Starcraft II GPU performance/IQ

I'd say it's more a contrast in scope. SupCom is designed for absolutely huge battlefields. StarCraft 2 is the more traditional RTS, like AoE and C&C, with close-in unit micromanagement and little maps. (I was in the beta, will likely get the final here eventually)

I'm currently trying to stomach SupCom2. Working through the sp campaign. I loved SupCom FA's sp campaign. SupCom2 is certainly different but I won't completely judge it until I've beaten it. It was a whole $11 when I bought it.

 

I think it has to be remembered that it was Blizzard's decision not to support FSAA in the game, and that both nVidia and ATi are being forced to try and ad-hoc FSAA support into the game after the fact. ATi's already confirmed that they are preparing a hotfix to support The Feature That Blizzard Didn't, and I'm not yet convinced that nV has really beaten ATi to the punch here, especially if we consider the quality of the FSAA implementations for Starcraft2. I mean, ATi could have taken one look at what nVidia did and said, "No way we're going to do anything that chintzy! We'll take a bit of time and do it better/right!" I don't know if this is the case, of course, but we'll be able to decide that after the ATi hotfix is released. If the ATi support for FSAA in the game is measurably better in terms of IQ, then I think few will care that it took ATi longer to patch in support for a feature that neither nVidia nor ATi should have had to patch in in the first place! It will all depend on how well the two FSAA implementations compare.

Well, I can't address that particular issue as we all know that often game benchmarks differ widely from site to site even when the same games and gpus are compared. A lot depends on who is doing the comparison and how they are doing it.

Yea, kind of reminds me of the benchmarks nVidia was plastering all over the web that purported to show how badly nV30 was thrashing R300 in various games.... I think it will be awhile before we'll have some really definitive info on this game as to not only the IQ the competing gpus produce, but also the "performance" they reach while providing it.

 

*nV has spent a lot of time optimizing its drivers for Sc2.
*ATi hasn't spent as much time or effort to date on optimizing its drivers for sc2, and is only now bringing its full attention to bear.
*Reviewers to date are still rushing to get out their Sc2 reviews as Blizzard forbade them from publishing reviews until the game was released for public consumption. In that haste, many things might have been overlooked that would tend to tilt the playing field in one direction or another.

The why of it really isn't important because both companies (ATi & nVidia) often do things that seem to defy the logic the other company is using at any given point in time...

 

Yeah I mentioned that. Each game has it's compromises and sacrifices it has to make. For SupCom it was large expanses and large number of units. So graphics, unit detail, environment detail, etc. all took a dive. It worked in that you really had a sense of scale, a sense that you were battling on a large world.

SupCom 2 really is just a horrible game. It tries too hard to be both a traditional RTS and a TA/SupCom RTS, and in the process fails miserably at both, with virtually no redeeming qualities, IMO.

Regards,
SB

 

I would be shocked if that was the case.

So it's nvidia's fault that amd cards crashed with Saboteur (a game sponsored by amd)! I knew it! Eventually Walt you're going to run out of excuses for amd; everyone already has.

 

I absolutely cannot believe that you managed to bring NV30 into this. My mind is completely boggled.

 

We used to call "in-game cinematics" by the acronym "FMV," which stood for Full Motion Video.
Often during that time games would advertise stunning FMVs as though they represented the actual game play the 3d engine provided the player at all times when playing the game. FMV fell out of favor pretty rapidly when it was discovered that the IQ provided by the game engine rarely if ever even came close to the IQ displayed in the "cut scenes," or FMV clips.

I much prefer it when a game's "cut scenes" or "in-game cinematics" are provided by the actual game's 3d-engine performing according to the dictates of a script. In this way the cinematics look exactly like the normal game play and the graphical consistency of the entire game is greatly enhanced. I can often remember being thrilled by a game's initial cut scene because the graphics were fantastic, but then being sorely disappointed a short time later when the cinematics ended and the normal game engine took over the game play--because by comparison the IQ of the game engine was much poorer than the IQ provided in the "cut scene" or the "FMV" or the "in-game cinematics" or whatever you'd like to call them.

