Dawn Of War 2 Retribution Graphics Mod



Dawn of War II - Retribution 3 beta on maxed out graphics (ultra graphics). System spec: Intel(R) Core(TM) i5-2500K CPU @ 3.30GHz (4 CPUs), 3.6GHz. Elite Mod (Elite for short) is a multiplayer mod for Dawn of War II: Retribution and features a multitude of changes including: Re-balancing of all factions with tweaks and changes to units, wargear, abilities and globals. New units and abilities for all factions; Changes to some of the core gameplay mechanics for multiplayer. Dawn of war 2 Retribution Mods Hey im looking for a good mod for the game and i cant find any i see alot of Elite mod but not much else and i dont like the look of it much either. Showing 1 - 1 of 1 comments.

ABOUT Dawn of War 2 Destroyer 40k Mod Destroyer 40k Mod for DoW 2. Destroyer 40k 2.0 is COMPLETE, and can be downloaded here. Program filessteamsteamappscommondow ii retributiongameassets/mods) if there is not a mods fold create one and put the 2 folders in it. Destroyer 40k Mod. POPULAR DISCUSSIONS. Better Easy Repainting Mod Mod Posted over 1 year ago; 21 downloads; This Warhammer 40,000: Dawn of War II - Retruibution mod contains all Elite Mod colours for the main game and unlocks all badges for all factions/races.

Warhammer 40,000: Dawn of War II – Retribution
Developer(s)Relic Entertainment, Feral Interactive(Mac OS and Linux)
Publisher(s)THQ (former)
Sega (current)
Producer(s)Jeff Lydell
Designer(s)Daniel Kading
Programmer(s)Beau Brennen
Stephen North
Artist(s)Michael J. Moore
Composer(s)Doyle W. Donehoo
SeriesWarhammer 40,000
EngineEssence Engine 2.0
Platform(s)Microsoft Windows, Mac OS, Linux
Release
Genre(s)Tactical role-playing game, real-time tactics, real-time strategy
Mode(s)Single-player, multiplayer

Warhammer 40,000: Dawn of War II – Retribution is the stand-alone second expansion to Warhammer 40,000: Dawn of War II, part of the Warhammer 40,000: Dawn of War series of real-time strategyvideo games. Set in Games Workshop's Warhammer 40,000 fictional universe, the single player campaign is playable with multiple races.

Mod

Imperial Guard is introduced as a new faction, and all races including the races from the original game and the first expansion (the Eldar, the Space Marines, Chaos, the Orks and the Tyranids) are playable in single-player.[2]

Gameplay[edit]

Campaign[edit]

Dawn of War II: Retribution offers a campaign for every race, including the Imperial Guard. The campaign takes place across Sub-sector Aurelia, which appeared in the previous two games. The worlds include the jungle world Typhon Primaris, the desert world Calderis, and the hive world Meridian from Dawn of War II; the arctic world Aurelia and the derelict ship Judgment of Carrion from Chaos Rising; and the dead world of Cyrene, mentioned in the original Dawn of War as having been subjected to Exterminatus (complete sterilization of all life on a planet corrupted by Chaos or alien influences) by Blood Ravens Captain Gabriel Angelos.

Setting[edit]

Dawn of War II: Retribution takes place ten years after events of Chaos Rising. Sub-sector Aurelia is now suffering from conflict between Ork pirates called the Freebootaz led by Kaptin Bluddflagg to pillage the sub-sector,[3] the arrival of the Eldar of the Craftworld Alaitoc led by Autarch Kayleth to seek about a prophecy and recovering an ancient artifact,[4] a Tyranid Hive Lord restoring the remnants of the Hive Fleet Leviathan and to the link of the Hive Mind,[5] the Blood Ravens defending the sub-sector led by Captain Apollo Diomedes to hunt down Chaos as well as investigating their Chapter Master, Azariah Kyras, for being corrupted by Chaos,[6] the return of Chaos Space Marines of the Black Legion led by Eliphas the Inheritor to fulfill his promise to Abaddon the Despoiler to annihilate the Blood Ravens,[7] and the newly arriving Imperial Guard of the 8th Cadian Regiment led by Lord General Castor, performing an Exterminatus under Inquisitor Adrastia to investigate for corruption within Sub-sector Aurelia as well as the Blood Ravens' Chapter Master Azariah Kyras. These events may have been caused by Gabriel Angelos' actions (from the original Dawn of War game) when he destroyed the Maledictum, a Chaos stone containing the bound essence of a daemon of Khorne, with the hammer 'God-Splitter'.[8]

Multiplayer[edit]

The Imperial Lord General will be added to the current selection of The Last Stand heroes as well as new environment and new waves of enemies. It can also update all the previous heroes from Dawn of War II and Chaos Rising by importing the Games for Windows account to the Steam Account.

