 | |||
|
|
|
|
|
 |
ATI RADEON X700 XT Performance Preview September 21, 2004 |
|
 |
|
| |
Summary: NVIDIA isn't the only one with a brand new mainstream card, as today ATI is unveiling their X700 lineup. Like GeForce 6600, X700 sports 8 pixel pipelines, GDDR3 memory, and high clock speeds. ATI one-ups NVIDIA in one area however: vertex processing. See how the X700 XT performs in comparison to the 6600 GT and GeForce 6800, RADEON 9600 PRO, 9700 PRO, and 9800 PRO, and two X600s inside. As an added bonus, we even threw in overclocking results!
 Introduction | Page:: ( 1 / 29 ) |
RADEON X700 XT: ATI’s RADEON 9500 PRO killer
When it comes to delivering the most bang for the buck, ATI’s RADEON 9500 PRO has been in a class of its own among DirectX 9 graphics cards. Thanks to its 8-pixel pipeline architecture (with four vertex units) and 275MHz core clock frequency, the RADEON 9500 PRO has the fill rate to pump pixels faster than many more contemporary DX9 mainstream cards. This allows it to perform well in situations with moderate screen resolutions, even with a little bit of anti-aliasing and anisotropic filtering thrown in for enhanced visuals.
[image]
| <% print_image("01"); %> | <% print_image("02"); %> |
When it was introduced nearly two years ago, enthusiasts flocked to the RADEON 9500 PRO. For about $200 you got a DirectX 9 card that played all existing games well, with the added bonus of DX9 compliance for the games of tomorrow. The RADEON 9500 PRO did so well that many gamers passed over its successor, RADEON 9600 PRO, even though in some cases the RADEON 9500 PRO was the more expensive option. In order to reduce manufacturing costs, ATI cut the number of pixel pipelines in half, they tried to offset this by cranking up the core clock speed to 400MHz, but it wasn’t enough, the RADEON 9500 PRO outperformed the 9600 PRO in all games except for Tom Clancy’s Splinter Cell, which is based on a derivative of the UT 2003 engine.
[image]
| <% print_image("03"); %> | <% print_image("04"); %> |
ATI took a second stab at truly replacing the RADEON 9500 PRO around this time last year with the RADEON 9600 XT, but even with nearly double the core clock speed (500MHz in the case of the 9600 XT) the RADEON 9600 XT wasn’t a dramatic performance enhancement over the RADEON 9500 PRO. With the graphics market’s brutal 6-month product cycles, the RADEON 9500 PRO’s run at the top has been unprecedented. If you bought one of these cards before they disappeared, neither ATI nor NVIDIA has really offered a true replacement in the $200 price segment that’s worth the cost of upgrading too up to this point. This is going to change this fall however, as both ATI and NVIDIA will be releasing a new generation of mainstream parts that should offer significantly more performance than previous midrange card offerings.
[image]
| <% print_image("05"); %> |
We’ve already shown you a sneak peek at NVIDIA’s mainstream solution, the GeForce 6600 GT. Its 8 pixel pipes and 500MHz core clock allowed it to outperform more expensive cards like the RADEON 9800 XT in many situations. But what does ATI have up their sleeve? Read on to find out!
| Core architecture | Page:: ( 2 / 29 ) |
Like the RADEON 9500 PRO and GeForce 6600 GT, for RADEON X700, ATI has adopted an 8-pixel pipeline architecture, with one texture unit per pixel pipeline. This 8x1 pipeline configuration was previously used on all of ATI’s high-end cards, such as the RADEON 9700 PRO and 9800 XT, so ATI is essentially now bringing it down to the mainstream consumer. After all, today’s high-end cards now boast 12 or 16 pixel pipelines.
Unlike the high-end cards of yesteryear (or even the GeForce 6600 GT) however, the RADEON X700 family has one additional advantage going for it: ATI equips X700 with six vertex units, two more than RADEON 9700/9800 and matching the spec of ATI’s high-end X800 series.
