Half-Life 2 Performance Preview: The Graphics Hardware Squeezer

The game that the gaming community has been waiting about half a decade for is finally here, and now everyone can enjoy the title many call “the most anticipated game ever”. But in order to experience Half-Life 2 fully, you will need the hardware the game demands. We can tell you for sure that the game does need a powerful central processing unit. But what about graphics cards? Find it out right now, with us!
UPDATE: Adding more game demos and results.

by Alexey Stepin , Yaroslav Lyssenko , Anton Shilov
11/18/2004 | 02:15 PM

UPDATE, November 21, 2004: Adding more game demos and results, adding more graphics cards’ performance numbers, changing demo settings for canals 09 map, adding instant fps graphs for certain hardware, adding tweaked performance for certain ATI RADEON hardware, adding more comments, expanding conclusion section.


Years have passed since the Black Mesa drama ended and the alien invasion was stopped. The tight-lipped Gordon Freeman was hired by a secret agency and… was suspended until his authorities felt they needed him in a European city code-named City 17, ruled by a scientist with huge ambitions who had brought some beasts down from other planets and deployed armed gangs to impose his law over the land.

Mr. Freeman arrives into City 17’s pretty deserted train station along with a few other people. Just like the others, the hero of the Half-Life 2 game is ushered into a special room where the floor is stained with blood – to be asked a few questions. But unlike the others arriving, Gordon Freeman won’t be asked anything as he meets his old buddy, a guard from the Black Mesa Research Facility. This guy tells Gordon that he had also been sent into City 17 to save the planet from the alien storm once again.


A man sitting in the train station’s canteen warns Gordon Freeman against eating anything there lest he should lose his memory, like all the others in City 17.


People are few in the streets of this godforsaken city, but there are lots of guards, abandoned vehicles, rubbish. After spending a couple of minutes here, you clearly realize that this is a place of misery, horror, unhappiness and fear. Once Gordon gets out of the train station building and enters a house (he can enter only a few buildings, as they are usually either heavily guarded or closed), he finds himself in the midst of a raid of an armed police group who have come to capture inhabitants of the house along with Gordon (you do not know why you are being arrested – because of your accidental appearance in the building or because their raid aims exactly at your arrest). With no weapons, map or plan, Gordon rushes through the roofs and backrooms to avoid the pursuers…

Throughout the game you will have to fight your way through alien beasts, masked gunmen, zombies, etc; you will solve puzzles and collaborate with your allies from among the rebels in the town.

On the whole, you are sure to have a few exciting days playing Half-Life 2, but now let’s turn to some technology-related things concerning the game that Valve Software took more than 5 years to develop.

Game Engine: DirectX 7, DirectX 8, DirectX 9

Five generations of graphics hardware have passed by since Valve started its Half-Life 2 project, and the DirectX API from Microsoft has grown three versions up in the meanwhile. As a result, the graphics engine of the game has been evolving along with the technology. Here is how Valve describes the capabilities of their Source engine in its current implementation depending on the DirectX version.

DirectX 7.0:

DirectX 8.0:

DirectX 8.1:

DirectX 9.0:

As you see, the most eye-pleasing and demanding rendering path of the Source engine is the DirectX 9.0 one that is intended for the latest generations of high-end graphics cards from ATI Technologies and NVIDIA Corp.

Graphics and Physics: Designed to Impress

Even though the specs and numbers can astonish, they cannot replace the actual impressions Half-Life 2 leaves in the player. Once you begin to play, you get absolutely immersed into this world, its atmosphere and obstacles on your way. The whole experience is absolutely fantastic, but some details are not.

The geometry, facial animation and textures are realized excellently: every face looks like real and, unlike in Doom 3, people do not have rectangular heads ;). The weapons, equipment and ammunition are all well worked out by the designers. The levels are huge, just like in Far Cry, and very detailed. However, some minor details are not quite perfect, like the coastline on certain levels. This does not affect the gameplay much, but all such details are made extremely accurately in Far Cry for example.

Physics is obviously a big thing in Half-Life 2: the environment is almost fully interactive and behaves like it would in reality. Unfortunately, you still cannot destroy buildings and thus leave a 100% enemy-free territory behind you.


