Half Life 2: Episode 2 Performance @ [H]

FrgMstr

Just Plain Mean
Staff member
2FA
Joined
May 18, 1997
Messages
58,694
Half Life 2: Episode 2 Performance and IQ - Is the Radeon HD 2900 XT as fast as a GeForce 8800 GTX in Half Life 2: Episode 2? You might just be shocked at some of the performance results. We also have some interesting image quality comparisons to look at between ATI and NVIDIA hardware in this game; all is not the same.

Unexpectedly the ATI Radeon HD 2900 XT made a stellar showing in performance with Quality Adaptive Supersampling Antialiasing enabled in Episode 2. Either the GeForce 8800 GTX is abnormally slow with Transparency Supersampling in this game or the Radeon HD 2900 XT is super-fast, because the 2900 XT nearly punched through the GeForce 8800 GTX performance envelope. Only with Quality Adaptive AA enabled however, once we disabled the feature the GTX greatly outpaced the 2900 XT.

Please Digg to Share!
 
First of all, Anand had more "numbers". However, the [H], per usual, looked into the more intersesting "things" which is where the real value is.

I also think it is a good thing for the 2900 owners to have something to hand their hat on for once. Nobody (well at least people of worth) likes to constantly watch one legged per in a kicking contest. Hopefully in the next series of cards by ATI they can take their fundamental technology + fixed AA and make this into a true fight again for the performance crown.

Thanks again.
 
Thanks for another excellent article chaps.

Interesting to see the beleaguered 8800GTS 320 MB cards doing well for a change. The game plays beautifully with everything on on my spec.

It would seem that the in the dual core comparisons the 2nd CPU isn't being fully utilised with this game. Even so it would be good to see some quad core comparisons in future....

I can't say I'm surprised the 2900XT cards do so well with this game, Value have always supported the ATi brand first. It's strange seeing ATi sponsoring and advertising in games these days - pretty much all the big release now are Nvidia sponsored.
 
Good review, and interesting numbers put out by the 2900XT. I was mostly happy to see the 320mb GTS doing so well :p

Can't wait to see the TF2 performance. I know I can play mine w/ everything maxed, including very high textures and 8x CSAA and haven't felt the FPS drop even with a lot of action... but that's for the next performance review. I still have to beat Ep. 1 :D
 
Always love the [H] reviews but why are 7900s missing? Or do you plan on including those in with the "2600 XT and 8600 GTS"?

I need to check my settings again because with my 7900 GTX I am having a hard time noticing motion blur in Episode 2, its very easy to spot in TF2 though.

Any noticeable difference between the .74s / .75s beta drivers? I know the .74s made a huge increase for me with the Orange Box.

Its funny that you mention the lighting/shadows because in the missile silo level I really used the flashlight and was amazed at how well the lighting looked. I just sat and played with the flashlight and at one point tried to make odd shadows on the walls while the in game characters were chatting it up.

One last thing the Vortigaunt’s textures are allot better in Episode 2 then the other Episodes/Halflife 2. They also have allot better animations and the eye is so nicely done.
 
My 8800GTX OC can do 16xQaa 16af, 1680 x 1050, supersampling, very high textures, everything on high, totally smooth. So I'm not sure where you get 8xaa for maximum playable settings on the GTX.
 
Is there a good article examining the difference between regular 4xAA and QA4xAA? There seems to be a big performance hit turning this on. Is it clearly worth it?
 
that shadow thing looks like a bug i don't see any flashlight effect with 2900 seems that there is no light coming from your flashlight
 
the frame rates seem much lower that what im getting with my 8800 320 at the same settings. i am also using the 163.71 not the 163.75.
 
Anyone else notice that the 2900xt had the same FPS with 4xAA and the Quality Adaptive AA Supersampling (4x)?

Can you guys please do some comparison shots between these AA modes as it looks like the Quality Adaptive AA Supersampling is not working and its using regular AA.

Note on the GTX / 2900 graphs, the it starts @ ~45, spikes @ 90, then spikes near 60 a few times for both settings, the graphs look identical for 4xAA and the 4xQAA.

http://enthusiast.hardocp.com/article.html?art=MTQwNCw0LCxoZW50aHVzaWFzdA==
 
After seeing these numbers and those at anands I placed an order for an upgrade for my poor old X2 3800+, I have at least 40 FPS to gain for just a small sum of money. Also, while other sites have more numbers, pictures and flash, I personally find that they lack the quality writing we're used to with [H] articles after all these years. I feel like I'm reading playboy just for the articles and I'm liking it.


The experiments with shadows cast from the flashlight - my 2600XT delivers something inbetween the two samples shown, the 8800 shadow looks too hard and out of place, the 2900 one looks way too soft and subtle, but when looking at the overcast light image (for lack of better words) on Alyx, both the 8800 and the 2900 cast the same strength shadow on the floor, which is weird.
 