With six vertex engines, X700 has twice the vertex power of GeForce 6600, and is capable of running over 5 billion vertex shader operations per second. ATI has released one technology demo that takes advantage of the added vertex power found in X700/X800: Crowd Demo. This demo involves a massive combat scene with a large number of characters on-screen at once (1,400 total). The added vertex power found in X700 and X800 allows these cards to render this sequence without a massive drain in performance, offloading work from the host CPU.
[image]
| <% print_image("06"); %> | <% print_image("07"); %> | <% print_image("08"); %> |
One shipping game that will take advantage of this is EA’s upcoming title “The Lord of The Rings: The Battle for Middle-Earth”. Like last year’s blockbuster movie, this game will have scenes with massive armies waging war with each other on the battlefield; the added vertex power found in X700 could yield a sizeable boost in performance. Other RTS titles will likely see similar benefits. Another title that should benefit is Far Cry, as vertex processing is used to render its lush jungle environments, with grass, trees, and other foliage found everywhere.
[image]
| <% print_image("09"); %> | <% print_image("10"); %> | <% print_image("11"); %> |
X700 supports the same SMARTSHADER HD shaders as those found in X800. These are officially designated as 2.0b pixel shaders versus the 3.0 shaders NVIDIA supports in GeForce 6600/GeForce 6800. The key additions 2.0b brings over 2.0 is support for longer pixel shader programs, and more instructions (1,536 versus 160 previously). ATI has also improved their F-Buffer performance.
Like the X800, ATI has also integrated 3Dc support into X700. 3Dc provides up to 4:1 compression for normal maps, this increases performance and reduces memory footprint in titles that use normal maps. CryTek has committed to adding 3Dc support to Far Cry in an upcoming patch; Serious Sam 2 is another game that will take advantage of 3Dc. The core itself is built on TSMC’s 0.11-micron manufacturing process, just like GeForce 6600.
Connected to the graphics core is a 128-bit memory interface. This is the same size used on previous mainstream offerings such as the RADEON 9500 PRO/9600/9600 XT. By limiting the memory interface to 128 bits, board complexity is reduced keeping production costs down, but the tradeoff is reduced memory bandwidth. To keep the graphics core fed with data ATI uses high-speed GDDR3 memory on their X700 XT line, clocked at 525MHz. This provides over 16GB/sec of peak memory bandwidth, a figure which bests NVIDIA’s GeForce 6600 GT (which uses the same Samsung 2.0ns memory modules as the X700 XT).
| X700 lineup | Page:: ( 3 / 29 ) |
ATI’s X700 family consists of three distinct product lines. On the high-end lies X700 XT. The X700 XT ships with a 475MHz core clock speed and 525MHz GDDR3 memory. This provides a texel fill rate of 3.8 Gigatexels/second, which is just shy of GeForce 6600 GT’s 4.0 Gigatexel/second, but it boasts 800MB/sec of additional memory bandwidth.
ATI offers the X700 XT in two memory configurations. At $249 is the 256MB X700 XT, while ATI also offers a 128MB X700 XT at the crucial $199 price point. Both boards boast the same clocks, with the only difference being the additional memory. We’ve run tests that show that newer titles such as Half-Life 2 and Far Cry are capable of taking advantage of the additional memory 256MB cards provide, particularly at high resolutions with AA/AF enabled.
Just below the X700 XT lies the X700 PRO. The X700 PRO sports a 420MHz core clock frequency, with its memory running at 432MHz (864MHz effective). The X700 PRO will ship with 256MB of memory and will be priced at an MSRP of $199.
Finally, on the bottom of ATI’s X700 lineup is the X700. X700 ships with the same core features as the other cards, including the 8 pixel pipes and 128-bit memory interface, only with slower clocks. ATI has committed to 400MHz on the graphics core and 300MHz DDR1 memory (600MHz effective). These are the same clocks as RADEON 9600 PRO/X600, but thanks to the boards 8 pipeline configuration, the X700 should offer dramatically more performance when it ships and will be priced at an MSRP of $149.
