 | |||
|
|
|
|
|
 |
DirectX 10 Performance Update: Is DX10 Really Worth It? September 28, 2007 Brandon Sandman Bell |
|
 |
|
| |
Summary: Up to this point, DirectX 10 hasn't really lived up to its next-gen hype, but recent game patches and driver updates have been designed to improve performance and image quality in newer DX10 games. In this article we go over what has changed, as well as introduce a brand new jaw-dropping DX10 benchmark. Is DX10 finally worth it? Read the article, check out the screens, and then you decide!
 Introduction | Page:: ( 1 / 7 ) |
Unlike previous DirectX launches where new games using the technology often trailed the introduction of the new API by a year or more, the transition from DirectX 9 to DirectX 10 has been a swift one so far: already a trio of DirectX 10 games have been released (Lost Planet, World in Conflict, and BioShock), while a pair of games have been patched to take advantage of DX10 – Company of Heroes DX10 patch came in May, while Call of Juarez was patched and released for the US market with DX10 support in June. Other games we can look forward to later this year include Age of Conan, Crysis, EVE: Online, Flight Simulator X (Patch), Gears of War PC, Hellgate: London, and Unreal Tournament 3.
As we’ve mentioned in the past, all of these games use DirectX 10 shaders in different ways. Some games will use DirectX 10 to improve performance, while others will use DX10 shaders to improve graphics. Some games will undoubtedly even combine a mixture of the two. Crysis, for example, is expected to try and do both, while Epic’s DX10 optimizations are focused mainly on improving performance (although based on this interview it also looks like AA support and higher detail textures will be DX10 exclusive features). Because developers are all using DX10 differently, we’ve tried to do our best documenting how the various games perform and use DX10, with the BioShock and World in Conflict series of performance articles being our most recent example just last month.
New patches bring updates to existing DX10 games…
But the first wave of DX10 benchmarks are outdated thanks to recent driver updates from AMD and NVIDIA, and in the case of Lost Planet, a new patch was released on August 16th that delivered numerous enhancements thanks to the use of the geometry shader. One notable addition that the 8/16 patch provides is a new DX10 fur shader. We’ll let these screenshots do all the talking:
[image]
| <% print_image("01"); %> | <% print_image("02"); %> |
With the new patch Capcom is also using DX10’s geometry shader to improve motion blur:
[image]
| <% print_image("03"); %> | <% print_image("04"); %> |
This technique uses geometry shaders as well as vertex texture fetch to improve image quality.
The game also supposedly boasts improved depth of field, although we couldn’t find any differences offhand. DX10 should also offer softer shadows, although again, we couldn’t find any great examples of this in-game. We took dozens of screenshots and while we did note differences it was hard to attribute them directly to DX10. With the latest Lost Planet patch, DX10 performance is also closer to the DX9 path in the game.
World in Conflict is another DX10 title we’ve tested in the past. Earlier this month the final version of the game was released, with the DX10 path in the final code of the game delivering improved performance over the beta build we tested last month. We also decided to throw Company of Heroes into this updated DX10 article since we didn’t publish a dedicated Company of Heroes DX10 performance article when that game was patched to support DX10 back in May (at the time the CoH patch was released we were preparing to board a Boeing 777 for Computex in Taiwan). Technically, little has changed with the game’s DX10 path, but we wanted to test the game anyway with the latest drivers from AMD and NVIDIA, as both companies have had time to optimize their driver for the DX10 path in this game.
[image]
| <% print_image("05"); %> | <% print_image("06"); %> | <% print_image("07"); %> |
…and a new DX10 benchmark arrives
We’ve also included a new game into this article, PT Boats: Knights of the Sea. This game is currently under development by developer Akella, creators of Sea Dogs and Pirates of the Carribean. In the words of the developer, PT Boats: Knights of the Sea is a “naval action simulator that places gamers in charge of a mosquito fleet of the Allied Forces, Russia or Germany during the height of World War II”. The game offers a strategic element in tactical mode, or if you prefer the hands-on approach you can take command of any crewmember among your fleet of ships. As you advance in rank, you’ll get more boats and warships under your command, while the game’s multiplayer component allows each player to take on various positions aboard the ship where you’ll be fighting online against another rival team.