The Last Standalone is a stand-alone version of the Retribution version of The Last Stand. It was released on April 20, 2011 as a separate Steam purchase. Owners of The Last Standalone' receive a discount on the full version of -Dawn of War II: Retribution.[9]

Multiplayer will introduce base building to a small degree, as well as every race getting a Global Ability Advance Unit. Retribution is a stand-alone title and does not require ownership of earlier games in the series to use any of the factions in multiplayer.[10]

Plot[edit]

The player's character arrives on Typhon Primaris engaging and battling an opposing faction and defeating their leader. (Space Marines vs. Chaos, Eldar vs. Orks, and Imperial Guard vs. Tyranids). It is learned that the Imperial Inquisition has deemed the sector beyond redemption, and will be arriving soon to perform Exterminatus on all the inhabitable worlds in the area. Later, the faction leader is given the objective to eliminate Azariah Kyras who intends to use the impending Exterminatus as a sacrifice to Khorne and ascend to daemonhood. The motivation varies depending on the player's faction, for example the Space Marines, Imperial Guard and Eldar wish to oppose Chaos, their ancient evil enemy; the warlike Orks simply want a good fight and to thump the strongest foes they can find; the Tyranid splinter wishes to overrun the sector and summon a new hive fleet to consume all the biomass; the Chaos faction are Kyras' rivals and wish to surpass him. Regardless, it is deemed by the player faction that Kyras must die. The player quickly attempts to secure a means of transport off Typhon, escaping a local cult along the way.

Arriving on Calderis, the player character fights against Kyras' Chaos-corrupted Blood Ravens Space Marines, operating under orders to purge the planet. After destroying a Warp portal on Aurelia, the faction learns of an attack on Meridian ordered by Kyras and arrives there killing the traitor guardsmen and uncovering a transmission from Kyras revealing his location on Typhon.

The player character returns to Typhon Primaris to confront Kyras himself, only to be ambushed by Eldar from Craftworld Biel-Tan. Wary of a ritual they are undergoing, the player's character kills the Eldar there. Following this, Kyras reveals that the Eldar ritual was preventing the Imperial Inquisition fleet from arriving at the sub-sector. The Inquisition fleet arrives, beginning Exterminatus on Typhon Primaris. The player escapes Typhon before the Exterminatus finishes. A cyclonic torpedo reduces Typhon to ash.

Finding themselves on the space hulk (a huge conglomeration of drifting space-borne detritus consisting of many wrecked ships) known as the Judgment of Carrion, the player's characters recover, and find their determination to stop Kyras. It is deduced that he is hiding on Cyrene, as the planet has already undergone Exterminatus decades ago, and therefore the Inquisition will not travel there to perform Exterminatus again. On Cyrene, the player characters launch an attack against a joint alliance of Chaos Space Marines, corrupt Imperial Guardsmen and traitor Blood Ravens by using their most powerful unit against them. Kyras begins to ascend into daemonhood. Gabriel and his 3rd Company launch an attack on Kyras' forces while Gabriel's own command unit confronts the daemon prince himself; however they are defeated by Kyras. The player's faction then launches their own attack, ultimately successfully killing Kyras.

Ending[edit]

After Kyras' death, the ending of the game will depend on which race the players character chose:

  • Chaos – Eliphas allows the Exterminatus to continue, thus sacrificing the sector to Khorne. He is thus granted daemonhood by the Blood God, usurping Kyras' place.
  • Eldar – Ronahn defeats his enemies and recovers the Spirit Stone containing the spirit of his twin sister, the Farseer Taldeer (taken captive by Kyras after being slain), and decides to return to Craftworld Ulthwé with her. This ending is confirmed as canon (as a sort of composite with the Space Marines ending) in Dawn of War III, where Taldeer and Ronahn both return, the former as a Wraithknight.
  • Imperial Guard – Inquisitor Adrastia returns to the Inquisition to suspend the Exterminatus on sub-sector Aurelia, by presenting Kyras' psychic hood as proof that the threat has ended, while Lord General Castor and Sergeant Major Merrick commend each other for their exemplary actions rather backhandedly.
  • Orks – Inquisitor Adrastia attempts to renege the deal between her and Kaptin Bluddflagg with assassination. Unfortunately, Kaptin Bluddflagg catches her off guard and takes her hat, which he wanted in their deal. Following that, he claims the Judgment of Carrion as his new Krooza, and uses it to leave the subsector.
  • Space Marines – Captain Diomedes contacts Inquisitor Adrastia to halt the Exterminatus. The Chapter is then purged of any remaining chaos taint and Gabriel Angelos, after being revived from the brink of death and rebuilt with bionics, is appointed as the new Chapter Master. This appears to be the canonical ending, as corroborated by other Warhammer 40,000 materials: a squad of Blood Ravens appears in Warhammer 40,000: Space Marine, in which their victory over the conflict in Aurelia is mentioned; meanwhile, the exact course of events of the successful Blood Ravens campaign in Retribution is likewise mentioned in some publications (such as Fantasy Flight Games' 'Deathwatch: Honour the Chapter' supplement). In Dawn of War III, the rebuilt Gabriel Angelos returns as Chapter Master of the Blood Ravens and Diomedes was made a Chaplain and Jonah Orion was made chief advisor. Meanwhile, at least a portion of the Eldar ending is considered canon as Ronahn managed to retrieve his sister's spiritstone before the two are waylaid by Kyre during their return trip to Ulthwé.
  • Tyranids – The Hive Tyrant's psychic strength summons a Hive Fleet, that launches a surprise attack and consumes the entire sub-sector, resulting in a 94% casualty rate for the Imperial Guard forces and the complete annihilation of the Blood Ravens, who refused to retreat.

