Introduction
In early October we brought you a small preview of the Rage Fury MAXX to give you an idea of what ATI has coming in the very near future. This future will soon become reality as we have been proven that things are moving along fairly well for the Fury MAXX. ATI once again dropped by the US lab for a few hours to let us crash-test the current status of their Rage 128 Pro AFR (Fury MAXX), which is very close to being released. The drivers have been improved as well as the board itself. This is still not a final release of the product but it should give us a good idea of what the product should and shouldn’t be able to do.
Rage Fury MAXX
For those of you who don’t recall what the Fury MAXX is, it’s basically two Rage Pro’s on a single board using AFR technology. This gives them a “brute force” method at getting higher fill-rate and memory bandwidth. Thanks to the two separate core’s (running at 125Mhz each) working in conjunction, we get nice fill-rate performance (500 MPixels/s) as well as broad memory bandwidth with each core having a dedicated frame buffer (two independant 32 MB sets of memory running at 143 MHz).
Before we get into the benchmark results, let have a quick look over the features of Fury MAXX in comparison with the other high-end 3D-solution, the GeForce from NVIDIA.
Fill Rate | |
ATI Rage128 Pro AFR | NVIDIA GeForce256 |
2 x 250 Mpixels/s = 500 Mpixels/s | 480 Mpixels/s |
In terms of fill rate, Fury MAXX has a slight edge over GeForce, but generally both solutions are close to identical. This will mean that both cards will score close to the same in games that don’t use T&L, if fill rate is the only limiting factor.
Memory Bandwidth | |
ATI Rage128 Pro AFR | NVIDIA GeForce256 |
2 x 2.288 GB/s = 4.576 GB/s | 2.656 GB/s as SDR-version 5.312 GB/s as DDR-version |
You can see that Fury MAXX is way ahead of GeForce w/SDR in terms of memory performance. This will make Fury MAXX look a lot better than GeForce’s SDR-version at high resolutions and high color-depths. GeForce’s DDR-version however comes with a higher memory-bandwidth than Fury MAXX, something that might cost a lot more money though.
Rage Fury MAXX, Continued
32-bit Color Rendering | |
ATI Rage128 Pro AFR | NVIDIA GeForce256 |
Switches down to 16-bit deep Z-buffer if ‘game doesn’t request 32-bit deep Z-buffer’ | always using 32-bit deep Z-buffer with 32-bit color |
This little trick used by ATI and Matrox can make Fury MAXX look a lot better at 32-bit color in many games that don’t use 32-bit Z-buffer.
Hardware Transform & Lighting | |
ATI Rage128 Pro AFR | NVIDIA GeForce256 |
not available | implemented |
Here Fury MAXX is clearly at loss. GeForce will look a lot better at upcoming T&L-titles, especially with a slower CPU. With fast CPUs Fury MAXX could look pretty much the same as GeForce, if the T&L computed by the processor is as fast as GeForce’s internal T&L-engine
Game Responsiveness or ‘Lag’ | |
ATI Rage128 Pro AFR | NVIDIA GeForce256 |
The time it takes until a user-input is actually displayed on the screen is at least two frames with the AFR-technology. | GeForce renders one frame after another with the same chip, thus it takes at least one frame for a user input to become visible on the screen. |
This topic is particularly important for hard-core gamers. I recently received an email from a reader who beat me up for my statement that frame rates above 60 fps are rather meaningless. He is correct to say that he wants to get as many fps as possible, when he e.g. makes a very fast 180 degrees turn. The more frames he sees in the time the turn takes, the more he sees of the game environment around him. With few frames he could miss the enemy approaching him from the side, with many frames he will spot it. It’s just as important to see the turn you make on the screen as soon as possible. In case of the Rage Fury MAXX, you will have to wait for at least 2 frames until you will see your turn actually happening on the screen, because the rendering time (the time it takes until a new frame is displayed on the screen) for a frame is still the same as with only one Rage 128 Pro chip. GeForce will most likely have a better game responsiveness, because it can render the same frame in pretty much half the time.
This graphic shall make it easier to understand this issue:
The green or red boxes are supposed to show the time it takes to render a frame. At the right end of each box the rendered frame is displayed on your screen. In this simplified graph it takes the same time for each frame. Now you can see that each Rage 128 Pro chip requires double the time for rendering one frame than GeForce needs. In this example Fury MAXX is still scoring the same frame rate by displaying the frames from each Rage 128 Pro chip alternatively.
Now see what happens if there comes a user input. Although in this example both 3D-solutions offer the exact same frame rate, it takes ATI’s Rage 128 Pro a lot longer to display the frame that was generated in response to the user input. At the time of the user input, GeForce can start rendering the changed frame right after the last frame was finished.
It’s a bit different in case of ATi’s AFR-solution. One of the two chips is the one that’s closer to finishing its current frame and this one will then start rendering the new frame that’s the response to the user input. By this time, the other chip is still busy rendering the previous frame, which has already been displayed by the GeForce at this time. You could also say that ATi’s AFR-solution is always lagging behind the actual game by one frame.