Half-Life 2 is not a hard game to pass. There are no crowds of monsters running to kill you, in contrast to id’s Doom 3 and this is actually a good aspect of the game as you have time to stop and think. Still, there are enough enemies to keep the adventure interesting. On the other hand, if you really enjoyed the hard battles with the alien beasts in Doom 3 or the masses of soldiers and monsters in Far Cry, you may find Half-Life 2 a bit unexciting.

Note also that the game itself does not have any clearly defined levels, just like the original Half-Life. You are just traveling freely around the gaming world, instead.

Testbeds and Methods

We have been hearing lots of requests from our readers to publish Half-Life 2 test results, but having a pretty small amount of time to accomplish the tests, we have decided to publish the article in two iterations. The first one includes three demos and only high-end and performance-mainstream graphics cards available to date, the second one will include five custom demos and a wide range of graphics cards released in the last couple of years. Later on, we will continue to investigate Half-Life 2 in various other aspects.

It has already become our tradition to use two testbeds to run the benchmarks on.

Testbed 1:

Testbed 2:


As usual, we set the graphics quality of the game to the same level so that the RADEON and the GeForce hardware produced an image of the same quality. We also wanted to check out the influence of NVIDIA’s new driver (ForceWare 67.02) on the performance, but found that the difference between the new version and ForceWare 66.93 was no more than 1-2fps in the three scenes we used in our tests. Considering that the absolute fps rates are about 60-100fps, this difference fits into the measurement error range.

After we expanded the list of graphics cards used for testing, we got the following products that we reviewed:

Brief Investigation of the Image Quality

We took several screenshots using ATI Catalyst 4.11 and 4.12 beta, and NVIDIA ForceWare 66.93 and 67.02. Here they are:


CATALYST 4.12 beta

ForceWare 66.93

ForceWare 67.02

It’s hardly possible to tell the difference between versions 4.11 and 4.12 beta of ATI’s CATALYST driver with a naked eye. The same goes for the two versions of the ForceWare driver, which produce even more similar-looking pictures. But one difference is certain – between graphics cards on ATI’s and NVIDIA’s GPUs. Distant objects look much darker on NVIDIA’s cards (the mountains in the first screenshot and the trees behind the dam in the third). It seems like the NVIDIA GeForce 6x00 cards don’t render the fog at all or make it thinner than the ATI RADEON X800 cards do.

A rendering error is also visible on the third screenshot – part of the dam behind the boulders isn’t visible for the player, but the water surface displays the reflection of the whole dam. This is probably a peculiarity of the game’s engine, but it doesn’t add any realism to the game, that’s certain. Overall, the image is free from obvious blunders with the ATI RADEON X800 XT/PRO as well as with the GeForce 6800 Ultra/GT.

We used five demos for the final version of the article. The demo sequences were recorded on Canals 09, Town 01, D3 C17 12, D2 cost 05 and D3 C17 02 maps of the game. ATI Technologies offered us demos made by ATI’s technical specialists, but we decided to use our own sequences for testing, except the Town 01 sequence, which had been made most professionally. We only renamed the Town 01 demo file to avoid any accusations that ATI used any unfair optimizations for the particular demo.


d1_canals_09, Pure Mode

The Canal map makes a heavy use of reflections and refractions on the water surface. During our demo a helicopter is dropping mines into water, causing alpha-blended sprays and explosions. Additionally, the sun in the scene is also blended using a special pixel shader. Experts say the rendering speed of this scene is primarily limited by the water shader speed, and all the scenes in the game that use the same water shader with refractions and reflections are likely to behave in the same way.

Initially we tested this demo with “reflect world” setting instead of “reflect all” setting in the game; typically the difference between “reflect all” and “reflect world” may be found on levels with water, which is why on levels with no water you will hardly notice any substantial differences. Still, even with “reflect all” enabled, the situation with performance in terms of ranks for different VPUs did not change.

This time the new-generation graphics cards from ATI deliver an excellent performance in all the display modes. Even the 12-pipelined RADEON X800 PRO provides playability, yielding 70 frames per second in 1600x1200 resolution. Alas, the competing solutions from NVIDIA behave worse, and the GeForce 6800 Ultra, the topmost model, is the only member of the NVIDIA team to have a high speed in all the resolutions. The GeForce 6800 GT is fast enough in two first resolutions only, and the GeForce 6800 only shows an acceptable result in 1024x768.