Development[edit]

On September 15, 2010 Relic Entertainment announced that Retribution would be dropping the Games for Windows – Live multiplayer platform in favor of using Steamworks as its primary and only platform.,[11] and the new multiplayer platform does not communicate with the old Live platform. This makes the game entirely stand-alone with all the races included (unlike Chaos Rising, which required the original Dawn of War II to use the original four races in multiplayer). Plot-wise, two playable characters (Cyrus and Tarkus) have been carried over from the original campaigns. A new multiplayer matchmaking service was developed for Retribution. The addition of Steamworks also allows inviting Steam friends directly to multiplayer matches as well as free-to-play multiplayer weekends and a much faster patching process.

There were speculations that the new playable race would be Imperial Guard and/or the Inquisition, due to the Inquisition's triple-lined 'I' used in the word 'Retribution' in the game trailer;[12] in addition, the expansion's wishlist icon in the Steam system features a female Witch Hunter Inquisitor. On December 21, 2010, the German PC gaming magazine Gamestar revealed the new race would be the Imperial Guard.

A multiplayer beta of the game was launched on Steam on February 1, 2011 and ended on February 25.

Release[edit]

Specific race packs were released for each of the six races within the campaign, containing special items for that race in the campaign of Retribution, such as armour, weapons and accessories. The Ork race pack is a Steam exclusive[2] and Tyranid pack a THQ online store exclusive,[13] although all were made available as DLC after release, and all come in the Dawn of War II: Retribution Collector's Edition retail box.[14]

Wargear packs were also released for The Last Stand multiplayer mode. Each pack added new wargear to a specific hero class.[15]

There were also two DLC packs released adding new sub-races to use in multiplayer game with unique models and color schemes for units: Dark Angels Space Marine chapter[16] and Eldar Craftworld Ulthwe.[17] Similar DLC pack adding the Ultramarines Space Marine chapter was released on August 16, 2011.[18][19]

The Tau Crisis Suit Commander is another hero for Last Stand mode available by download.

Reception[edit]

Aggregate scores
AggregatorScore
GameRankings81.65%[20]
Metacritic82%[21]

The game received generally positive reviews upon release. It received an aggregated score of 81.65% on GameRankings based on 36 reviews and 82/100 on Metacritic based on 52 reviews.[20][21]

Sequel[edit]

A few months after Retribution's release, Relic revealed that work had begun on a sequel.[22]

Shortly before THQ's filing for bankruptcy in December 2012, Sega secured a licensing deal agreement with Games Workshop.[23] Sega then purchased Relic Entertainment from THQ in January 2013.[24] This made a future of the sequel uncertain and, in May 2014, Relic would not comment on its status.[25]

However, in September 2015, Relic opened up a new web page featuring Dawn of War III.[26] In October 2015, PC Games News wrote that they expected Dawn of War III to be released in 2016.[27]Dawn of War III was displayed at E3 2016[28] and was released on April 27, 2017.

