chip, of course, is the NVIO chip which handles all input and output signaling to and from the card. It’s responsible for the dual link DVI ports, VGA, component inputs, and also SLI communications among other things.
 8800 GTX bare board shot |
 8800 GTS bare board shot |
 "G80 graphics core (under the heatspreader, natch)") G80 graphics core (under the heatspreader, natch) |
 Samsung 900MHz GDDR3 memory |
 NVIDIA's new NVIO chip |
 8800 GTX dual-SLI connectors |
 8800 GTS single SLI connector |
On the GTX there are not one but two SLI connectors. NVIDIA hasn’t stated exactly why there are two of them, only to say that they will enable ”future enhancements to SLI”, but speculation is that with two SLI connectors per board it will allow NVIDIA to daisy chain more than two cards together. Curiously, however, the GTS only has the one SLI connector which apparently means GTS owners aren’t going to be treated to any such enhancements.
Here are a pant loads worth of more photos, just because:
 BFG 8800 GTX |
 "BFG 8800 GTX (back)") BFG 8800 GTX (back) |
 XFX 8800 GTS |
 XFX 8800 GTS |
 BFG 8800 GTX |
 BFG 8800 GTX |
 XFX 8800 GTS |
 XFX 8800 GTS |
 BFG 8800 GTX |
 BFG 8800 GTX |
 XFX 8800 GTS |
 XFX 8800 GTS |
 BFG 8800 GTX |
 BFG 8800 GTX |
 XFX 8800 GTS |
 XFX 8800 GTS |
Games Benchmarks (click for settings)
Image Quality Settings
In most of our benchmark charts we provide links at the top of the charts to switch between these settings. Simply click the setting to view the results. We do this to reduce clutter on the benchmark pages and to make it easier to see the impact that applying antialiasing and anisotropic has on the graphics cards performance. Note that for this review, with its emphasis on the ultra high-end hardware, the 4xAA/16xAF results are the default.
System Setup
| Test Systems | NVIDIA | ATI |
| Motherboard ( chipset ) |
Foxconn C51XEM2AA (NForce 590 SLI) |
MSI K9A Platinum (ATI RD580/SB600) |
| CPU | AMD Athlon FX-62 @ 2.8GHz | |
| Graphics Card | BFG 8800 GTX |
ATI Radeon X1950 XTX |
| Driver Version | Forceware 96.94 Beta | Catalyst 6.10 WHQL |
| Memory ( Timings ) |
2GB (2x1024MB) SuperTalent PC2-6400 DDR2 @ 800MHZ ( 4-4-3-8 2T) |
|
| Hard Disk | Western Digital Caviar WD2500KS SE16 250GB | |
| Sound | Onboard | |
| Network | Onboard | |
| PSU | PC Power & Cooling 1KW Turbo-Cool Quad-SLI | |
| OS | Windows XP Pro SP2, DX9c (August 2006) | |
| Card Specifications | BFG 8800 GTX | XFX 8800 GTS | NVIDIA 7950 GX2 | BFG 7900 GTX OC | ATI X1950 XTX |
| Core | G80 | G80 | G71 | G71 | R580+ |
| Silicon Process | 90nm | 90nm | 90nm | 90nm | 90nm |
| Transistor Count (millions) |
681 | 681 | 278 | 278 | 384 |
| Core Speed MHz | 575 | 500 | 500 | 670 | 650 |
| Memory Speed MHz (Effective) |
900 (1.8GHz) |
800 (1.6GHz) |
600
(1.2 GHz) |
820 (1.64GHz) |
1,000 (2.0GHz) |
| Memory Size | 768 MB | 640 MB | 1,024 MB
(512 MB x 2) |
512 MB | 512 MB |
| Bus Standard | PEG x16 | PEG x16 | PEG 16x | PEG 16x | PEG x16 |
| Bus Width | 384bit | 320bit | 512bit
(256bit x 2) |
256bit | 256bit |
| ROPs | 24 | 20 | 32
(16 x 2) |
16 | 16 |
| Pixel Shaders | 128 | 96 | 48
(24 x 2) |
24 | 48 |
| Vertex Shaders | 128 | 96 | 16
(8 x 2) |
8 | 8 |
| Peak Memory Bandwidth (GB/s) |
86.4 | 64.0 | 76.8
(38.4 x 2) |
51.2 | 64.0 |
| Pixel Fillrate (million pixels/sec) |
13,800 | 10,000 | 16,000
(8,000 x 2) |
10,720 | 10,400 |
| Texel Fillrate (million texels/sec) |
18,400 | 12,000 | 24,000
(12,000 x 2) |
16,080 | 10,400 |
| API Compliancy | DX 10 | DX 10 | DX 9.0c | DX 9.0c | DX 9.0c |
| MSRP $US | $599 | $449 | $549 | $449 | $449 |
The Windows XP desktop was set to 1280x960 with a 32bit color depth and 85Hz refresh rate for all tests. Refresh rate locks for 3D graphics modes, as supported by both NVIDIA and ATI graphics control panels, was not enabled. V-Sync was forced off via the graphics card control panel as well. For the NVIDIA 7- and 8-series tests image quality was set to "High Quality".