This probably comes as the performance of a graphics card in this scene directly depends on the speed of the water-imitating shader. Cards from ATI Technologies have been traditionally good at processing pixel shaders which are rich in complex math1ematical computations. Added the higher GPU clock rates, they ensure their victory in this test.

Note also that the RADEON 9800 XT graphics card, with a GPU of the previous generation, almost matches the speed of the NVIDIA GeForce 6800 GT and is clearly faster compared to the GeForce 6800 and GeForce 6600 GT; RADEON 9800 PRO. Unfortunately, we did not have RADEON 9600-series in our possession to test them, however, the RADEON 9500 PRO typically performs in line with the RADEON 9600 XT and, based on the scores in pure mode in mid-resolution, it does not seem that with everything set to the max the RADEON 9500- and 9600-series are really rapid performers in the Half-Life 2 Canals 09 map.

Graphics cards of the GeForce FX family seem to be doing fine, but don’t forget that they’re working in the DirectX 8.1 mode and thus do not provide the same image quality as modern DirectX 9.0 graphics cards do.

As for the cards with the PCI Express interface, it’s the same as with the AGP hardware: top products from ATI Technologies are in the lead, and the RADEON X700 XT is as fast as the much more expensive GeForce 6800 GT. The RADEON X600 XT cannot run this game at a comfortable speed even in the lowest resolution.

d1_canals_09, Eye Candy Mode

Only two graphics cards allow playing all resolutions with full-screen antialiasing and anisotropic filtering enabled. They are the ATI RADEON X800 XT Platinum Edition and the RADEON X800 XT. The RADEON X800 PRO is again faster than the GeForce 6800 Ultra and 6800 GT, while the GeForce 6800 is competing with the RADEON 9800 XT. The GeForce FX and the RADEON X600 XT don’t run the game at an acceptable speed in this mode at all.

Probably in 1024x768 resolution the game will run at playable levels with eye candy caps enabled on a number of performance-mainstream graphics cards, including such as the GeForce 6600 GT and RADEON 9800 PRO, however, once we boost the resolution up, performance mainstream segment will not deliver sufficient speed. If you play on mainstream boards like the GeForce FX 5900-series or RADEON 9500-/9600-series or lower, then certainly FSAA and anisotropic filtering in Half-Life 2 is not for you and you will have to find certain compromises.

The results suggest that we really deal with a new-gen game. Half-Life 2 takes and consumes everything the graphics card can yield. Weak cards have nothing to do with this game.

d3_c17_12, Pure Mode

The battle scene on the d3_c17_12 map is probably among the most gorgeous episodes of Half-Life 2: it includes a number of combatants, gunfire, explosions, smoke, rocket trails. In short, this is a perfect example of hot street fighting.

Unfortunately, this demo is pretty much limited by the CPU, and some graphics cards can’t show their best. Still, such demos reflect the actual performance of the actual game, which is important for those who plan to play Half-Life 2.

Almost all the participating cards have the same results, and this indicates the dependence of this scene on the CPU performance. Some older graphics cards like the RADEON 9500-/9600-series, RADEON X600 XT, the GeForce FX 5700 Ultra, GeForce 6600, GeForce PCX 5750 cards are the only to fall out of the group, and only in 1600x1200 resolution, to tell you the truth..

d3_c17_12, Eye Candy Mode

In the eye candy mode, the graphics cards don’t differ among themselves in 1024x768 (save for the products based on RV350/RV360 as well as NV36 chips), but the leadership of the RADEON X800 family becomes apparent since 1280x1024 resolution. This tendency keeps on in the higher resolution, too. The GeForce FX 5950 Ultra outperforms the GeForce 6600 GT, but the latter is honestly working in the DirectX 9.0 mode, so it would be incorrect to compare the two cards.

Given significant dependence on the CPU in this particular demo, ATI’s RADEON X800 leadership over the GeForce 6800-series is not really substantial, in our point of view; however, it is pretty evident that the RADEON X800 family do the job better than the RADEON 9800-series and GeForce 6800-series. Currently the difference in performance is not that large, but once faster processors come out, the performance gap between different graphics architectures may become more apparent.

d1_town_01, Pure Mode

The Town 01 demo may be considered as a kind of artificial, because the hero of the game blasts some gas balloons, causing much fire. While this demo is interesting from the technological standpoint, it is absolutely unnecessary to burn everything that is on your way when you’re actually playing.