References[edit]

  1. ^'Dawn of War II: Retribution preorder bonuses, CE detailed'. New Game Network. January 6, 2011. Retrieved 2011-01-06.
  2. ^ abDawn of War II: Retribution on steam
  3. ^Retribution - Ork Campaign Heroes
  4. ^Eldar Campaign Heroes
  5. ^Tyranid Campaign Hero
  6. ^Space Marine Campaign Heroes
  7. ^Chaos Campaign Heroes
  8. ^Imperial Guard Campaign Heroes
  9. ^[1], Dawn of War 2 Community.
  10. ^Dawn of War II: Retribution on Steamworks
  11. ^As a result of this, the campaigns do not import or continue directly from Dawn of War II and Dawn of War II: Chaos Rising
  12. ^Retribution Trailer , 'I' seen at 0:51 seconds into the video
  13. ^Dawn of War II: Retribution pre-order page on the THQ estore
  14. ^PC Gamer article revealing Dawn of War II: Retribution Collector's Edition
  15. ^The Last Stand Wargear Packs - Dawn of War II Community
  16. ^Dark Angels Invade Retribution - Dawn of War II Community
  17. ^Eldar Ulthwé Pack Now Available - Dawn of War II Community
  18. ^Ultramarines DLC Preview - Dawn of War II Community
  19. ^Warhammer 40,000: Dawn of War II Ultramarines Pack at Steam
  20. ^ abWarhammer 40,000: Dawn of War II - Retribution
  21. ^ abWarhammer 40,000: Dawn of War II - Retribution
  22. ^Dawn of War 3 will let you 'build your own custom mega army' - PC Gamer
  23. ^SEGA secures Games Workshop licensing deal - VideoGamer
  24. ^SEGA has purchased Relic from THQ, now owns Warhammer and Company of Heroes - Escapist Magazine
  25. ^Relic's five-year plan for Company of Heroes 2 - Eurogamer
  26. ^Sega Registers Warhammer 40,000: Dawn of War 3 Domain - Gamespot
  27. ^Warhammer 40,000: Dawn of War 3: release date - PCGamesN
  28. ^E3 2016: New Dawn of War 3 gameplay trailer - Gamezone

External links[edit]

  • Warhammer 40,000: Dawn of War II – Retribution at MobyGames
Retrieved from 'https://en.wikipedia.org/w/index.php?title=Warhammer_40,000:_Dawn_of_War_II_–_Retribution&oldid=995932412'

by Relic Entertainment and Intel Corporation

Download Article

Dawn Of War 2 Retribution Graphics Mod

Download Warhammer* 40,000*: Dawn of War II*: Retribution* Performance case study [PDF 1.8MB]

Abstract

During the development of Warhammer* 40,000*: Dawn of War* II: Retribution*, engineers from Relic Entertainment and Intel Corporation analyzed the game for performance on mainstream PCs. We used a PC with a 2nd generation Intel® Core™ processor with Intel® HD Graphics 3000. At the time, Retribution was GPU-bound on that system. This case study describes the steps we took to find and speed up the most time-consuming areas of the game. Looking at the GPU workload, the pixel shaders for the terrain took more time than any other parts of the frame, so we experimented to find the best ways to speed up the terrain. We tried several techniques: clamping the maximum MIP level, removing ‘texkill’, eliminating overdraw of the pixels under the game’s HUD, and reducing texture sampling. After studying the terrain shaders in detail, they were improved by adding a threshold test to avoid certain texture loads. After this change, Retribution now runs 1.4x faster at the target resolution and quality settings on the test machine.

Overview

As they were developing Retribution, Relic engineers wanted the game to run great on Intel graphics hardware. Older generations of this hardware have not been up to the task, but the latest generation can run games like Retribution quite well. Relic and Intel engineers studied the game’s performance and worked together to optimize the game. After optimization, Retribution looks and plays great on Intel graphics hardware, and we hope you enjoy playing it!

This case study will help you better understand how to find and eliminate similar issues in your games. For general guidance on tuning games on Intel graphics parts, see the http://software.intel.com/en-us/articles/intel-graphics-developers-guides. In this case study, we used the Intel® Graphics Performance Analyzers (Intel® GPA) to measure and analyze Retribution in order to set the stage for tuning the game. Intel GPA also let us try many graphical experiments in real time before making any changes to the game code. If you’re not familiar with GPA, you’ll get a sense of it in this case study, or you can get more info from the Intel GPA site.

During development, the most expensive part of each frame, by far, was rendering the terrain. In that part of the frame, time was mostly being spent in pixel shaders, which were sampling and blending a large amount of textures. The pixel shaders were changed to be more selective, so that they only load/blend textures above a certain threshold. After implementing that change in the pixel shaders, the game runs much better and looks great.

We tried several other experiments to speed up the game, but the resulting improvements were small in comparison with this one change.

Picking a representative scene

Before optimizing, we needed to know where to look (any time you optimize, you need a well-defined “workload” to measure ). To get a quick sense of the game’s performance under various conditions, we brought up different scenes from the campaigns and did a few simple experiments while watching the frame rate. For this, a simple frame rate monitor like Fraps works great, or you can use the Intel® GPA System Analyzer Heads-up Display (HUD) to watch the frame rate (this HUD is added to your game when you profile in real time). The idea here, before doing any analysis, is to get a quick sense of where your game shows average and worst-case performance, so you can study those cases.