Now you might say that this little time it takes for one frame might not be noticeable, but consider the following. Let’s say the game is running at an average frame rate of 60 fps. This means that it takes 16.7 ms for each frame in average, more realistically 15-20 ms. Now it really depends what kind of gamer you are. If you are used to playing multi-player games on a LAN with a ping of 10-20 ms, an additional lag of 15-20 ms is definitely noticeable. Even if your ping is usually 50 ms you will still be able to feel a graphics card that adds another 15-20 ms. People that play single-player games or modem-users with pings of up to 200 ms won’t be bothered by Fury MAXX’ additional lag. It’s up to you to decide if the immanent lag of ATi’s AFR-architecture is able to bother you or not.
Rage Fury MAXX, Continued
MPEG2-Decoding/Encoding | |
ATI Rage128 Pro AFR | NVIDIA GeForce256 |
motion-compensation and iDCT/DCT hardware implemented | motion-compensation support unclear, no iDCT/DCT |
For people who would like to watch DVDs on their computer, ATI has definitely the better solution. The same is valid for people who like to encode MPEG2-movies on their system, since ATI’s DCT is the only valid solution today that lets you do this with a mainstream graphics chip in an acceptable fashion.
Looking at all those topics shows that GeForce is more future proof than Fury MAXX. However, it really depends on how long it will take until T&L-titles will become available. Fury MAXX has a pretty good standing for the current crop of games and it offers a significantly better ability to display DVD-movies. It will take GeForce with DDR-memory to really beat Fury MAXX, but we neither know when this version will become available nor what price tag it will carry. The importance of the ‘lag’-issue will depend on you. Die-hard gamers should definitely consider it.
ATI Rage Fury MAXX drivers
The ATI driver has improved quite a bit since we last saw them in our various tests. The OpenGL driver seems to be solid while the D3D (especially DX7) driver still seems to require some work. Applications either ran fine without any visual issues or the programs outright failed, something that’s currently acceptable considering the beta-stage of the drivers. The only game that worked and showed visual issues was Descent 3 in DirectX. We saw heavy amounts of frames dropping while the demo was running. It appears the driver had issues keeping the two chips in sync. I hope to see this disappear once we get our hands on a final rev of the driver. All in all, the D3D driver is far from final but OpenGL appears to coming along very well.
The ATI Rage 128 drivers are used for the Rage Fury MAXX and Rage Fury Pro. You can look at the drivers here.
The Benchmark Setup
We ran the Fury MAXX under the same system-setup that we used for the previous 3D-card evaluations, so that the scores could be compared with the other cards we’ve already tested. A Pentium III 550 may not be the fastest CPU currently available, but its performance should lie where the majority of gaming systems is at right now. We should also not forget that the currently small T&L-impact coming into account with Quake3 is camouflaged when using a very fast CPU. It’s logical that the impact of GeForce’s internal T&L-engine only shows in games that support it properly. Games that only take little advantage of GeForce’s T&L will look good on Fury MAXX as long as the CPU is fast enough. We were very disappointed to see that Fury MAXX completely failed to run Dagoth Moor Zoological Gardens, the only T&L-title that can be used for testing right now.
I’d also like to mention that ATI did not give us enough time to do some really major testing with Fury MAXX. While the 5 hour-visit of a manufacturer bringing his new product may enable a publication to present an early review, the resulting hectic test-session is not the right base for an actual thorough in-depth review. We apologize for that.
Hardware Information | |
CPU | PIII 550 |
Motherboard (BIOS rev.) | ABIT BX6 2.0 (BIOS date 7/13/99) |
Memory | 128 MB Viking PC100 CAS2 |
Network | Netgear FA310TX |
Driver Information | |
ATI Rage Fury MAXX/PRO | BETA/6.30CDH34 |
NVIDIA GeForce 256 | 4.12.01.0353 |
NVIDIA TNT2 Series | 4.11.01.0353 |
Voodoo3 Series | 4.11.01.2103.03.0202 |
Matrox G400 Series | 4.11.01.1300 (5.30.007) |
Environment Settings | |
OS Version | Windows 98 SE 4.10.2222 A |
DirectX Version | 7.0 |
Quake 3 Arena | V1.08 command line = +set cd_nocd 1 +set s_initsound 0 |
Shogo | V2.14 Advanced Settings = disable sound, disable music, disable movies, disable joysticks, enable optimized surfaces, enable triple buffering, enable single-pass multi-texturing High Detail Settings = enabled Fortress Demo |
Expendable | Demo Version Setup = use Triple Buffering Audio = disable sound |
Descent III | Retail version Settings = -nosound -nomusic -nonetwork -timetest |
DMZG | Demo Version Command line = -bench -tl on |
Benchmark Results – Shogo results
A brief note before we get into the benchmark section. Due to the beta drivers, we were unable to run our full test suite. We did however test what would completely run without crashing.
At low resolution we can really test how good the drivers are because typically the fill-rate is out of the picture. Here we can see that the Fury MAXX does well by keeping at the top of the pack. The visual quality of Shogo was very good. I saw no visual signs of sync problems.