All X700 cards will initially support PCI Express, just like the GeForce 6600. To serve the existing AGP add-in card market, ATI also plans to ship AGP-based X700 cards, but these cards won’t hit retail until later this year. Since the X700 supports PCI Express natively, ATI will have to integrate a bridge chip on their AGP X700s.
In addition to adding X700, ATI has made one other change to their lineup. In the coming weeks, X800 XT will make its way to the AGP platform. Up to now, all X800 XTs have been PCI Express-only. But with the lack of AGP X800 XT Platinum Edition cards at retail, ATI has decided to introduce an AGP-based X800 XT at 500/500 to help satisfy demand for high-end cards.
X700 XT board design
ATI’s X700 XT relies on a single slot copper heatsink to keep the graphics core cool. Like previous coolers, the X700 XT’s cooling unit is a ducted design. The cooler draws in air from within your system’s case and blows it across the graphics core and memory modules, which are partially covered by the copper heatsink.
[image]
| <% print_image("12"); %> | <% print_image("13"); %> |
A variable speed fan is used to keep everything cool, although we never saw the fan run at its highest mode outside of system bootup. Even when the card was overclocked the card remained in its intermediate mode, which was fortunately very easy on the ears.
| CATALYST driver/OVERDRIVE | Page:: ( 4 / 29 ) |
CATALYST AI
One new feature ATI has integrated into their display driver is known as CATALYST A.I. CATALYST A.I. is designed to improve your graphics card’s performance without degrading image quality. ATI uses a combination of technologies to accomplish this, with the implementation varying depending on the particular title.
So far, CATALYST A.I. is designed for five games: Splinter Cell, Unreal Tournament 2003/2004, DOOM 3, Race Driver, Price of Persia, Crazy Taxi 3, and Half-Life 2. If one of these applications is detected, CATALYST A.I. kicks into action.
In the case of Unreal Tournament, up to now trilinear filtering was only in use in the first texture stage, all other stages got bilinear filtering. With CATALYST A.I. you can now run trilinear filtering on all texture stages.
[image]
| <% print_image("14"); %> | <% print_image("15"); %> | <% print_image("16"); %> |
In the case of DOOM 3, shader replacement is used to enhance performance. ATI replaced the lookup table for one lighting shader that was used extensively in DOOM 3. According to ATI the new table is mathematically precise, no banding or other artifacts should be introduced by the change, but performance improves by up to 20% on some cards, with the performance improvement varying based on the card.
[image]
| <% print_image("17"); %> | <% print_image("18"); %> | <% print_image("19"); %> |
A.I. serves a much simpler purpose in Splinter Cell. Ever since the game was released, it has lacked support for AA: turning AA on in Splinter Cell with ATI cards results in visual corruption, while NVIDIA cards “appear” to run fine. NVIDIA got around this by simply auto-detecting for Splinter Cell. If SC was launched with AA enabled in the control panel, the driver automatically disabled AA. With CATALYST A.I., ATI does the same thing, if Splinter Cell is running with AA enabled in the control panel, A.I. turns it off. Finally, for Half-Life 2 engine games, texture caching is used to improve performance.
We only received the A.I.-enabled driver over the weekend, so we haven’t had much time to play around with it, but we didn’t notice any artifacts with it enabled. On low-end systems it’s possible that the algorithms found in CATALYST A.I. could put too much of a strain on your CPU, so ATI provides two settings “Low” and “High”. By default, A.I. is set to the “Low” setting, the “High” setting is more aggressive and is designed to improve performance even further.
We stuck with the default setting of low for all of our benchmarks, except in the case of DOOM 3, HL2 VST and CS: Source, where we provided results with all three settings with X700 XT.