For games that have single-player campaigns and multiplayer modes, campaigns are generally easier to benchmark. The Retribution campaigns have similar performance with multiplayer, so we started with the campaigns for simplicity. If the two were very different, we would have analyzed both.

We tried a variety of resolutions, but the most interesting results were at 1280x800. This was the resolution range where some of the quality settings ran fine and others ran too slow, so it gave us a great starting point to understand performance while the game was in development. Figure 1, Figure 2, and Figure 3 show the performance at 1280x800 resolution, at the predefined “Low”, “Medium”, and “High” quality settings (see the game’s graphics options). While the game does let you zoom the camera in and out from the default camera height, all tests were done at the default zoom level.

Figure 1: Prologue

Figure 2: Ork Campaign – Typhon Primaris

Figure 3: Space Marine Campaign – Calderis

Of those scenes, the Ork scene (Figure 2) had the lowest frame rates.

Retribution also includes a built-in Performance Test that shows a representative battle. It’s a great way to simulate a multiplayer game workload in a repeatable way. If you’re developing a multiplayer game, you should consider adding a built-in performance test like this; it’s great to give your customers a quick way to see the performance of their configuration settings. It’s also a very handy way to benchmark. (To find the Performance Test in Retribution, look in Graphics Options.)

Figure 4 shows the results of running the Performance Test at “Low”, “Medium”, and “High” quality settings. Here we didn’t need to use Fraps to collect the data. The game reports average, min, and max results with the average shown here.

Figure 4: Performance Test

Even though we were just picking a workload for benchmarking, this was a good time to start thinking about what we saw. These tests suggested a number of possible contributors to frame rate, but the number of characters had no large effect on the frame rate. Later experiments would help us narrow it down. At this point, our working theory was that the terrain was affecting the frame rate more than other elements of the scene were. There are other tools in the HUD to help with a rough assessment of this. See the discussion of wireframe below.

Looking across all of the campaign scenes and the Performance Test, we saw good performance at Low quality settings. Although the game was playable at 15 FPS, both Medium and High quality left some room for improvement. We decided to focus our attention on Medium quality first. The Performance Test and the Ork scene had about the same average frame rate, which was slower than the other two cases. Of the two, it was easy to capture nearly-identical frames each time we brought up the Ork campaign. This let us easily make changes and consistently test them. We expected that performance improvements we made for the Ork scene would improve the Performance Test as well (and we confirmed that later).

So, the rest of this analysis was done using the Ork scene shown above, at 1280x800 resolution and Medium quality, unless noted.

Major bottleneck: On the CPU or GPU?

With a workload to study, it was time to locate the major bottleneck. If the game was mostly GPU bound, we could speed up the game by making graphical changes.

Intel GPA can install a HUD that runs on top of your game. Figure 5 shows it running on Retribution. This HUD shows real-time performance metrics and lets you take different kinds of captures of your game as it runs. The HUD also includes some simple tests so you can check in real time if your game has a GPU bottleneck. These tests (known as overrides) let you change aspects of how the graphics pipeline works or change in real time how the game presents content to the driver and hardware, so you can measure the impact of those changes. Since you see and measure the changes from each override in real time, you can get a very quick sense of how the game is spending its time.

Figure 5: GPA HUD on top of the Ork scene, showing real-time metrics

The simplest override is the Disable Draw Calls override. With this override, all draw calls into the graphics driver are short-circuited, so they return immediately without doing anything. Since draw calls are discarded, the game display goes black while the override is in use. This does not disturb the rest of the rendering, so the game continues to run although we can’t see anything. The relative change in frame rate shows whether the draw calls are consuming a small or large amount of the frame time.

We also have the Null Hardware override. When Null Hardware is used, the driver processes all calls as usual, but simply discards all calls and data as they would normally be provided to the hardware. The Null Hardware override thus includes all the overhead of the driver; the Disable Draw Calls override does not include any driver overhead for draw calls. The Null Hardware override lasts for a few seconds before reverting (you can configure its duration in Properties).

By comparing the performance using these overrides with the original frame rate, you can see how much of each frame was spent in the hardware, in the graphics driver, and in game code plus DirectX* and the operating system.

The HUD also supplies a number of other overrides. Table 1 shows the ones we tried, along with their results for this scene.