Great article, and thank you for including 2560x1600 in your review.

It is awsome that a new game is playable at that res. Lately, it seem most games strugle at that res with a GTX which I find insane!!
 
My theory: The difference between the single core & dual core, probably has to do with the drivers, utilizing a helper thread. (multithreaded drivers, not game) Thats why the performance is slightly better and there is some activity on the 2nd core.
 
My 8800GTX OC can do 16xQaa...
"playable" means different things to different folks. I actually found 8xQ TSAA @ 1680x1050 on a GTX @ 648/1512/2000 to be less than ideal, so I backed it down to 4x. The difference in IQ's just not enough for me to live in 45-50, when I can be in 55-60 territory. It also provides a more consistently smooth experience and there's a lot less jarring min fps drops too.
 
"playable" means different things to different folks. I actually found 8xQ TSAA @ 1680x1050 on a GTX @ 648/1512/2000 to be less than ideal, so I backed it down to 4x. The difference in IQ's just not enough for me to live in 45-50, when I can be in 55-60 territory. It also provides a more consistently smooth experience and there's a lot less jarring min fps drops too.

Exactly correct on "playable." I think we hit a good medium as to what will make most gamers happy. As always, YMMV.
 
"playable" means different things to different folks. I actually found 8xQ TSAA @ 1680x1050 on a GTX @ 648/1512/2000 to be less than ideal, so I backed it down to 4x. The difference in IQ's just not enough for me to live in 45-50, when I can be in 55-60 territory. It also provides a more consistently smooth experience and there's a lot less jarring min fps drops too.

True true. Personally I can play most FPS at 30 or more, mostly because most of my computer time I had a very mid to lower range computer, but still liked the graphics up, so 30+ FPS works just fine for me. So when I get to play at 40-50 or even 60 FPS it's effin beautiful :D
 
I'm wondering how much of a performance boost the 2900 is receiving due to missing grass and/or weak lighting effects. I think this needs to be looked into more. Seems like the 8800 hardware is working harder, hence the close fps between the two cards.

And maybe its just me, but I'm getting better fps on my 320 than [H] was. I'll get some hard numbers, but its seemed to be running better than 35ish for me at those settings.
 
Anyone else notice that the 2900xt had the same FPS with 4xAA and the Quality Adaptive AA Supersampling (4x)?

Can you guys please do some comparison shots between these AA modes as it looks like the Quality Adaptive AA Supersampling is not working and its using regular AA.

Note on the GTX / 2900 graphs, the it starts @ ~45, spikes @ 90, then spikes near 60 a few times for both settings, the graphs look identical for 4xAA and the 4xQAA.

http://enthusiast.hardocp.com/article.html?art=MTQwNCw0LCxoZW50aHVzaWFzdA==

I covered this in the article, there isn't much of a performance hit enabling Quality Adaptive AA on the 2900 XT, which makes it almost as fast as the GTX. You can clearly see it is working in the screenshot comparisons. As mentioned in the article we double and triple checked, and 4X Q ADAA WAS working.

That's kind of the whole point here, ATI doesn't take a hit with Q ADAA in this game, while NV does take a large hit.
 
What drivers were used for the ATi card?
I installed the 7.10's the other day and kept getting horrible texture corruption in Portal, I didn't even bother seeing what it was like in HL ep2.

Going back was a pain in the butt because after I uninstalled the 7.10's to got back to the 7.9's it said the 710's were still installed.

:mad:

anyone get corruption with these drivers also?
 
that shadow thing looks like a bug i don't see any flashlight effect with 2900 seems that there is no light coming from your flashlight

There is light, it is just not as bright as the NV side. Again, that is the point of the screenshots, the NV GPUs are clearly providing a better flashlight IQ and shadow quality.
 
Always love the [H] reviews but why are 7900s missing? Or do you plan on including those in with the "2600 XT and 8600 GTS"?

We are concentrating on current generation video cards, it is possible we may look at older cards, down the road.

Any noticeable difference between the .74s / .75s beta drivers? I know the .74s made a huge increase for me with the Orange Box.

We used the latest driver at the time of testing on NZONE which is 163.75, and it is still the latest driver.
 
rarely ever used the flash light to begin with, forgot it was even in the game :p
 
"playable" means different things to different folks. I actually found 8xQ TSAA @ 1680x1050 on a GTX @ 648/1512/2000 to be less than ideal, so I backed it down to 4x. The difference in IQ's just not enough for me to live in 45-50, when I can be in 55-60 territory. It also provides a more consistently smooth experience and there's a lot less jarring min fps drops too.

Correct, also consider we used regular stock speeds. You should be able to hit at least 8X TR SSAA on a GTX for playable performance. You may indeed find you can go higher. For us though, we saw a large hit in framerate at 16X TR SSAA in outdoor areas such ep2_outland_10a

the frame rates seem much lower that what im getting with my 8800 320 at the same settings. i am also using the 163.71 not the 163.75.

