Rage3D ATi HD 4850 Single Card DX9 Tests
By Alex 'Morgoth Bauglir' Voicu - reviews@rage3d.com
June 26th, 2008
[ Print ] - [ Close ]
Introduction & Setup

Ready to go back to that really really robust DX9 API? We certainly hope so because that's what's going to happen throughout the following pages.

Again, before we start pouring out the graphs, lets revisit the changes we've implemented with this set of reviews:

This is a single-card performance investigation and Crossfire numbers will follow shortly. We'll also be looking at more "exotic" features like Edge-Detect or AAA quite soon, but those deserve their own separate piece.

Here are the specs for the system we used during testing:



With that out of the way, we'll start our DX9-fest with some UT3 action.


Unreal Tournament 3

Version used: 1.2

Testing method: FRAPS run using the UT3Bench with the WAR-Torlan_bot timedemo, results are the average of 6 three minute runs

Another encounter in UE3, after the two we already had earlier in the form of Bioshock and Gears of War. Counter-intuitively, despite of being released at a later date than either of those two titles, UT3 does not include a DX10 pathway. Without DX10 support, and given the way it implements shadow rendering (deferred), there is no out of the box support for AA.

Lucky for us then that a driver workaround ("hack" for those of you with a less politically correct mindset) permits forcing AA through the CCC. It's not perfect, mind-you, as without developers using DX10/10.1 and handling AA themselves, some bugs are unavoidable, but it's good enough.

We're using FRAPS+UT3Bench+Bot timedemo because that's closer to what happens when actually playing the game, yet keeps our very poor UT3 skills from skewing the results (staring at the death screen continuously for 3 minutes might not be an accurate representation of game-play). With the bot timedemo we have some control and can switch perspectives if the bots decide to do something retarded like stare at a wall, allowing us to ensure it's as close to the UT3 gaming experience as possible.

Oddly shaped pictures of checkboxes:


We're controlling AA through the CCC and AF through UT3Bench's nifty menu. Numbers to use in heated forum battles:




Now, without AA, UT3 isn't the most demanding of games as it is a primarily multiplayer game, geared towards catering to a lot of possible configurations out of which most aren't exactly powerhouses. This translates into a small gap between the competing architectures, which widens sharply once you enable AA. The small drop in performance associated with enabling 8X AA on the 4850 is also most impressive.


Be advised that the FRAPS run we're using is slightly more demanding than the demo we were using before, so results aren't directly comparable with the ones we got on other occasions in Crysis, not to mention the fact that we have to correct those results anyhow since an error made its way amongst them.


Half-Life 2, Episode 2

Version used: whichever Steam decided was the latest

Testing method: FRAPS run through the rocket launch part of the "T-Minus One" chapter, the results are the average of 6 three minute runs

Uproar, anger, disappointment, condescension fill the room: "Still testing HL2? PUHLEEEZE. Review-padders!!!". It's not cutting-edge, it doesn't look that good and it's no Crysis in terms of requirements and capability of sending GPUs back to their mommies, so why is it still being tested? Elementary, our dear Watsons: a lot (that should be all caps) of folks still play HL2/Source-engine based games, and they probably would like to see how a new toy might improve their experience.

The Lost Coast is truly lost this time and we move on to using Episode Two which should be a fairly accurate representation of where Source is now.

Here are the settings we used:


AA and AF levels are controlled through the game's menu.




Lo and behold, even an older engine can become quite demanding once AA and AF are ramped up. Whilst the 3870 was becoming a little slow with 8X AA (slow for a game like HL2), the 4850 runs like a champ.



Call of Duty 4

Version used: 1.5 (1.4 for single player executable)

Testing method: FRAPS run through the "Blackout" level, results are the average of 6 three minute runs

Considering the number of copies this particular game sold, it's fairly safe to assume quite a few of you want to know how it behaves with the new GPUs. Also, in spite of not being on the bleeding-edge technology wise, it combines a lot of older tricks in order to achieve an appealing look.

Here are the settings we used:

AA and AF were controlled through the game's menu, with the mention that for enabling 8X AA we used the r_aaSamples console command after which we triggered a reboot of the video subsystem in order to apply it.




Hey, who gave the new guy steroids? You'll recall from earlier articles that ATi cards liked COD4 a lot ... this does not change, and the 4850 seems to coast through the game with no problems, up to an impressive lead with the Extreme Quality preset (it's nearly 2.5 times faster, close to where it's theoretical specs place it, in spite of the lower clockrates). Take home note is that a 4850 should ensure you a very pleasurable trip through the world of Modern Warfare.



S.T.A.L.K.E.R.

Version used: 1.0.0.6+TheFloatingPoint 2.0

Testing method: FRAPS run through the battle between the stalkers and the military at the Agroprom Research Institute, results are the average of 6 three minute runs

Stalker, despite a long development cycle, was actually worth the wait. The engine is still quite impressive and reasonably demanding, making it a good test for putting GPUs through their paces. Since the game hasn't been patched in a couple of ages (well, perhaps not that much), we've opted to also use the FloatingPoint mod. This mod adds a more accurate lighting model, more complex surface materials, enables higher resolution shadow maps by default and tweaks some of the game's shaders. In short, it's good stuff, and we recommend that you grab it from www.thefloatingpoint.org.

Embellishments applied to the Zone:


Stalker is a deferred renderer, so no AA for it until the Clear Sky expansion lands later this year. We have 16X AF enabled for the tests though.