Table 1: Overrides and their performance

Dawn of war retribution races

When we compared the original frame rate to Disable Draw Calls, we saw a large performance jump. This told us that the draw calls (in both the driver and the hardware) were taking a large portion of the frame time. With this override, the frame rate went from 16 FPS to 105 FPS, a 6.6x speedup! This override shows how your game would run if the graphics driver and hardware were infinitely fast. Without any graphics driver or hardware time, this game ran at 105 FPS. Clearly, this game was bound somewhere in the GPU hardware and/or driver.

Looking at the other results, we saw Null Hardware run at 52 FPS, about a 3.3x speedup. This is about half of the Disable Draw call frame rate, so clearly a large amount of each frame is spent in the graphics driver, but an even larger part of the frame is spent in the hardware.

To better understand how the hardware was getting used in real time, we tried a few other overrides. But what do they show? The 2x2 Textures override shows how the game runs with simple textures (reducing texture load times and texture processing in pixel shaders). The 1x1 Scissor Rect override turns off pixel processing by passing only a single pixel through the pipeline. Finally, the Simple PS override replaces all pixel shaders with a trivial shader that emits a constant color.

These overrides told the rest of the story. Reducing the game to a single pixel gave a frame rate about the same as null hardware. Swapping to a simple pixel shader also gave about the same performance as null hardware. Using simple textures we saw a 1.6x speedup, which is only a modest gain when compared with the Null Hardware result. So, some amount of each frame was dominated by the use of more complex textures, but the majority of the frame time in the hardware was spent processing pixels in the pixel shaders.

Other overrides will sometimes show interesting results at this point, but this combination gave us a clear direction for more analysis. In particular, for other games you might learn much from the Z, culling, and alpha overrides, and from showing overdraw visualization and wireframe.

To confirm that we were bottlenecked in the hardware, we also used GPUView (from the Windows* Performance Toolkit, part of the Windows SDK). When we looked at the CPU queue, we could see that the queue was full all the time. This showed that the hardware always had queued work available for it, so we knew that the game was running faster than the hardware could process.

Figure 6: GPUView shows a full context CPU queue

In Figure 6, the GPU HW queue is at the top, and the context CPU queue is the large, filled region on the bottom. You can see the commands flow down in the queue over time. Here we clicked on one Present packet to highlight it. You can see it progress down the queue until it was processed on the right of the figure.

In this trace, the queue held about 2.5 frames worth of commands at all times (when the queue is this full, you can tell by scrolling through the queue and counting the Present packets). It’s good for a game to pipeline its frames to keep slower graphics hardware from going idle between frames, as long as you’re careful to not run too far ahead. Later experiments confirmed that Retribution pipelined at most three frames in the queue. Any more would have started to add a noticeable UI latency, so this looked like the right amount of queued work.

It was now clear that our bottleneck was in the GPU hardware. Armed with this information, it was almost time to capture a frame for deeper analysis. But first, we could learn more by looking at real-time data in the HUD.

More real-time data: Roughly what were we doing?

The real-time counters displayed by the HUD can give a good sense of how a frame runs. Using Intel GPA on any DirectX graphics hardware, the HUD shows HW and SW counters from the CPU and DirectX. When Intel GPA run on Intel graphics hardware, they also show hardware counters that are specific to Intel graphics hardware.

The HUD can show four counters at a time. You can show different counters by looking at the HUD monitor’s profile settings. Changes take effect immediately.

First, we noted all the metrics that the HUD can display. Then we did some experiments with the overrides in the HUD and captured those values as well to see how the data changed.

The data in Table 2 shows a few areas for investigation, but mostly we confirmed what we found by looking at the frame rate changes from the overrides. A large amount of each frame was spent on the GPU, and most of that was in pixel shaders.

Table 2: Counter detail with overrides

With this information, it was time to capture a single frame, so we could study how we were spending time in the pixel shaders.

Moddb Dawn Of War 2

Looking at a frame: Details emerged

After capturing a frame in the Intel GPA HUD and exiting the game, we used the Intel GPA Frame Analyzer to show the frame with detailed stats about how it was drawn. In Frame Analyzer, the frame is also “live” and can be experimented on, so you can try various changes and see the impact.

Figure 7 shows how Frame Analyzer displays all the calls across the captured frame. Each unit of work (an “erg”) in the frame is drawn at the top, and the rest of the window holds various summary and detail panes. We can display the ergs in various ways to better understand them. The default view is shown here, but it can also be useful to plot GPU breakdown vs. GPU duration. See the dropdowns in the corner for various options.

Dawn of war 2 retribution elite mod

Figure 7: GPA Frame Analyzer shows terrain is the most expensive

First, we looked at the summary information for areas of interest (highlighted below) and then started running experiments to better understand how to speed up the frame.