The game supports DX10 and AGEIA’s PhysX physics processor, and graphically sports some of the best graphics we’ve seen in a naval simulator.
[image]
| <% print_image("08"); %> | <% print_image("09"); %> | <% print_image("10"); %> |
PT Boats: Knights of the Sea uses DX10 shaders for advanced effects like smoke and water. You can see the amazing reflections in this batch of screenshots:
[image]
| <% print_image("11"); %> | <% print_image("12"); %> | <% print_image("13"); %> |
| <% print_image("14"); %> | <% print_image("15"); %> |
Here are some more PT Boats: Knights of the Sea screenshots we captured from the DX10 benchmark:
[image]
| <% print_image("16"); %> | <% print_image("17"); %> |
| <% print_image("18"); %> | <% print_image("19"); %> |
| <% print_image("20"); %> | <% print_image("21"); %> | <% print_image("22"); %> |
You can really see the lighting/shadowing difference between DX9/10 in this pair of screenshots:
[image]
| <% print_image("23"); %> | <% print_image("24"); %> |
The DX10 benchmark for PT Boats: Knights of the Sea was released last Friday and can be downloaded here. The DX10 benchmark weighs in as a 256MB download and requires the installation of AGEIA PhysX (don’t worry, it runs fine if you don’t have a PhysX PPU).
| System Setup | Page:: ( 2 / 7 ) |
Intel Core 2 Extreme X6800
EVGA nForce 680i SLI motherboard
ASUS P5W DH Deluxe (for Radeon cards)
2GB Corsair TWIN2X2048-6400C4
ATI Radeon HD 2900 XT 512MB
Diamond Viper Radeon HD 2900 XT 1GB
Catalyst 7.9
NVIDIA GeForce 8800 Ultra
NVIDIA GeForce 8800 GTX
NVIDIA GeForce 8800 GTS 640MB
NVIDIA GeForce 8800 GTS 320MB
ForceWare 163.71
300GB Western Digital Caviar SE
Windows Vista Ultimate 64-bit
Benchmarks
Lost Planet w/ 8/16 patch
World in Conflict
Company of Heroes 1.71
PT Boats: Knights of the Sea DX10 benchmark
| Lost Planet | Page:: ( 3 / 7 ) |
Notes
While the performance of Lost Planet’s DX10 codepath has improved in comparison to DX9, the game still runs a bit faster under DX9 for both AMD and NVIDIA’s hardware. The GeForce 8800 GTX for instance ran 5% faster under DX9 compared to DX10 at 1600x1200, although the actual difference was just a little over 1 fps. SLI and CrossFire setups also scale well in this game under both APIs.
AMD’s Radeon cards clearly take a backseat to the GeForce boards in this game under the DX10 path, although under DX9 the 2900 XT runs neck-and-neck with the GeForce 8800 GTS 640MB.
