Original Link: http://www.anandtech.com/show/1450


If anyone were asked to name the two most anticipated games of this year, no thought would be required before putting forth the inevitable reply of "Doom 3 and Halflife 2". Order may very depending on the gamer's background and taste, and the occasional "Sims 2" response may be heard in the minority. So far, we have Doom 3 (even The Sims 2 is already Gold), but Halflife 2 is still MIA. We keep hearing rumors (though nothing substantial), but even last year's certain release wasn't set in stone.

Ever since the incendiary remarks made by Valve's Gabe Newell about the ability of Valve's programmers to come up with code that ran as fast on NVIDIA's hardware as it did on ATI's (and putting the blame for this square on NVIDIA's shoulders), and the following source code leak, both the buzz over the Halflife sequel and its delay have increased in what feels like an exponential manner.

Of course, last year, ATI "bundled" Halflife 2 with its cards, but consumers who made the purchase were left with little in the way of fulfillment of this offer. That is, until now. Last week, Valve pushed out a one level beta version of the Counterstrike mod fitted to the Halflife 2 core over steam for those customers who had registered their ATI HL2 coupons. Eventually, the game will be released as Counterstike: Source, but, for now, the beta version shows off the bells, whistles, and capabilities of the new Source engine that powers HL2.

Light blooming through windows, illuminating dust in the air in Counterstrike: Source Beta

As an added bonus, Valve included a video card "stress test" in the beta version of CS: Source. Now, with an updated version of the engine, refined drivers, and brand new cards, we take a look at another bit of the story behind the ever lengthening saga of Halflife 2.

Source, CS, and Halflife 2

Before we take a look at the numbers, we have to know what we are dealing with. There are really three important components to address when looking at the beta version of Counterstrike: Source. First, there's the fact that the game is based on the Source engine and has a built in graphics stress test to push hardware to its limits and beyond. Next, there's Counterstrike and how that franchise fits into the equation. Finally, the crux of the matter, what do our observations reveal about the upcoming Halflife 2 title?

The opening of the Video Stress Test benchmark

In taking a look at the source engine test level supplied with CS:Source Beta, one of the first things we noticed was a difference in feeling between Source and id Software's latest engine. We've only got a test room and a sandy CS level to test the Source on, so we really haven't seen all that it has to offer, but our first impression of the engine is that it is very "clean". Clean like listening to an undistorted guitar after hearing some intensely heavy rock, or the feeling of looking at an HDTV next to an old UHF box. So far, the scenes we've seen rendered with Source show us that the engine is very good at producing crisp, clear images. Our initial impression of Doom III wasn't that it was any less real or vivid (we still got that warm, squishy, HD feeling inside), but that it was real in a dark, dirty, and atmospheric way. We will probably see how well Valve can coax their engine into producing similar situations at different points in Halflife 2, and surely there will be Doom III engine licensees who will create games with a slightly (for lack of a better word) happier feel. But at this point we don't have enough info to say that one engine is "better" than the other. So far, we'll just call them different. We are absolutely looking forward to seeing how Valve will integrate some of the very cool effects produced by the stress test in gaming situations.

Some of the more complex rendering that goes on in the Video Stress Test is overlapping reflective/refractive surfaces

Counterstrike: Source is somewhat of an interesting choice for Valve to lead off with. Historically, CS players have been very demanding on framerate. While we can't get 200fps at high resolutions, the subtle (and not so subtle) enhancements to the game engine add a depth to Counterstrike that could revitalize the franchise. Unlike the stress test, the enhancements in the one level available for play in Counterstrike: Source Beta (dust) aren't all about fire, water, reflection/refraction, and tv monitors. The main benefits we see in CS are in the added clarity, feeling of realism, and enhanced physics. Kicking a box or rolling an oil drum down a hill are fun enough to distract players from the game at hand. Actually, such real world interactions with the environment provide a new way to play: kicking a tire down one side of a hill to draw sniper fire, then running down the other is a very effective trick. Smoke and flashbang effects, muzzle flares, and explosions add to the experience as well. Lighting effects inside buildings pile on the realism. With the intensity of effects kept to a subtle level, high frame rates are still achievable, and if we decide to include CS: Source as a test game in future articles, we will put a larger emphasis on higher frame rates than we will from the stress test we will see benched here.