Anti-Aliasing and Anisotropy were applied in the game engine where the options existed. For games that did not support those options natively, the graphics card control panel was used.
Custom batch files were used when possible for automated benchmarking, which are available upon request. When manual benchmarking was necessary Fraps 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.
Sound and networking interfaces were enabled for all tests.
Windows XP was installed fresh on two seperate but identical hard-drives prior to testing (one for the ATI system, one for the NVIDIA system).
Image quality was always one of the biggest issues we had with NVIDIA’s 7-series. The AA was decent, especially the mixed modes, but what we took particular issue with the absolutely atrocious texture filtering optimizations NVIDIA forced on us. It wasn’t just us either; I think everyone would unanimously agree that NVIDIA was behind in the IQ department; especially after ATI introduced HQ Anisotropic Filtering with the R520 late last year, rubbing salt in NVIDIA’s open wound.
While NVIDIA couldn’t do much to improve the 7-series IQ, they seemed to be listening to the end users and made some really great improvements in this area with the 8-series. So great was the focus on IQ this time, in fact, those NVIDIA marketers put a shiny new wrapper around all the new features and called it the Lumenex Engine (it’s latin for something, apparently).
On this page and the next we will explore those improvements and draw some possibly subjective conclusions.
We’ll start with anti-aliasing quality and performance first, and then move on to texture filtering on the next page.
Anti-Aliasing
There are a few things worth mentioning about AA in general, things that have changed from the way things were, and some new things I’d like to point out before we get on to specifics.
In the control panel AA options is a new setting called “Enhance the Application setting”. This option exists so that games can control AA, as it should be, while still allowing us to use NVIDIA’s new AA modes. So instead of forcing AA on every edge, you can enable “Enhance the Application”, set your desired AA mode, and then the game determines what and what-not should be anti-aliased. This is a welcome and very clever compromise to forcing AA outright (which still exists, if you want to use it) that I expect won’t be exclusive to NVIDIA for too long.
We should also mention gamma corrected AA. ATI introduced gamma corrected AA with their R300 in 2002, and NVIDIA followed suite with the 6- and 7-series a bit later. Why mention it now? Well, for the simple reason that, unlike the 6- and 7-series, gamma corrected AA is now enabled by default with the 8-series cards.
You can see both of these settings in the screenshot of NVIDIA’s 96.94 control panel right here.
The first thing NVIDIA will point out when you ask them what new in AA is their new CSAA, or Coverage Sampling Anti-Aliasing. It’s like… coverage, or something… and sampling of samples. It looks good and goes really fast, according to NVIDIA, but getting a reasonable explanation out of them of how it works exactly is pretty hard. Even now they continue to hold their cards pretty tight to their chests. Who, besides ATI engineers and the zany crowd over at Beyond3D, really cares about the filthy technical details about how it works anyway? All us mere mortals need worry about is whether or not CSAA does what NVIDIA claims it does, and what they claim is improved quality and speed.
| CSAA | |||||
| Quality Level | 4x | 8x | 8xQ | 16x | 16xQ |
| Texture/Shader Samples | 1 | 1 | 1 | 1 | 1 |
| Stored Color/Z Samples | 4 | 4 | 8 | 4 | 8 |
| Coverage Samples | 4 | 8 | 8 | 16 | 16 |
What we’ve been given are the 2x and 4x AA modes we’ve grown accustomed to, and four new modes: 8x, 8xQ, 16x, and 16xQ.
The 2x and 4x AA modes are pretty much unchanged from the 7-series cards with identical subsample pattern locations. It’s worth mentioning though that 4x AA is now done in a single cycle on the G80.
The 8x AA mode is the first of the new modes to take advantage of CSAA. It stores 4 color/Z Samples to save bandwidth while providing quality that can be as good as a true 8x MSAA mode on most edges. On stencil shadows and with Transparency AA you only get 4x AA though.
