[H]ardOcp : Just Cause 2 performance and image quality

[Madcatatlas]

In which case Nvidia would perform better the higher the tess level gets.

Anything else?

Have you checked your crystal scrying ball again? Whats happening in 6 months Keys? whats happening in 1 year?

AMD/ATi would be fools not to increase the tesselation powers of their 6xxx series to around current Fermi levels. They have had tesselation for far longer and implementing it in a way that equals Fermi should not be a problem.
It might be a problem for their 6xxx if its a 40nm, with half SI and half NI, but it shouldnt be a problem once Tesselation ( IF tesselation) does take off in about a year and their full 28nm NI hits the market.

But then again, these schedules, atleast on nVidias part have missed their targets by over 6 months, maybe AMD/ATi will pull a Fermi in terms of missing schedule.


edit; And whats with the "your not as smart as me" goading/taunting i see from you lately Keys? I could quote several posts from you with something along the lines of "try to keep up with me" "are you following me", "keep up" etc. No need for that imo.

 

[happy medium]

Wish they would have done some benches with equal image quality , equal resolution, and with the Nvidia effects turned off and on.

 

[Qbah]

Wish they would have done some benches with equal image quality , equal resolution, and with the Nvidia effects turned off and on.

They did. There are apples-to-apples sections for each test (same res and settings) and the influence each effect has on performance on the nVidia cards. You did click the link, right?

 

[Keysplayr]

Have you checked your crystal scrying ball again? Whats happening in 6 months Keys? whats happening in 1 year?

AMD/ATi would be fools not to increase the tesselation powers of their 6xxx series to around current Fermi levels. They have had tesselation for far longer and implementing it in a way that equals Fermi should not be a problem.
It might be a problem for their 6xxx if its a 40nm, with half SI and half NI, but it shouldnt be a problem once Tesselation ( IF tesselation) does take off in about a year and their full 28nm NI hits the market.

But then again, these schedules, atleast on nVidias part have missed their targets by over 6 months, maybe AMD/ATi will pull a Fermi in terms of missing schedule.


edit; And whats with the "your not as smart as me" goading/taunting i see from you lately Keys? I could quote several posts from you with something along the lines of "try to keep up with me" "are you following me", "keep up" etc. No need for that imo.

Go ahead. post them up IN context. Have at it.

 

[Dark4ng3l]

Looks like this thread has turned into an nvidia marketing thread already wow I never expected that!

 

[Madcatatlas]

Go ahead. post them up IN context. Have at it.

id rather keep it a notch over that level of debate Keys. reported and who knows what that leads to, if anything.

 

[badb0y]

OOOOkayyyy?????
Did not mean to piss in your oatmeal there sparky. Re- lax.
Slow Spyder seems to be keeping up though. Not sure where you, or I lost track of each other here. ?

Look basically you quoted me and said that nVidia would perform better using tesselation to make water, which I don't have a problem with but that fact that you quoted me and said that made it look like I was somehow promoting ATi.

All I said was, tesselated water>CUDA water. Therefore I would assume companies are going to use tesselation over the CUDA path. Your response?

In which case Nvidia would perform better the higher the tess level gets.

Anything else?

Okay Captain Obvious, thanks for the information.

 

[Attic]

In which case Nvidia would perform better the higher the tess level gets.

Anything else?

I question your place as a moderator when I read posts like this.

I feel the marketing shtick is out of place.

It's in gamers interests for game developers to not be pulling stunts like only implementing features for nVidia or ATI cards based an paycheck from either of the companies.

 

[dguy6789]

I question your place as a moderator when I read posts like this.

I feel the marketing shtick is out of place.

It's in gamers interests for game developers to not be pulling stunts like only implementing features for nVidia or ATI cards based an paycheck from either of the companies.

What do you want Nvidia to do? If they can't compete with actual products, they need to do something to keep people thinking they are the better option.

 

[Lonyo]

In which case Nvidia would perform better the higher the tess level gets.

Anything else?

But would that higher tessellation performance manage to overcome the general slowness in the game? I mean, when they start off being slower, a smaller impact from tessellation in Just Cause 2 if implemented might still mean they are slower.

 

[happy medium]

They did. There are apples-to-apples sections for each test (same res and settings) and the influence each effect has on performance on the nVidia cards. You did click the link, right?

Hmm, missed that. Thanks.

Seems they didn't use aa at all in those comparisons?
Wonder why?

Alienbabletech ran some benchmarks with 8x aa and with the 2 avalable Nvidea options on and the 5870 came out ahead by a hair. They also note that the 58xx series cards use a different method of AA?

According to the Hard/ocp review turning on the Nvidea options cost the gtx cards 16%.

So add 16% to these benchmarks.
http://alienbabeltech.com/main/?p=17869&all=1

 

[JRW]

Just ordered a EVGA GTX 480, Im guessing it'l be a worthy upgrade from my current EVGA GTX 260.

 

[Keysplayr]

But would that higher tessellation performance manage to overcome the general slowness in the game? I mean, when they start off being slower, a smaller impact from tessellation in Just Cause 2 if implemented might still mean they are slower.

Really can't say. If tesselation "was" used in stead of the CUDA water, I really can't say if it would look better, worse or perform better or worse. So, your guess is as good as mine. Performance could be abysmal or stellar.