| Company of Heroes | Page:: ( 4 / 7 ) |
| Company of Heroes Performance 1600x1200x32 | ||
| Card | Min FPS | Max FPS |
| GeForce 8800 Ultra DX10 | 10.7 | 62.9 |
| GeForce 8800 Ultra DX9 | 38.6 | 151 |
| GeForce 8800 GTX SLI DX10 | 19.6 | 59.9 |
| GeForce 8800 GTX SLI DX9 | 69.7 | 273.4 |
| GeForce 8800 GTX DX10 | 10.7 | 60.2 |
| GeForce 8800 GTX DX9 | 31.9 | 139.5 |
| GeForce 8800 GTS 640MB SLI DX10 | 16 | 54.4 |
| GeForce 8800 GTS 640MB SLI DX9 | 63.6 | 256 |
| GeForce 8800 GTS 640MB DX10 | 8.6 | 55.7 |
| GeForce 8800 GTS 640MB DX9 | 20.7 | 62.1 |
| GeForce 8800 GTS 320MB SLI DX10 | 11.4 | 56.4 |
| GeForce 8800 GTS 320MB SLI DX9 | 13.3 | 62.5 |
| GeForce 8800 GTS 320MB DX10 | 9.9 | 53.6 |
| GeForce 8800 GTS 320MB DX9 | 24 | 60.3 |
| Radeon HD 2900 XT CrossFire DX10 | 2.9 | 49 |
| Radeon HD 2900 XT CrossFire DX9 | 69.1 | 226 |
| Radeon HD 2900 XT DX10 | 7.5 | 50 |
| Radeon HD 2900 XT DX9 | 17 | 141.5 |
| Diamond 2900 XT 1GB DX10 | 3.4 | 46.8 |
| Diamond 2900 XT 1GB DX9 | 19.8 | 76.4 |
| ||
Notes
Company of Heroes was the first game to be patched to support DX10, with the 1.7 patch being delivered at the end of May. The game’s DX10 path offers improved lighting, grass, and additional litter objects – random objects such as rocks are strewn across the environment. We’re using the game’s built-in benchmark here.
As you can see, performance takes a significant hit in CoH once DX10 is enabled. For single-card configs, DX10 performance is roughly half that of DX9, while SLI systems see a performance hit of 4X. The GeForce 8800 GTS 320MB card doesn’t have enough memory to run the game’s ultra settings with 4xAA at 2560x1600, which is why those results are left blank.
Company of Heroes is another game that shows good SLI scaling, although in the case of the Radeon HD 2900 XT, CrossFire only scales under the DX9 path – DX10 performance is practically identical to the single Radeon 2900 XT card.
The GeForce 8800 GTS 320MB SLI config performs similar to the single GeForce 8800 GTS 320MB. We asked NVIDIA about this and they think the memory slowdown issue is what’s causing this. Turning the texture setting down to medium resolved this. Hopefully this is an issue NVIDIA can resolve in a future driver.
The Radeon HD 2900 XT cards continue to perform poorly under DX10 in this title, although you can see the 2900 XT pulls ahead of the GTS in this game under DX9, falling just 5% shy of the GeForce 8800 GTX at 1600x1200.
| World in Conflict | Page:: ( 5 / 7 ) |
| World In Conflict Performance 1600x1200x32 | ||
| Card | Min FPS | Max FPS |
| GeForce 8800 Ultra DX10 | 17 | 79 |
| GeForce 8800 Ultra DX9 | 19 | 85 |
| GeForce 8800 GTX SLI DX10 | 17 | 97 |
| GeForce 8800 GTX SLI DX9 | 16 | 86 |
| GeForce 8800 GTX DX10 | 17 | 70 |
| GeForce 8800 GTX DX9 | 16 | 81 |
| GeForce 8800 GTS 640MB SLI DX10 | 13 | 63 |
| GeForce 8800 GTS 640MB SLI DX9 | 13 | 78 |
| GeForce 8800 GTS 640MB DX10 | 11 | 59 |
| GeForce 8800 GTS 640MB DX9 | 12 | 72 |
| GeForce 8800 GTS 320MB SLI DX10 | 9 | 57 |
| GeForce 8800 GTS 320MB SLI DX9 | 16 | 72 |
| GeForce 8800 GTS 320MB DX10 | 8 | 44 |
| GeForce 8800 GTS 320MB DX9 | 13 | 74 |
| Radeon HD 2900 XT CrossFire DX10 | 2 | 12 |
| Radeon HD 2900 XT CrossFire DX9 | ||
| Radeon HD 2900 XT DX10 | 5 | 13 |
| Radeon HD 2900 XT DX9 | 15 | 50 |
| Diamond 2900 XT 1GB DX10 | 4 | 14 |
| Diamond 2900 XT 1GB DX9 | 16 | 51 |
| ||
Notes
Since the open beta in August Massive has been working hard on improving the performance of the DX10 code path and it shows. DX9 still runs a little faster than DX10 though. We should mention that this game is extremely CPU-bound, even with a Core 2 Extreme X6800, and as such the SLI setups don’t scale as well as they did in previous games earlier in this article.