On the other hand, this demo uses a flashlight, a very helpful device in some scenes of the game, adding load on visual processing units and also heavily activates gravity gun, which adds pressure on the central processing unit.

In the lowest resolution nearly all the graphics cards of the new generation, save for the GeForce 6800, have similar results, with a minor advantage on the part of the RADEON X800 XT PE/X800 XT. This advantage grows up in the higher resolutions, but the GeForce 6800 Ultra looks anyway better than on the Canals 09 level, while the GeForce 6800 GT performs in line with the RADEON X800 PRO. The RADEON 9800 XT and the AGP version of the GeForce 6600 GT share the top place in the mainstream sector with the RADEON 9800 PRO lagging only a bit behind.

Mainstream graphics cards, also belonging to the new generation, namely the GeForce 6600 GT and the RADEON X700 XT, have similar speeds.

The graphics cards with the obsolete 4-pipelined architecture, and the GeForce FX 5700 Ultra with the RADEON X600 XT, only allow for a comfortable play in 1024x768. Once again we should state that Half-Life 2 is a new-generation game with the appropriate hardware requirements.

d1_town_01, Eye Candy Mode

The RADEON X800 XT PE and X800 XT increase their advantage over the GeForce 6800 Ultra and GT in the eye candy mode, so the latter card can only compete with the RADEON X800 PRO. All new-generation graphics cards, including the GeForce 6800, ensure a comfortable play in all the three resolutions.

Despite its venerable age, the RADEON 9800 XT won’t lose to the GeForce 6600 GT and even beats it in high resolutions (where there’s no playability, though). One more good-old graphics card – ATI RADEON 9800 PRO – also yields pretty high framerate in 1024x768.

New age performance-mainstream graphics cards, dubbed RADEON X700 XT, RADEON X700 PRO and the GeForce 6600 GT perform almost similarly with slightest of the advantages on the RADEON X700 XT’s side. Lower speed products obviously cannot run Half-Life 2 with full-scene antialiasing and anisotropic filtering enabled.

d2_coast_05, Car demo, Pure Mode

The “Car demo” we recorded is a pretty simplistic sequence that involves riding a vehicle on a road throwing off the way dozens of abandoned and disabled cars. Finally, the small car comes close to a building flooded by masked armed “policemen” and Gordon Freeman starts to shoot them using his weapon with great deal of success.

Even though the demo is clearly bound to the CPU performance and thus the difference between comparable ATI RADEON and NVIDIA GeForce hardware is not that substantial, even though it does exist and it’s slightly on ATI’s side: the RADEON X800 PRO still outperforms the GeForce 6800 GT in 1600x1200 resolution.

Among the less expensive graphics cards the demo shows pretty nice scalability: the GeForce 6800 is a bit faster than other graphics cards in its price segment, primarily products like RADEON 9800-series, whereas the GeForce 6600 GT demonstrates excellent speed for $199 street price.

d2_coast_05, Car demo, Eye Candy Mode

With full-scene antialiasing and anisotropic filtering activated, the Car demo begins to offer even better scalability, even though its microprocessor dependence is also clear.

Given more efficient full-scene antialiasing pattern of ATI RADEON hardware, the X800-parts increase their lead over the comparable GeForce 6800-series as the resolution grows. Still, we must say that even in 1280x1024 performance mainstream graphics cards with up-to-date architecture and rapid pixel and vertex shaders engines can deliver pretty normal performance. In 1600x1200 ATI’s hardware lead over NVIDIA’s products is being accentuated, even though NVIDIA’s GeForce 6 series provides pretty fine speed for every segment.

d3_c17_02, Dog demo, Pure Mode

The demo recorded on the d3_c17_02 level is a yet another battle on the streets against masked gunmen. But this time Gordon Freeman has another ally – a robot dog that was made for Alex, a friend of Gordon. The dog grabs cars and drops them onto the shooting soldiers. While the demo is yet another CPU-limited, the robot itself has reflective body and there are multiply graphics effects, as in the real game.

It is pretty obvious that this particular level and map are very dependant on the drivers and central processing unit’s performance, which is the RADEON 9800-series deliver the highest speeds in1024x768.1280x1024 resolution puts older graphics cards down and arranges graphics cards according to their market positioning. ATI’s RADEON X800 is ahead of the GeForce 6800, but not significantly.