When it comes to Halflife 2, we still don't have a very good picture of how Counterstrike and the included stress test will relate to game play. We will very likely see less sheer graphical impact in the final version of HL2 than in the stress test. After all, Valve packed tons of water, translucent surfaces, and special effects into a tiny room in order to push graphics cards to the limit. We wouldn't expect (or want) to see a level like that in the game. Effects and eye candy are wonderful things, but game designers always need to be careful not to heap too much of a good thing on gamers. One of the most important aspects of the latest in the FPS genre is to put the player in the game as much as possible. At the same time, HL2 will have AI and huge levels with all kind of stuff going on. We really feel that HL2 will be much more CPU intensive than the graphics stress test (which we would hope), or even Counterstrike. Graphically, HL2 will likely be more intense than CS (as it will have more opportunity for strange alien special effects), though this remains to be seen.

So, with Halflife 2, we're looking at something more CPU intensive than both CS and the gpu stress test, and between the two in average graphical intensity. No, it's not easy to call. Especially since we haven't found a "good" way to benchmark our Counterstrike level yet. We take a brief look at CPU scaling with the video stress test on the A64FX platform, but this won't offer as much insight as we would like until we can get our hands on HL2. An in depth analysis of the impact of the cpu on the graphics engine as far removed from game play as possible could help us separate the graphics factor from the game benchmark later on if we decide to use HL2 as a CPU test (we are hoping to be able to test the impact of the new AI algorithm if possible).. Of course, that depends on the type of benchmarking software that comes with HL2 (if any).

But that's enough background. Now on to the tests.

The Test

Our benchmarks will take a look at the graphics stress test at 3 different resolutions with and without AA/AF (where available). We decided it was time to add 2048x1536 to our tests formally. These latest drivers make it much easier to take advantage of the highest resolution available to the highest end cards these days. It's been a much easier experience on NVIDIA hardware (all we had to do was pick the resolution and it worked). On ATI hardware, we've had to fiddle with monitor settings and maximum refresh rate settings in order to finally get 2048x1536 @ 75Hz to become a reality. We don't run this resolution with AA; if you've got a monitor with a dot pitch to do it justice, you just won't need AA anymore. We do still run with 8xAF at playable resolutions on some of the newer cards (which is truly amazing).

 Performance Test Configuration
Processor(s): AMD Athlon 64 FX53 (oc to 2.6GHz)
RAM: 2 x 512Mb OCZ 3200 Platinum Rev 2 (2:2:2:10)
Hard Drives Seagate 120GB 7200 RPM (8MB Buffer)
Video AGP & IDE Bus Master Drivers VIA Hyperion 4.51
Video Card(s): NVIDIA GeForce 6800 Ultra Extreme
NVIDIA GeForce 6800 Ultra
NVIDIA GeForce 6800 GT
NVIDIA GeForce FX 5950
ATI Radeon X800 XT Platinum Ed.
ATI Radeon X800 Pro
ATI Radeon 9800 XT
NVIDIA GeForce 4 Ti 4400
Video Drivers: ATI Catalyst 4.8
NVIDIA ForceWare 61.77
Operating System(s): Windows XP Professional SP1
Motherboards: MSI MS-6702E (VIA K8T800 Pro Chipset)

We do have to note that this isn't an apples to apples test. The NVIDIA 5950 and Ti 4400 couldn't run AA at all. The 5950 will not run in DX9 mode (though we are still looking for ways to force it on just for testing purposes). We suspect that AA is currently tied to DX9 functionality, but the beta makes note of a known bug that could be causing the problem. In addition to not being able to run with AA on, these cards would also not run with water reflections enabled. So far, this seems to be the only DX9 only feature we've seen in the engine. This may or may not change by the time HL2 ships. The effect is lost in the video stress test (there's just too much going on), but could prove very dramatic if Valve made proper use of it in the game.