Unlike the 8x AA mode, 8xQ is a true 8x MSAA mode. It stores 8 color/Z Samples in memory and provides 8x AA on stencil shadow edges and stores 8 samples with Transparency AA.
Both 16x modes take advantage of CSAA. With the first 16x mode it stores 4 color/Z Samples so it should perform about the same as the 8x mode, but should look better than the 8xQ mode in most cases. This mode provides 4x AA on stencil volume edges and stores 4 samples with Transparency AA.
Finally the 16xQ mode: it stores the same 8 color/Z Samples from the 8xQ mode, and provides 8x AA to stencil shadow volumes and Transparency AA.
Here’s a little app that shows the AA sample positions within a pixel. It doesn’t show CSAA positions, but it will give us an idea as to what’s going on:
So with a new 8xQ mode that really does take 8 discrete samples and a new 16xQ mode that has the potential to provide some really spectacular image quality, you have to figure the quality claims are going to hold up to some close scrutiny. You also have to figure that with the bandwidth and resource saving techniques that CSAA offers, and the OMFG z-rate of the G80, that the speed claims are going to be pretty close on too.
So to help test things out, starting with image quality since that’s the most fun, here are a couple of simple JavaScript apps designed to let you easily compare the main contenders. The first app makes uses of Futuremark’s 3DMark06 Game Test 4 (Deep Freeze), which uses HDR effects and serves well to illustrate the HDR+AA ability and quality of the 8800 and X1950 cards. The second app uses a scene from Half-Life 2 with a lot of thin and thick geometry at varying degrees and loads of alpha transparent textures that illustrates the edge anti-aliasing and transparency adaptive AA of the 8800, the 7900, and the X1950. There’s more commentary in each of the apps. Note that these may take a long time to preload all the screenshots, especially the Half-Life 2 app which has almost 12MB worth of them. You also need JavaScript enabled in your browser for them to work, of course.
 Click for HDR+AA Quality Tests in Futuremark's 3DMark06 Game Test 4 (Deep Freeze) |
 Click for AA tests with Transparency/Adaptive AA in Valve's Half-Life 2 |
So if you’ve looked inside those apps you’ll probably agree that AA quality seems to be very good, certainly every bit as good as NVIDIA has claimed. What came as a bit of a surprise, however, was how well ATI’s AA stands up to the 8800’s new modes. Considering ATI hasn’t significantly changed their AA since the R300 was released over four years ago, I think that says a lot.
Great image quality, however, is only good if it’s actually useable. Not much point in nice AA if it turns your gaming session into a slideshow after all.
To that end we ran benchmarks on all the AA levels with and without Transparency/Adaptive AA on all the cards compared here using our standard Half-Life 2 Episode One test. Click the links at the top of the chart to show results with and without Transparency/Adaptive AA.
|
[ Without Transparency/Adaptive AA ]
[ With Transparency/Adaptive AA ]
|
| Performance drop using TAA/AAA | ||||||
| 2x | 4x | 6x/8x | 8xS/8xQ | 16x | 16xS/16xQ | |
| BFG 8800 GTX | -3% | -16% | -14% | -43% | -14% | -34% |
| XFX 8800 GTS | -3% | -16% | -13% | -37% | -13% | -29% |
| 7950 GX2 | -2% | -7% | -8% | -8% | -8% | |
| 7900 GTX | -3% | -8% | -8% | |||
| X1950 XTX | -4% | -13% | -20% | |||
In some games you can expect to see virtually no difference in performance between 2x and 4x AA on the 8800 cards, but in this case there is a difference, albeit slight. Framerates using the 8x and 16x modes are virtually identical, as expected, and 8xQ performance is pretty close as well. The only significant performance drop off comes with the 16xQ mode, though it is still definitely playable at these settings.
With the 8800 GTS we see the same performance characteristics and the results show that it still remains playable with 16xQ AA enabled, but just barely.
The 8800 takes quite a hit with Transparency AA enabled though, more so than any of the other cards involved in the test. The 8xQ mode in particular shows a 43% drop in performance on the 8800 GTX with TAA enabled.
Without doubt the biggest issue many people had with NVIDIA's 7-series was the terrible texture filtering it offered by default. Texture Shimmering, a side effect of overzealous filtering optimizations on NVIDIA’s part, was always an huge issue for 7-series cards, and while it was reduced over time, it never really went away completely.
With the tremendous texture filtering capability of the G80 it’s good to see NVIDIA take advantage of it and drastically improve their default texture filtering quality. Texture shimmering is completely gone now from what I can tell, even in the default control panel “Quality” mode.