Performance mainstream graphics cards in 1600x1200, such as the RADEON 9800 XT, GeForce 6600 GT and RADEON X700 XT and RADEON X700 PRO demonstrate pretty much equal performance even on different platforms, which means that speed for those products is mostly limited by rendering power, not CPU.

d3_c17_02, Dog demo, Eye Candy Mode

Full-scene antialiasing does not bring us a lot of surprises: as the resolution rises, RADEON X800 products leave the competing GeForce 6800 family behind, however, as in the majority of cases, their performance is limited mostly by the CPUs. To get high speed in the Half-Life 2 in 1600x1200 resolution with eye candy enabled, you will most definitely need a RADEON X800 XT-series graphics card.

Fast Z Reject: ATI Performance Booster?

Even though ATI RADEON 9800- and X800-hardware performs quite well in Half-Life 2, there may be an additional way to improve the speed of the RADEON graphics processing units. Due to some reason, Valve disabled the so-called Fast Z Reject capability of the RADEON hardware.

According to information from Beyond3D, activation of  Fast Z Reject enables Z fill geometry pass in order to speed up rendering of the full scene rendering pass.

Fortunately, FastZReject can be enabled by a console command 'fastzreject 1'. We tried to activate the feature that is present by default on the GeForce FX and GeForce 6 hardware, but is unavailable, at least for now, on ATI's VPUs. We used two of our demos along with the CATALYST 4.12 beta driver to test whether the Fast Z Reject brings any advantages to the RADEON X800 XT Platinum Edition and RADEON X800 PRO.

Well, at least in our conditions the Fast Z Reject brought no performance gains at all; maybe this should be attributed to our demos, maybe CATALYST4.12 beta driver turns this on by default, maybe Valve has quietly enabled Fast Z Reject for the RADEON X800 in a minor update for the game (since we run the game via the Steam software, we receive updates automatically).

Instant FPS: Performance in Graphs

In order to get better knowledge of our demos behaviour and also find out which graphics architecture delivers higher minimal framerate, we measured a couple of our demos along with top graphics cards from ATI and NVIDIA using Fraps utility.

The results did not bring anything new to us: ATI RADEON X800 appears to be faster than the GeForce 6800 in terms of minimal as well as average frames per second.

Some consider minimal frames per second number more important compared to the average FPS because it reflects the most complicated part of the testing sequence, such as a heavy battle that crucially needs additional speed.


So, Valve’s long-awaited project has fleshed out into shiny compact discs with the Half-Life 2 logo. With its excellent graphics, complex physics of the game world and thrilling gameplay, the second Half-Life is really a representative of the new generation of computer games.

Talking about performance of the Half-Life 2, we should keep in mind that there are two peak situations that represent the new game from Valve:

In case you are buying a processor, you most-definitely have to look at the CPU-bound demos. If you are buying a graphics card, you should look for demos that depend on the GPU performance. The truth is that you do need a fast central processing unit and a rapid graphics card to play the Half-Life 2, even though the game will most probably run comfortably on a machine with Intel Pentium 4 2.40GHz processor and a RADEON 9800 XT graphics card with 1GB of RAM.

Half-Life 2 evidently prefers graphics cards with graphics processors of the new generation developed by ATI Technologies. The RADEON X800 XT Platinum Edition, X800 XT and X800 PRO GPUs run this game at the highest speed. On the other hand, graphics cards on the GeForce 6800 Ultra and GT chips from NVIDIA do not look completely hopeless against their competitors. Using such cards you will still have enough performance for a comfortable play. Moreover, the difference between the top-end solutions from ATI and NVIDIA depends heavily on the specific game scene. The gap is wider in scenes with water surfaces, and narrower in others, especially when there’s mass fighting onscreen and the main load falls on the central processor of the system.

Based on different demos we have examined so far we can state that the RADEON X800-series graphics cards and AMD Athlon 64 processors are superior hardware for the Half-Life 2 compared to NVIDIA GeForce 6800-series and Intel Pentium 4 chips, at least in the so-called peak situations, when gamers need the speed of certain components of their PCs most of all. In the mainstream market segment RADEON 9800- and RADEON X700-families deliver pretty much equal speed with NVIDIA’s GeForce 6600- breed of products.

Given that the Half-Life 2 game was released days ago, we believe graphics processors designers will spend quite some time optimizing drivers to run the Half-Life 2 faster, therefore, the final conclusion will probably be out in future, not now.