You can also see that while AMD’s DX9 performance is competitive to NVIDIA, DX10 performance is nowhere close to being optimal. Under DX9, the GeForce 8800 GTS 640MB outruns the 2900 XT by 10% at 1600x1200. CrossFire doesn’t work in this game either.
The GeForce 8800 GTS 320MB SLI config performs similarly to what we saw with CoH.
| PT Boats | Page:: ( 6 / 7 ) |
| PT Boats Performance 1280x1024 | ||
| Card | Max FPS | Min FPS |
| GeForce 8800 Ultra | 17.8 | 38 |
| Geforce 8800 GTX SLI | 18.3 | 63.9 |
| GeForce 8800 GTX | 16.2 | 33.1 |
| GeForce 8800 GTS 640MB SLI | 14.6 | 43.9 |
| GeForce 8800 GTS 640MB | 11.7 | 23.6 |
| ||
Notes
PT Boats: Knights of the Sea really pushed the systems we tested. As such we tested from 1280x1024 to 1920x1200. The higher-end GeForce cards perform well in this game, and SLI scales nicely, although the GeForce 8800 GTS 320MB card had problems running this benchmark with high quality settings, as did the Radeon HD 2900 XT. As you can see, performance with this group of cards was in the single digits, even at 1280x1024!
| Conclusion | Page:: ( 7 / 7 ) |
It’s also great to see developers like Capcom tweaking their DX10 code even after their game (Lost Planet) was released. The 8/16 Lost Planet patch incorporates numerous improvements designed to improve performance and image quality. Hopefully other developers will be equally willing to improve their games over time as well.
With all that being said, it’s also clear that DirectX 10 isn’t going to completely change the landscape of gaming. As we said nearly a year ago in our DirectX 10 preview: “Is DX10 as significant an update as going from DX7 to DX8? Definitely not.” In that same article, Tim Sweeney said: “Realistically, DirectX 10 doesn't introduce fundamentally new capabilities, but brings many new features that will enable developers to optimize games more thoroughly and thus deliver incrementally better visuals and better frame rates.”
In other words, DX10 is an evolutionary upgrade over DX9, not revolutionary.
And just how does the latest high-end DX10 hardware stack up in today’s DX10 content? Unfortunately as it stands right now, we aren’t seeing the true potential of AMD’s DX10 hardware due to immature drivers, so we can’t even speculate on how the Radeon HD 2900 XT performs in comparison to the GeForce 8800 from NVIDIA. Quite simply, NVIDIA’s GeForce 8800 line takes the DX10 performance crown unchallenged. And as our tests show, only one DX10 app scales with CrossFire: Lost Planet. This means that AMD’s driver team not only has to tweak their DX10 driver for more performance, but CrossFire needs to be implemented as well if they’re going to mount an effective challenge to NVIDIA’s SLI.
Fortunately AMD knows this and they’ve told us that they’re hard at work on addressing these issues. We should see the first fruits of this in upcoming drivers later this year, hopefully as soon as Catalyst 7.10. We also know AMD has some interesting developments in the works in terms of CrossFire – already AMD has demonstrated 3-way CrossFire on their upcoming RD790 platform, with 4-way CrossFire in the works for that chipset as well. If everything goes according to plan, we should see this debut later this year.
Of course, NVIDIA won’t be resting on their laurels. Their nForce 680i chipset has boasted 3 PCI Express Graphics slots for quite some time now, and the second SLI connector present on their GeForce 8800 GTX/Ultra has been sitting there untapped for nearly a year now. It has also been an awful long time since we’ve seen a new GPU from NVIDIA, a company which is well known for establishing the 6-month product cycle in desktop graphics.