Such cinematics are 2d in nature, and can be run through a variety of image processors to make the resulting "cinematic" look so much better than the IQ provided by a game's actual 3d-play engine, that the IQ difference between them is night and day. This approach was popular for awhile because it helped companies market games with much better looking images than the games actually provided while you were playing them--and it was that very disparity that has so greatly de-emphasized the importance of FMV in current 3d games. That, and the fact that 3d game-play engines have improved so much in terms of IQ that it's preferable to use the game's own engine to provide the in-game cinematics so that the entire game's graphical consistency remains intact--which is very important in preserving the desired level of realism and/or believability the developer wishes to impart.

I have no idea how Sc2 handles its "in-game cinematics," but I hope that they all use Sc2's 3d engine instead of being strictly 2d animations filtered and processed to death to create a level of IQ that isn't present throughout the game. I think that games that utilize 2d cut scenes simply to create the illusion of better IQ than the game's 3d engine can produce are inferior to games that rely solely on the game's actual 3d engine to create the cut scenes and in-game cinematics. Would you say that Sc2 uses its own 3d engine to display the cinematics, or that the game uses cinematics the quality of which is not possible during normal game play? Just wondering, as your post above dredged up some things I haven't thought of in quite awhile.

 

Really? I have been shocked many times by what various 3d IHVs have both done or said in the past...

I wasn't offering "excuses" for anybody--my intent was to show how misleading published game benchmarks can be at times.

 

The point was, of course, not nV30 at all. The point of the remark, as I explained to Willard, was to demonstrate how published benchmarks can be erroneous, sometimes egregiously erroneous.

Really, too, I think we've all matured past the point where "nV30" is taboo, for some reason. I thought just barely mentioning it was an excellent example of how benchmarks can be misleading. My mind would boggle if you disagreed with that premise...

Can someone please explain the mystical curse that apparently comes down on someone who simply says, "nV30"? Is the nomenclature truly that horrifying that it may not be muttered even in the dead of night? I don't get it. If I was to mention in passing, "V5 6k" in order to illustrate a point, would it elicit the same reaction?

 

So is that imminently to be released fix that you were chastising me for criticizing out yet Walt? Y'know, the one that was hours away yesterday?

 

Nah v5 6k would surely not elicit the same reaction. It'd be a welcome sign of maturing past the nv30 broken record phase you've been stuck in for the last 7 years.

 

Now lets do a quick comparison of pre (OP) vs post SC2 release benchmark results. Wow, the performance looks nothing like the pre release of the game, gee I wonder why?
Still waiting for the hotfix though.

 

One thing's for certain ... Rampage would have supported AA in SC2 from the get-go.

 

Hmm well my guess would be because you can't read German.

But wonder no more because i'm here to help with a translation:

Hey I think it means these results could be CPU-limited! Or something. Hmm, wait that could be the reason why the graph says 'Results not necessarily comparable with earlier reviews'..

 

It actually means (without blabla) In singleplayer, the fps usually are significantly higher because in that mode there's not so many units and thus not as cpu limited as our scene"

Looking at how Starcraft was played, it's natural to assume that the overwhelming majority of playing hours will be in multiplayer.

No, that's a standard in our benchmark, because we're getting links from non-german speaking countries and often the benchmarks pictures are posted elsewhere without the necessary context - for example mentioning different circumstances for the test.

But in a sense you're right: We've included that line to make people more cautious when comparing "just benchmark bars" without necessary context. That way, it's harder to just dismiss our work with "they're dumb".

 

There are two types of cinematics in SC2. First is the good-old pre-rendered video of the traditional Blizzard quality (= awesome). Most of the cutscenes are done with the in-game engine. They use separate models there for characters and environment and they look really great, though have very little in common with the actual graphics you see during the RTS part.

 

Can someone reveal how Nvdia got MSAA working in SCII?

 

I think it's super sampling, rather than MSAA, hence the performance hit.

 

I asked PCGH's Marc Sauter, what kind of AA is it, and he answered it's MSAA, you can add TAA over CP and SSAA over Nvidia's SSAA tool.

The performance hit is "normal".

 

The results between cards are very close between the 5800 vs the 400 series vs the OP's results. And it was done online instead of SP mode. Unlike the previous reviews. IMO, people would prefer to see online results instead of SP results. And if you recall, the OP link clearly states at 5% variance in results between runs. Which is why I really didn't care for it.