Table 3 shows several areas that might be making the frame run slow, but the pixel shader time was the single biggest piece of the frame, so we focused our attention there first. Selecting the longest-running erg, we saw from the render target that this erg drew the terrain. Selecting several of the most expensive ergs confirmed that most of the work in each frame was dedicated to rendering the terrain.

Table 3: Frame Analyzer summary shows a few areas for further study

Frame budget: Confirming that terrain was expensive

It can be useful to look at your game’s frame “budget”, to see how much of each frame is spent on each part of its rendering. This lets you decide if the right amount of each frame is devoted to the right geometry and the right look. By selecting different ergs among the longest-running group of ergs, we could see that most of the longest-running ergs were dedicated to rendering the terrain. But how much?

To find out, we manually picked different ergs to see which ones rendered the terrain. Tip: often, terrain work is grouped together, so the terrain ergs are next to each other in the timeline, and can be multiply selected at once. When we knew the first and last ergs that contributed to the terrain (or post-processing, or characters, or whatever), we selected all the interesting ergs to see a summary of the terrain work.

When we selected the terrain ergs, we saw that just over half of the frame time was spent on the terrain. The total frame time was 63.8 msec or 15.6 FPS. Of that, terrain time was 32.6 msec or 51% of the total frame.

Looking closer at the longest-running ergs, we saw that they had the most complex pixel shaders in the game. They tended to have many texture reads, with as many as 36 texture reads per pixel. Each shader sampled up to 8 terrain texture layers (at 4096x512), plus some additional textures.

War

The terrain textures looked like this:

Figure 8: Sample terrain texture

Experiments: Try to speed up the terrain

Now that we had located slow parts of the frame, it was time to try to speed up the frame. In Frame Analyzer we can change many aspects of how a frame is drawn and see the effect immediately.

Experiment: Clamping Max MIP level

With so many textures, we hoped that reducing the MIP levels of the textures might give us a large speedup.

In Frame Analyzer, we selected the ergs in question and manually overrode the MIP level on each of the terrain textures. Here’s what it looked like.

Table 4: Higher MIP levels ran faster but at much lower quality

MIP level 1 still looked OK, but the visual quality beyond that was unacceptable. We found only a modest speedup at MIP level 1. This change might have been useful, but it would not solve the larger performance issue.

Experiment: Eliminating ‘texkill’

In Frame Analyzer, you can edit the pixel shaders directly to try various scenarios. Some of the shaders included use of ‘texkill’, which can sometimes be replaced with faster code. We tried replacing the ‘texkill’s with a fall back path that simply outputs a transparent black pixel. We found no significant performance change; the original code was fine.

Experiment: Eliminating unnecessary texture reads

Looking at the shaders more closely, we found areas that had many texture lookups. In the Terrain Tile shader (PSTileLow), every terrain texture layer is sampled and then blended. The blend uses a number of computed Weightfactors and other parameters. Figure 9 shows the original shader code from PSTileLow.

Dawn Of War 2 Campaign Mods

Figure 9: PSTileLow shader showing layer blending

That was the PSTileLow shader. There was also a higher-fidelity version, PSTileHigh (see Figure 10). That version has similar sampling logic, but it will only sample terrain layers if the Weight is above a certain threshold value. This keeps the shader from working with textures that don’t contribute much to the desired result. Changing the PSTileLow shader to use the same technique gives a more complicated looking shader, and one we hoped might run faster.

Figure 10: PSTileHigh shader blends only above thresholds

The revised shader used tex2Dlod instead of tex2D. We used tex2Dlod since the sample was inside a dynamic branch and tex2D instructions aren’t allowed inside dynamic flow control. The tex2D instructions couldn’t be used here because the GPU always works on 2x2 pixel blocks at a time, and it needed to calculate the derivatives used for the MipMap LOD and anisotropic filtering. To add dynamic flow control to our texture instructions, we needed to directly compute the pixel’s LOD and use that in the explicit tex2Dlod instruction. This made it possible to change the shader so the texture lookups only happened when the layerWeight was above the threshold.

Although Frame Analyzer isn’t a full shader debugger, we can change the shader by hand (dynamically), to test our changes. In this case, when we changed the shader, we saw some significant changes.

Table 5: Threshold check speeds up the frame

Adding the threshold check gave us a major speedup on the slowest part of the frame and a significant speedup to the whole frame.

To understand this change even better, we wondered if reducing the memory reads per pixel would give a speedup. So we looked at the MIP level of the textures in question to give us smaller textures to load. Table 6shows the results, highlighting the significant differences.

Table 6: Adding threshold check makes shaders run faster

The modified pixel shader with the threshold check did run faster. As we expected, the overall frame ran faster, and we showed a significant reduction in GPU duration and specifically in PS duration. There is a reasonable drop in post-filter texels, as we would expect, and a significant drop in GPU texture reads. While clamping the MIP level to 1 also ran slightly faster, that gain was small in comparison. Also, the frame looked nearly identical to the original frame, so changing this shader had no negative effects.