The 3870 was already reasonably fast, and the 4850 improves on that, but the lead isn't as solid as we've seen under other circumstances. Without AA, even with the added load of the FloatingPoint mod and at a reasonably high resolution, it seems that the game isn't that demanding for either of the Radeons, at least in the tested scenario (for what it's worth, indoors it's even faster on both cards). The solid increase in minimum FPS does indicate that the added muscle of the 4850 helps in the most performance intensive scenes of the testing run.



The Witcher

Version used: 1.3

Testing method: FRAPS run through the "Old Vizima" level, results are the average of 6 three minute long runs

If you must play something, play The Witcher (no, we're not game pimps) - it's very very good for those that like RPGs, and is surprisingly demanding of GPUs. Even if it is another title that doesn't rely on a throng of demanding features, it does put a lot of little things together and manages to achieve a very stylish look, which places a rather hefty burden on those little silicon chunks we all know and love.

The potions we drank before going monster hunting:

AF is controlled through the game's menu, whilst we use the CCC to force the differing levels of AA.




A good lead, especially once we start adding AA, but still a bit lower performance than we'd like. The average even with 8X AA is probably sufficient for a RPG, but those minimum framerates we don't quite like. This might well be a driver issue, as there are some strange hitches in performance that would suggest some underlying problem, and we'll have the opportunity to verify this assumption once we start playing with Crossfire, as we'll be using newer drivers there.



Gothic 3

Version used: 1.6

Testing method: FRAPS run going from Gotha, passing by Montera on to Silden, following the road, results are the average of 6 three minute runs

Two reasons pushed Gothic 3 into our test suite:

  1. it's an old guy that still kicks, doing a decent amount of work per surface and relying on a nice, dynamic lighting model

  2. we like it ... hmm, perhaps this one was actually the main reason

The Unnamed Hero asked for these to be enabled:


Gothic3 is a game historically plagued by numerous bugs (it still is to a certain extent) and, as it never gained AA support in spite of hints of it being hidden deep within the executable, this is another title where we're limited to a single graph that shows performance with 16X AF enabled.


With a 10 FPS bump, the 4850 manages to make the game more enjoyable compared to the 3870 and, since Gothic doesn't play nice with CF at all, it actually beats the 3870 X2 as well. The minimums of 0 FPS are caused by the way the game loads tiles for different zones, so it's not due to the GPUs.



TimeShift

Version used: 1.02

Testing method: FRAPS run through the opening level of the game, "Arrival", results represent the average of 6 three minute long runs

What do we have here? Probably one of the nicest games you've never heard about, really. Since our product pimping quota is spent, we won't harp about its gameplay merits, but we'll mention that it's packed to the gills with graphical goodies, but hampered by a less stellar lighting model.

Time had to be reversed for these:


Why are "Projected Shadows" disabled? Those are a feature exclusive to nVidia hardware, so we can't test them (yet). TimeShift steals a page from Gothic3's "How to do odd things with grace" manual by not supporting AA in spite of a configuration file indicating support at some point in time ... probably before the huge graphical overhaul that took place at a certain point during its development cycle. You'll hear no complaints from us as it makes the job easier, with only 16X AF performance being tested.


Timeshift does quite a bit of math, so the increase in SIMD count of the 4850 helped here, probably more than the added TUs did as superficial inspection doesn't indicate an elevated texturing load. If you disagree, write a 10000 word essay justifying your position and email it to us ... or simply let us know about it and we'll recheck.



Clive Barker's Jericho

Version used: 1.0.0.1

Testing method: FRAPS run through the first level after entering the Al-Khali complex, results are the average of 6 three minute runs

Jericho shows that it was primarily developed on the Xbox360 by avoiding using large textures (consoles have to be more frugal about memory). It makes up for this by applying numerous shaders per surface, which in short makes everything look good, and causes video card coolers to spin-up.

Innumerable flies were battled in order to rescue these arcane inscriptions:


AA is controlled through the game's menu, whilst AF is managed through its configuration file. We smacked undead at these exact rates:




The game is a good sport and allows us to show how the 4850 improves on the 3870. It's also very demanding, so sky high framerates are out of the question. It should be noted that at 8X AA we were running into VRAM limitations, so some hitches associated with that were experienced.



Colin McRae's DIRT & Conclusion

Version used: 1.2.0.0

Testing method: FRAPS run through the "San Ysidro Mountains USA" truck race, results are the average of 6 three minute runs

Finally done ... for this piece. Now, let's quickly write something and go to bed ... or not. DiRT is one of the better engines out there, being smart about getting the best out of GPUs and CPUs, which is rather rare these days.

Wear sunglasses before looking ... bloom alert:

The game supports AA up through 4 samples in its menu, and we went around and messed with its configuration file to get 8 sample AA. AF is forced through the CCC . Pedal to the metal:




HD 4850 is faster etc etc etc ... you already know that by now. What's interesting here is the 8X AA numbers- either our config hacking skills aren't producing the expected results, or the game is smart about how it handles AA. Subjectively, 8X AA seemed to be applied properly, but we'll take a closer look and apply corrections, if needed. So consider the 8X AA numbers with that in mind.


A Closing Teaser

No, the conclusion is not here, maybe later. We're done with looking at the 4850 in solace though, next up Crossfire performance. Stay tuned!

Ready to enjoy the RV770 experience yourself? Don't miss on your chance to win a Radeon HD4850 Crossfire combo, sponsored by ATi!