[image]
| <% print_image("20"); %> |
Also, in case you were wondering, CATALYST A.I. is only available with CATALYST Control Center. ATI has no plans to integrate it into the regular control panel, so if you want to run with A.I., you’ll have to use CCC.
OVERDRIVE plans
ATI continues to offer OVERDRIVE to those who would like to use it. In case you’re not familiar with OVERDRIVE, it’s a hardware monitoring/overclocking utility that’s built into the control panel, providing dynamic temperature-based overclocking when enabled. While ATI’s plans aren’t written in stone just yet for OVERDRIVE with X700 XT, ATI expects to provide two settings above the default level of 475MHz; overclocking will occur in 7MHz increments.
This means OVERDRIVE support will top out at 489MHz, with the intermediate setting at 482MHz. The X700 XT’s memory on the other hand maxes out at 550MHz (we never saw a middle setting). The only aspect that hasn’t been finalized is the temperature thresholds. In our testing with OVERDRIVE enabled, our X700 XT card remained at the intermediate core setting, with the memory clocked at 550MHz, our core temperature at that setting was measured by OVERDRIVE to be 42 degrees Celsius.
Our testing
We’ve provided benchmark results for up to eight cards, ranging from the mainstream RADEON X600 PRO all the way up to the GeForce 6800. Those of you with RADEON 9600 PROs should look at the X600 PRO results, as your card shares the same clocks, while RADEON 9500 PRO/RADEON 9600 XT users should follow the X600 XT results. Your card mirrors its performance in many situations.
| Test systems | Page:: ( 5 / 29 ) |
System Setup
Intel Pentium 4 3.4GHz “Prescott”
ABIT IC7-G Max II Advance
ABIT AA8-DuraMax
1GB OCZ EL DDR400 Platinum Edition Rev 2
1GB Corsair XMS2 5400 DDR2
ATI X700 XT 128MB
ATI X600 XT 128MB
ATI X600 PRO 128MB
ATI RADEON 9800 XT
ATI RADEON 9800 PRO
ATI RADEON 9700 PRO
Driver version beta CATALYST 4.10
Leadtek WinFast A400 (GeForce 6800)
NVIDIA GeForce 6600 GT reference card
Driver version 65.76
250GB Maxtor Hard Drive Maxline III SATA Hard Drive w/16MB Cache
Windows XP Professional SP1
DirectX 9.0c
Benchmarks
Half-Life 2 Video Stress Test
Lock On: Modern Air Combat (Mig-29 custom demo)
Unreal Tournament 2004 (T3 custom demo)
IL-2 Sturmovik: Forgotten Battles (The Black Death track)
Splinter Cell (FS custom demo)
Tomb Raider: Angel of Darkness (Beyond3D custom demo)
Halo: Combat Evolved (stock benchmark)
Far Cry 1.2 (custom demo for mp_jungle, SM2.0 path used)
DOOM 3 (gameplay custom demo)
| IL-2 Sturmovik: Forgotten Battles | Page:: ( 6 / 29 ) |
IL-2 Sturmovik: FB - OpenGL
| IL-2 Sturmovik: Forgotten Battles 4xAA/4xAF | Page:: ( 7 / 29 ) |
IL-2 Sturmovik: FB - OpenGL
| Lock On: Modern Air Combat | Page:: ( 8 / 29 ) |
Lock On: Modern Air Combat – Direct3D
| Lock On: Modern Air Combat 4xAA/4xAF | Page:: ( 9 / 29 ) |
Lock On: Modern Air Combat – Direct3D
| UT 2004 | Page:: ( 10 / 29 ) |
Unreal Tournament 2004
| Splinter Cell | Page:: ( 11 / 29 ) |
Splinter Cell – Direct3D
| Tomb Raider | Page:: ( 12 / 29 ) |
Tomb Raider – Direct3D
| TR 4xAA | Page:: ( 13 / 29 ) |
Tomb Raider – Direct3D
| Halo | Page:: ( 14 / 29 ) |
Halo – Direct3D
| Far Cry mp_jungle | Page:: ( 15 / 29 ) |
Far Cry – Direct3D
| Far Cry 2xAA | Page:: ( 16 / 29 ) |
Far Cry – Direct3D