So, the MIP change was not worthwhile, but the pixel shader edits were a major speedup. These results were measured when the few most expensive shaders were changed by hand. When all similar shaders were changed across the game, the frame rate went up to 24 FPS.

Experiment: Avoiding overdraw of pixels under HUDs

Looking at the game, there are two significant HUD areas: the mini-map and the unit info/command area. Both areas contain irregular opaque objects that get rendered on top of everything else on the screen. They’re highlighted in Figure 11.

Figure 11: These HUDs cover ~10% of the screen real estate

All pixels rendered beneath the HUD are simply overdrawn when the HUD is rendered. They take up roughly 10% of the screen real estate. What if there was a way to avoid that overdraw by not rendering those pixels in the first place?

It sounds simple to avoid calculating the wasted pixels by reordering, so the HUD is rendered first. That way, we could let the graphics hardware’s early-Z detection save us from calculating any pixel beneath the HUD. This was not so simple here because the game uses multiple render targets. The game has render passes for color, lighting, and post processing. It wouldn’t make sense to draw the HUD first, because then the HUD would be subject to post processing passes.

It is possible to block out the HUD regions in the earlier render targets using simple quads with a depth of zero. This could be as simple as issuing one extra Clear call right after clearing the surface at the start of the frame. This Clear would set the depth buffer to zero for the rectangular parts of the frame that would eventually hold the HUD. When the Intel engineers were considering this, they did not have access to the source code to make this change. Instead, an internal development tool was used to block out these rectangles in all the overdrawn render targets. This increased the frame rate from 24 to 27 FPS, for a 1.1x speedup. Because of the added complexity of in-game toggles for the HUD and varying HUD locations (e.g., between letterbox and widescreen modes), implementing this would have required significant engineering as well as content pipeline changes, so this change was postponed. There are also other HUDs that might have theoretically yielded a still bigger speedup, but they’re very small and would have had minimal impact.

Results in the final game

After incorporating the shader threshold changes discussed above, the game runs much faster. Before optimization, the Ork scene ran at 17 FPS at 1280x800 and Medium quality. Running on the latest graphics drivers, the released game runs at 24 FPS, or a 1.4x speedup. The same change also yields a similar speedup at low quality and similar speedups on other levels and on the game’s built-in Performance Test.

Now, when we examine a frame, we see much less time spent on the terrain. Figure 12 shows the frame capture with the terrain ergs highlighted. From the highlighted parts of the frame preview, you can see that this is terrain.

Figure 12: Terrain now takes up a smaller part of the frame

Looking at the same view but displaying the ergs with GPU duration and GPU breakdown, it is even more obvious. In Figure 13 we can see that the terrain takes up about 26% of the frame, which is a big improvement. Before these changes, terrain had taken over half of the frame. This confirms that the single biggest bottleneck on the GPU has been optimized.

Figure 13: Terrain shown with GPU duration and breakdown

Looking at this another way, how much did the CPU and GPU workload balance change from optimization? The HUD overrides give us a fast way to tell.

Table 7: Overrides before and after changes

When we run with draw calls disabled, the game runs at about the same speed as before. This shows that the performance of the game on the CPU is largely unchanged. The actual FPS, though, tells the whole story; it’s significantly faster due to speeding up the slowest terrain pixel shaders.

Conclusion

This case study shows some useful techniques for finding graphics bottlenecks and making and measuring experimental changes. There are more techniques that can be used. Retribution did not have this issue, but if your game has intermittent drops in frame rate, you can set up triggers from the Intel GPA Monitor to capture the slow frames. If you’re not sure what the data means, the newest versions of Intel GPA include a feature called Advisor that can make recommendations based on your measurements. Advisor was first released as a Beta feature in version 4.3.

We hope these techniques are useful for you on your project! You can learn more about Intel GPA and download it from www.intel.com/software/gpa.

About the authors

Paul Lindberg is a Senior Software Engineer in Developer Relations at Intel. He helps game developers all over the world to ship kick-ass games for the PC.

Doraisamy Ganeshkumar is a Senior Software Engineer on the Intel Developer Relations team. He helps PC game developers optimize pre-launch titles using the latest Intel technologies. His current focus is to help provide the best Out of Box gaming experience for PC gamers.

Special thanks to Juancho Buchanan and Karl Schmidt from Relic and Mark DeLoura and Morten Haugaard from THQ for all their work on performance improvements to the game and contributions to this case study.

*Other names and brands may be claimed as the property of others.