Probably most notable, though, is the almost perfectly angle-independent anisotropic filtering that the 8800 series uses as the default setting. This is probably the biggest image quality improvement NVIDIA could have made, and they definitely deserve extra points for making it the default setting. I was a little disappointed to see that they are still doing their trilinear optimization by default (aka “brilinear”), but in all fairness it is virtually impossible to tell the difference now with and without those optimization enabled (very much unlike the 7-series cards). Even so, High Quality mode disables those trilinear optimizations if you run into some games where it becomes evident. Check out the JavaScript app to the right where we compare sampling patterns on the 8800, the 7900, and ATI’s X1950 using D3D AF-Tester.
Anisotropic Filtering Pattern Tests using D3D AF-Tester
Screenshots aren’t the best way to showcase filtering quality (you really need movement to spot texture shimmering and other issues), but we make do.
Below we have provided several more JavaScript apps to help you see up close what the texture filtering quality of the 8800 is like compared to the 7-series and ATI’s X1950. In lieu of doing one game and providing shots of all quality levels, we are doing several games from different genres and comparing the highest quality possible (High Quality 16x AF on the 8- and 7-series cards, 16x High Quality AF on the X1950 XTX). The first game is the new SimBim racing sim GTR 2, the second is the excellent RTS Company of Heroes from Relic, and the third is the old shooter Far Cry from Crytek. Commentary on the texture filtering quality for each game is included within the apps.
 AF quality with GTR2 |
 AF quality with Company of Heroes |
 AF quality with Far Cry |
It’s difficult to quantify the results shown here. In Company of Heroes the 8800 filtering looks noticeably better, and in GTR 2 ATI hardware appears to make the textures cleaner and sharper. In Far Cry, conversely, it looks to be a virtual tie between the 8800 and the X1950. Overall, at least when directly comparing the settings used in this review, I would say that both the 8800 and ATI’s X1950 do an excellent job at texture filtering. As for default control panel settings, NVIDIA clearly has the better filtering pattern which they definitely deserve some credit for.
As with the AA tests on the previous page, AF quality is only good if it doesn’t hurt performance too much. The performance results below were gathered using the second timedemo in out Half-Life 2: Lost Coast benchmark test. I’ve found that this timedemo in particular shows performance differences on the various texture filtering levels more-so than any of our other tests.
|
[ Default Settings ]
[ High Quality ]
|
| Performance drop using High Quality filtering vs default settings | ||||
| 2x | 4x | 8x | 16x | |
| 8800 GTX | -0.4% | -0.9% | -1.6% | -2.2% |
| 8800 GTS | -0.9% | -1.8% | -3.0% | -3.4% |
| 7950 GX2 | -3.0% | -5.1% | -7.6% | -9.7% |
| 7900 GTX | -4.0% | -6.6% | -9.6% | -11.5% |
| X1950 XTX | 0.0% | 0.5% | 0.3% | 0.3% |
Even though they are doing more work you can see that the 8800 cards suffer much less of a performance impact when compared to the previous 7-series cards with High Quality texture filtering enabled. The ATI X1950 XTX, though, isn’t affected at all when HQ AF is enabled (slightly faster, even).
Here we’ll look at Power and Temperature of the 8800 GTX and 8800 GTS and compare with cards from NVIDA’s previous generation and the immediate competition from ATI (sorry, I don’t have a dB meter so no noise tests). Idle tests were run with the test system powered on and sitting on the Windows desktop. To get Load results we used ATITool’s 3D View (the spinning furry cube view). This applies to both the Power and Temperature tests.
Power consumption was one of the big things many people talked about leading up to the launch of the G80 boards. The assumption was that the G80 would require massive amounts of power and that most everyone would need a PSU upgrade to accommodate the thing.
Surprisingly, however, NVIDIA managed to get the power profile of even the GTX board down to very manageable levels. NVIDIA claims that the GTX requires only a little more power than ATI’s X1950 XTX, a bold claim considering the 681m transistor beast hiding under the heat-spreader. To test the claim we plugged our test systems into a digital watt meter (Kill-A-Watt from P3 International).
At peak load the GTX draws a significant amount of power, and even idling on the Windows desktop our test results show that it’s drawing very near 200 watts. 35 watts over the X1950 XTX is not insignificant, especially when you consider how much power the X1950 XTX is pulling.
Looking at the 8800 GTS though we see much lower power requirements, 44 watts lower than the 8800 GTX and 9 watts lower than the X1950 XTX under load. It’s safe to say that if you’re already running an X1950 XTX then an upgrade to the 8800 GTS should be rather painless.