Performance Tests: No AA/AF

Each of these benchmarks come out the same: The Radeon XT PE on top, followed by the three NVIDIA cards, then everything else.

Unfortunately, no ammount of coaxing could get 20x15 to run properly on the X800 Pro card we had. The Radeon cards seemed more resilient to increasing resolution in this benchmark, so we may have seen the X800 Pro pass the 6800 GT if it had worked.

Also, our Ti 4400 card hit 33 fps in the stress test, but in order to test a more playable frame rate we looked at 10x7 and 8x6 for that card. The GF4 hit 46.6 and 68.4 respectively, showing that older cards will still be able to play Source based games well.

Source Graphics Performance

Source Graphics Performance

Source Graphics Performance

Performance Tests: 4xAA/8xAF

In looking at performance with AA and AF enabled, we can see that ATI increases their lead at the top end, and are able to pull ahead of NVIDIA at the upper mid range level with the X800 Pro leading the 6800 GT.

Again, it would have been nice to see the X800 Pro in the 20x15 slot (which only used 8xAF and not the 4xAA setting). It seems that, even if NVIDIA cards run the highest resolution easier, ATI cards run it better when they finally get there.

Source Graphics Performance

Source Graphics Performance

Source Graphics Performance

CPU Impact Teaser

For our CPU tests, we just wanted to demonstrate that our 2.6GHz Athlon 64 FX system eliminated as much of the CPU as possible from the equation. We don't see any CPU bottle necking until we hit 1024x768. Even then, the FX line of processors is fairly resilient, only slipping a few percent each clock speed drop.

Source CPU Scaling

Source CPU Scaling

Final Words

ATI is the performance leader when we're talking about Source Engine performance at the high end. Unfortunately, we didn't have an X800 XT to compare with our 6800 Ultra, nor did we have our 6800 nonultra that we predict would have fallen between the 9800 XT and X800 Pro (closer to the latter) in our tests. Apparently, some of our graphics cards decided to go on vacation this week to visit a penguin. When it comes to upper midrange, NVIDIA's 6800 GT seems to have a leg up on the X800 Pro in most tests (though this may have changed if we could have gotten our Pro to run 20x15). This is just further proof that (so far) the GT offers some of the best value in NVIDIA's lineup.

Overall, the framerates we saw in these tests were higher than we expected. Doom III will bring just about anything to its knees at the highest settings, and 2048x1536 wasn't even an option on the list. We still expect to see very high framerates when gameplay elements (more CPU usage) are introduced into the mix. This follows the traditional view (that id Software broke from with Doom III) that higher resolutions and higher framerates are always the better option. Certainly, these aspects have their place, but id has proven they aren't the be all end all of graphics engine design. This fundamental difference in viewpoint helps explain our initial impressions of each game. Source can look incredibly crisp running at a steady framerate at 20x15, and Doom III can look incredibly frightening at 10x7 with its intense shadows, atmosphere and lighting effects, and well executed low contrast edges between overlapping objects.

We will absolutely still have to wait for Halflife 2 before we can make any further judgment calls about relative goodness of the engine. Obviously the outcome of our tests revealed that even when source is pushing its hardest against a graphics cards, modern hardware doesn't have any major trouble rendering scenes.

From our brief look at CPU scaling, we can see that none of our tests were really CPU bound. This helps us know we were pushing our graphics hardware as hard as possible. We can also expect Valve to use as much of the CPU headroom they can for other things in the actual game. This is why we haven't taken as in depth a look at CPU scaling yet.

We hope our coverage of Valve's latest beta release has been informative, and if there is anything further anyone would like to explore, please feel free to drop us a comment and let us know.

Log in

Don't have an account? Sign up now