| Far Cry Jungle 4xAA | Page:: ( 17 / 29 ) |
Far Cry – Direct3D
| Jungle 2xAA/8xAF | Page:: ( 18 / 29 ) |
Far Cry – Direct3D
| DOOM 3 High Quality 8xAF | Page:: ( 19 / 29 ) |
DOOM 3 – OpenGL
| DOOM 3 High Quality 4xAA/8xAF | Page:: ( 20 / 29 ) |
DOOM 3 – OpenGL
| Video Stress Test | Page:: ( 21 / 29 ) |
Half-Life2 – Direct3D
| HL2 VST 2xAA | Page:: ( 22 / 29 ) |
Half-Life2 – Direct3D
| VST 2xAA/8xAF | Page:: ( 23 / 29 ) |
Half-Life2 – Direct3D
| Counter-Strike: Source beta | Page:: ( 24 / 29 ) |
Half-Life2 – Direct3D
| CS 2xAA/16xAF | Page:: ( 25 / 29 ) |
Half-Life2 – Direct3D
| CS w/4xAA | Page:: ( 26 / 29 ) |
Half-Life2 – Direct3D
| CS 4xAA/4xAF | Page:: ( 27 / 29 ) |
Half-Life2 – Direct3D
| Overclocking | Page:: ( 28 / 29 ) |
Half-Life2 – Direct3D
DOOM – OpenGL
| Conclusion | Page:: ( 29 / 29 ) |
In a matter of weeks, the mainstream graphics card market is about to get one serious performance upgrade in the form of ATI’s RADEON X700 lineup and NVIDIA’s GeForce 6600. These $200 cards are poised to deliver performance comparable to last year’s $500 graphics cards – what a difference a couple of quarters makes! But which card is right for you?
That’s a difficult question to answer right now, as kinks in both cards are still being worked out. The hardware is complete, but the drivers still need a little bit of work. The performance of ATI’s CATALYST Control Center still leaves a lot to be desired, while ForceWare 65 has its fair share of issues that need to get tweaked also.
Looking strictly at the hardware itself, ATI’s use of six vertex units could give X700 a performance advantage in RTS titles such as LOTR: Battle for Middle Earth, but the added units should also provide one additional bonus: better performance for those of you with slower CPUs. The additional vertex units should take more load off your system CPU.
X700 also performed well in Far Cry 1.2, sweeping GeForce 6600 GT in all tests (although we ran both GeForce 6600 and X700 with the default 2.0 path). And if that weren’t enough, ATI’s performance in Far Cry will be getting even better shortly thanks to the addition of 3Dc, which will be coming in Far Cry 1.3.
On the other hand, NVIDIA’s GeForce 6600 GT supports SLI, allowing you to connect two GeForce 6600 cards for nearly double the performance. Shader Model 3.0 support also ensures that GeForce 6600 GT is designed for the next generation of games. Because of these factors, if you plan on keeping your card for the long haul, GeForce 6600 GT would probably be the better choice.
It’s a tough call at this point, because so much is still up in the air. We want to see how both cards perform with Half-Life 2, and follow-up titles based on the Source engine, we’ve already seen Counter-Strike: Source performance take a huge nosedive in the past few weeks, the game is much more CPU-bound now than it was when it was first released last month. We also want to see how ATI’s CATALYST A.I. initiative matures. We don’t expect ATI to take the crown in DOOM 3 anytime soon, but if they can close the gap a little more, gamers may not complain too much.
In all honesty, either one of these cards will likely satisfy the needs of the mainstream consumer. As you saw in our benchmarks, the X700 XT triples the performance of its predecessor, X600 XT in some cases. Who can complain about that?