To get the temperature results we first loaded up the graphics hardware and recorded the peak temperature level. Without rebooting we then removed the load (closed the furry cube view) and let the cooler do its job. Once the temperature leveled out we recorded the Idle results.
As you can see, even with the higher power draw, the GTX remains cooler under both idle and load conditions than the X1950 XTX (it’s also much quieter under load, but you’ll have to trust my ears for that).
Now for what everyone came here to see; hardcore nudity!
Some ugly bugs in nTune prevented us from using that for our overclocking tests so we resorted to the newest beta of ATITool instead, which seemed to work. One problem we did run into was that it would crash whenever we tried to get either card to go over a core speed of 635MHz. It seems a little too coincidental to suggest that both the 8800 GTX and the 8800 GTS cards can't overclock beyond this exact point, so we are chalking it up to an application issue instead. We ultimately settled on core overclocks of 625MHz for both cards, which is 50MHz over the stock GTX speed and an extremely impressive 125MHz over the stock GTS speed.
Memory overclocking went smoother however, netting us again impressive results. For the GTX we reached 1100MHz (that's 2.2GHz effective), or 200MHz over the stock speed. For the GTS we reached 980MHz (or 1.96GHz) over the stock speed. Very nice results from both cards.
Performance results using Prey and Splinter Cell Chaos Theory are below:
|
[ 4x AA / 16x AF ]
[ No AA / No AF ]
|
|
[ 4x AA / 16x AF ]
[ No AA / No AF ]
|
The G80 is a giant leap forward, architecturally, over anything that came before it. It is well beyond the leap in performance and features we saw with ATI’s 9700 Pro four years ago. While there are a lot of things about the G80 that caught a lot of people off guard, I’m willing to bet that for most people the fact that it features a fully Unified Shader Architecture is number one on their list of G80 surprises. It wasn’t all that long ago where even NVIDIA was saying that a USA probably wasn’t appropriate “at this time”. NVIDIA has become well known for their smoke and mirrors act, and comments like that are prime examples of it. Number two on the list of surprising things is probably the fact that it has 681 million transistors on a 90nm process and doesn’t require the energy output equivalent of a small star to power it. When everyone was assuming that the GTX would require an extremely beefy PSU, with some suggestions that anything short of one rated for 1KW wouldn’t do, NVIDIA revealed that it wouldn’t need much more power that what the previous generation cards were using. That alone is a testament to the amazing job NVIDIA has done in engineering the G80.
The level of performance you get with the 8800 GTX is nothing short of astounding. In most cases it’s fully twice as fast as the 7900 GTX and X1950 XTX, and the previous King of the Hill, the 7950 GX2, can’t touch it either. After seeing how fast the thing was I even decided to throw in SLI and Crossfire tests too, just to illustrate the point further and show that, more often than not, the 8800 GTX is faster than 7900 GTX SLI and X1950 Crossfire too. And that’s not even considering the image quality improvements you get with it; looks better, goes faster.
If you had told someone six months ago that this card would be this fast you would have gotten a swift kick in the head for your troubles and, upon reflection, would have agreed you deserved it.
Not only do you get ground-breaking speed and image quality, you also get full support for DirectX 10. There aren’t any DX10 games out yet or, for that matter, DX10 itself, but when you’re dropping this kind of money on a graphics card it’s good that it’s future-proof and gives us something to look forward to.
The 8800 GTS is no slouch either. It’s clearly faster than the 7900 GTX and X1950 XTX, and in most cases beats the 7950 GX2 as well. $449, the same asking price for a 7900 GTX or X1950 XTX, and cheaper than a 7950 GX2, for a card with this level of performance and with all the same features as the GTX, that’s a pretty good deal if you ask me.
I can’t think of anything really negative with the G80 or either SKU that’s worth mentioning. The GTX board is long enough that it would probably not fit in many average computer cases, but this is an enthusiast class board and enthusiasts have enthusiast class enclosures, so the board length shouldn’t be an issue for most of the people this is targeting at. I also encountered some minor driver issues during my tests, but nothing remotely show-stopper serious. Given NVIDIA’s track record with drivers I don’t expect to have to wait too long to see the few things I reported fixed.
With the G80 sitting atop their lineup, and more G8x GPUs quickly coming to fill in the rest, NVIDIA finds themselves in an absolutely undeniable performance and features leadership position. This holiday season the only choice you have to make is between the 8800 GTX and the 8800 GTS, everything else has been made obsolete.
Unbelievable speed, fantastic image quality, and true next-gen features are what make the 8800 GTX the card to have
All the features and image quality of the 8800 GTX, with class leading performance, but in a smaller and more affordable package