The above explanation applies here as well
 
That's kind of the whole point here, ATI doesn't take a hit with Q ADAA in this game, while NV does take a large hit.

Isn't that simply because G80 is supersampling while R600 is doing alpha-to-mask? I thought we went over this a long time ago....
 
I covered this in the article, there isn't much of a performance hit enabling Quality Adaptive AA on the 2900 XT, which makes it almost as fast as the GTX. You can clearly see it is working in the screenshot comparisons. As mentioned in the article we double and triple checked, and 4X Q ADAA WAS working.

That's kind of the whole point here, ATI doesn't take a hit with Q ADAA in this game, while NV does take a large hit.

Thanks Brent, I only skimmed the article before heading out to work this morning. Can you do some comparison shots on just the 2900xt with reg 4x and 4xQADAA?
 
Thanks Brent, I only skimmed the article before heading out to work this morning. Can you do some comparison shots on just the 2900xt with reg 4x and 4xQADAA?

The ones we have in the IQ section are sufficient, they show no differences for regular AA on objects, and do show the differences we experienced with Q ADAA and TR SSAA on alpha textures. I need to stress though, when you are actually playing the game, it is hard to notice the differences on alpha textures, we only discovered them upon examining the still screenshots.

In our forthcoming evaluations where we use this game we will provide appropriate AA screenshot comparisons.

In my opinion the flashlight/shadow issue is the bigger issue.
 
The ones we have in the IQ section are sufficient, they show no change for regular 4X AA on objects, and do show the differences we experiences with Q ADAA and TR SSAA on alpha textures. I need to stress though, when you are actually playing the game, it is hard to notice the differences on alpha textures, we only discovered them upon examining the still screenshots.

In our forthcoming evaluations where we use this game we will provide appropriate AA screenshot comparisons.

In my opinion the flashlight/shadow issue is the bigger issue.

Oh I do agree that the flashlight problem is a big one, esp when you need to use it to look in some areas.
 
Great writeup. I was surprised that more wasn't discussed about the material shaders (is that the correct term?) except for the Vortigaunts' skin.

Everything in the game has really mad textures now. In fact I found myself just staring at a cube in Portal for ages, moving it in and out of various light sources. It's especially evident on the skin on all the main characters in EP2. Compare the G-Man in EP2 to the one at the start of Point Insertion, for instance. Magnusson also has really amazing skin. It's so lifelike that it's a bit disturbing actually. Usually games have either dull skin, or rubber skin, or ultra shiny skin but EP2 actually looks like all those tech demos from nvidia and ATI that we got with our cards several years ago.

The materials shaders (again, not sure if that's the right term) used in the new Source engine are the big highlight imo.
 
What does it take to get 1680x1050 resolutions tested?

Newsflash: gamers are buying today's hardware, not yesterdays. Widescreen monitors are the future.

Stop testing 1600x1200 and start 1680x1050 please.
 
There is light, it is just not as bright as the NV side. Again, that is the point of the screenshots, the NV GPUs are clearly providing a better flashlight IQ and shadow quality.
hmmm strange was there same issue with other radeon cards in half life 2 ? i think when i played first one with 9600xt i clearly saw bright light coming from my flashlight don't remember shadows but light was much brighter may be as bright as nvidias
 
The ones we have in the IQ section are sufficient, they show no differences for regular AA on objects, and do show the differences we experienced with Q ADAA and TR SSAA on alpha textures. I need to stress though, when you are actually playing the game, it is hard to notice the differences on alpha textures, we only discovered them upon examining the still screenshots.

How are they sufficient for regular 4xAA vs QA/TR 4xAA? There isn't one single comparison showing regular 4xAA.

Is this worth invoking on NV cards given the big performance hit?
 
How are they sufficient for regular 4xAA vs QA/TR 4xAA? There isn't one single comparison showing regular 4xAA.

Is this worth invoking on NV cards given the big performance hit?

The very first two AA screenshots compare regular AA.
 
What does it take to get 1680x1050 resolutions tested?

Newsflash: gamers are buying today's hardware, not yesterdays. Widescreen monitors are the future.

Stop testing 1600x1200 and start 1680x1050 please.

Maybe if you bitch some more? Just kidding... but seriously.... considering 1680*1050 is less than 1600x1200, then just assume that if the card does alright at that res, it'll do just slightly better at 1680x1050.
 
hmmm strange was there same issue with other radeon cards in half life 2 ? i think when i played first one with 9600xt i clearly saw bright light coming from my flashlight don't remember shadows but light was much brighter may be as bright as nvidias

I don't think we've seen this issue in previous versions of Half Life 2.
 
Back
Top