Fury MAXX pushes its way to the top but falls slightly behind the GeForce for top honors. With some driver tuning we might see a performance jump here but only enough to creep a bit closer to the SDR GeForce board.
The Fury MAXX does admirably well but not good enough to knock the GeForce off the top. I’m a bit surprised that it didn’t match if not surpass the GeForce at this resolution with its 500 Mpixel/s fill-rate capabilities.
Benchmark Results – Descent 3 results
Interestingly enough, the Fury MAXX is able to run Descent with its OpenGL ICD well, where other companies can’t even get it working. It isn’t impossible for ATI to squeeze another 10% after the driver mature a little bit more. At this low resolution, driver maturity means quite a bit.
As we increase the resolution and lift the responsibility from the driver, we start to push the cards a little bit. Fury MAXX gains a little bit of ground and managed to slightly pass the TNT2 Ultra while coming up very short of the GeForce cards.
Look at the efficient gains the Fury MAXX performs when compared to the Fury Pro. I’m sure the double memory bandwidth lends a big helping hand. This is a similar story as we saw with Shogo at this resolution; the Fury MAXX gets within about 10% of the SDR-GeForce. Given a theoretical higher fill-rate than the GeForce we should be seeing higher scores in such a simple application where TnL isn’t present.
Benchmark Results – Quake3 Arena Normal
Clearly this setting in Quake Arena is testing the driver performance as the graphic cards with similar drivers are all tied. Although the Fury MAXX is near last place in our line-up, it still provides plenty of performance. I have no doubt that with final drivers, this situation will change.
We’re seeing some pretty consistent performance differences between the Fury MAXX and the GeForce cards in all our test applications. At 1024 the Fury MAXX is coming in a decent 3rd place but about 25% slower than the SDR-GeForce board. When we transition to 32-bit modes I’m sure this gap will narrow due to the massive memory bandwidth of the Fury MAXX.
The Fury MAXX creeps within the range of the SDR-GeForce board and I wouldn’t be too surprised if we see some driver enhancements put it on par if not ahead of the SDR GeForce. We’ll have to wait and see what ATI can pull off.
Benchmark Results – Quake3 Arena High Quality
There is nothing too surprising here at low resolution with the Fury MAXX driver hindering the potentially higher scores. A decent 73 FPS is nothing to yawn at but it’s not good enough to catch the GeForce boards.
We’ll look what we have here! Look closely at what’s going on. Notice how important memory bandwidth is to the GeForce boards. The DDR crushes the SDR board by a whopping 50%. At this resolution at in 32-bit color, you need not only fill-rate but also a strong memory performance. This is why the Fury MAXX bursts ahead of the SDR GeForce board. It still comes a little over 20% short of the DDR board but impressive nonetheless.
Woah, look at that fight for the top spot. The GeForce DDR board shares the top honors with the Fury MAXX. The SDR GeForce lags in a distant 3rd place. At this resolution Quake3-Test shows that it’s a great tool to find out about the memory performance of a 3D-card. Geforce w/DDR has almost the same fill rate and only a tiny bit better memory bandwidth than Fury MAXX, thus the almost identical results. Unfortunately T&L doesn’t come into account here though.
Conclusion
It seems ATI is well on it’s way to producing a card with impressive fill-rate and memory bandwidth. The board has proven that it can perform with the best of them on a flat out fill-rate and memory-bandwidth competition at higher resolutions and especially when 32-bit color is involved. With a brute force approach ATI took to the fill-rate and memory bandwidth problems, we should see the Fury MAXX do extremely well in high-resolution 32-bit color non-T&L games. Another thing to keep in mind is that on top of all of this, you get excellent DVD playback performance second to none in its class.
Please remember however, that this article is merely a preview for several reasons. First of all, we didn’t have the chance to compare Fury MAXX to GeForce in a T&L-supporting game-title. The driver would not allow DMZG to run, which may make Fury MAXX look better than it actually is. We will test Fury MAXX with T&L-titles as soon as we can. We also need to find out how much the CPU impacts Fury MAXX’ performance. Basically, a very fast CPU could equalize the missing T&L-feature of Fury MAXX.
ATI promises to have a product out early next year that’s basically a Fury MAXX with T&L, but that’s in a few months when we might see a faster GeForce product released as well as the latest offerings from S3 that has T&L. I’m also not too impressed with the release price I’m hearing. The price is said to be that of the SDR GeForce boards. While the board is in the same ballpark as the SDR board in many cases, you get SDR like performance without the benefit of T&L. In extreme situations Fury MAXX managed to pull near the lead but not nearly fast enough to actually use to play. At that price range, I expect to have the T&L to match with my great fill-rate performance to be able to play with upcoming leading edge software titles. It will come down to the final pricing of the product, the performance of the released drivers and how much you value DVD performance. We’ll have to wait and see what ATI offers in the coming weeks when the Fury MAXX is officially released and you can be sure that we haven’t said the last word about Fury MAXX yet. A much more thorough test will follow soon.