And there are many many levels of tesselation that can be used. Metro2033 would be an example of minimum usage of tesselation, would you agree? While Unigine Heaven 2.0 is probably as extreme as tesselation is going to get. But it still seems it keeps "playable" framerates. As devs adopt the Unigine Engine for their games, we "may" see higher usage of tesselation in those games.

 

[ugaboga232]

Unigine is a very inefficient use of tessellation. Just look at the wireframe. One could easily optimize this to run much better. It does serve as a future look at tessellation though. But the current rumors of this summer/falls SI and next years NI looks like a doubling in tessellation in SI and another doubling with NI.

 

[BenSkywalker]

They have had tesselation for far longer and implementing it in a way that equals Fermi should not be a problem.

It will be a significant increase in transistor cost over their current tesselation implementation and is considerably more complex to implement(although, it also scales better so the penalty is linear based on the layout of the chip and not an absolute as it is with their current setup).

I think the choice to go CUDA over tesselation in this game makes a lot more sense from a development standpoint at this point in time even ignoring nV likely footing the bill. I offer a feature available to ~150Million users or 6Million users- as a developer the only reason to go with a solution that is available to 4% of the potential users seems like something you would only reasonably do if someone else was footing the bill(yes, I fully realize that a lot of those 150Million CUDA parts aren't gaming based, but a bunch of the 6Million also fall into that same category).

 

[ugaboga232]

I doubt all 150 million cuda possible parts can even begin to use it for a visual advantage over tessellation.

Also, ATI has a pretty easy way (supposedly) to increase tes power. In the current 5XXX series, there's 1 tessellator. In SI, there should be 2. In NI, 4. These are the current numbers and combined with the rumor of a change to a 4 shader unit instead of 5 should definitely allow it to compete and possibly beat the nvidia tessellation.

 

[SHAQ]

Did they enable supersampling on the Nvidia cards? Image quality with 2x supersampling is better than 8XAA. 32XAA is still not as good colorwise.

 

[BenSkywalker]

I doubt all 150 million cuda possible parts can even begin to use it for a visual advantage over tessellation.

How well do you think the 5450 is doing with tesselation?

Also, ATI has a pretty easy way (supposedly) to increase tes power.

Easy frequently isn't cheap, and simply increasing the brute force of their current method isn't going to be as robust as what nV offers(not saying it isn't a viable approach, but it isn't going to be directly comparable).

 

[ugaboga232]

How well do you think the 5450 is doing with tesselation?



Easy frequently isn't cheap, and simply increasing the brute force of their current method isn't going to be as robust as what nV offers(not saying it isn't a viable approach, but it isn't going to be directly comparable).

I said nothing about ATi or tessellating cards. However, it is fairly obvious, that few of the cuda cards can do it good enough to mach tessellation. In fact, we basically only have the 260, 275, 280, 285, 295, and fermi's versus 5770, 5830, 5850, 5870, and 5970, and fermi. While the 2XX series does definitely have more cards out than the 5XXX series, it isn't a 150 million to 6 million as you said before.

Why don't you think a doubling or quadrupling of tessellation units when tessellation is the supposed bottleneck is going to increase their tessellating power? Are you saying that Ati is lacking in implementation? Do you think Ati can't come back with a 4X tessellation improvement? Will Nvidia have a brand new architecture when NI is out?

 

[BenSkywalker]

However, it is fairly obvious, that few of the cuda cards can do it good enough to mach tessellation. In fact, we basically only have the 260, 275, 280, 285, 295, and fermi's versus 5770, 5830, 5850, 5870, and 5970, and fermi. While the 2XX series does definitely have more cards out than the 5XXX series, it isn't a 150 million to 6 million as you said before.

http://www.youtube.com/watch?v=gVbKbC_y1wM

Only GTX series?

Why don't you think a doubling or quadrupling of tessellation units when tessellation is the supposed bottleneck is going to increase their tessellating power?

It would, never said it wouldn't.

Are you saying that Ati is lacking in implementation?

Currently? Absolutely, it isn't even close.

Do you think Ati can't come back with a 4X tessellation improvement?

Of course they could, but I thought we were talking about them catching up in tesselation performance?

http://www.bjorn3d.com/read.php?cID=1831&pageID=8804

A 400% improvement in raw tesselation throughput just isn't going to get it done for heavy loads, unless being roughly half as fast is OK.

Will Nvidia have a brand new architecture when NI is out?

If we are talking strictly about tesselation performance, NI would have to be ~800% faster under heavy tesselation loads to match the current nV parts that are available today.

 

[StinkyPinky]

So instead of using tessellation which is an open standard, they use CUDA. And for what?

For a big fat check.

People shouldn't be encouraging that crap, no matter which side you're on.

 

[MrK6]

Graphical improvements seems to be a good reason for most.
And per that same article, they noticed distinct improvements with the feature on. Sure it reduces framerates. Very rare that a feature that worsens image quality reduces framerates, right?

But the card is already about 18% slower than the 5850, why turn on more graphical improvements if the card is struggling? Also, where is this "sweet" water that everyone is talking about? The three screenshots in the review don't provide anything great. Yes, in the first shot the water looks a lot better, but that's because the water on the right looks like something out of 2003. The second and third pictures show absolutely zero difference in IQ, with the exception that the waves in the third picture being rendered further out. Crysis did better water two and a half years ago.

Anyway, buying a slower, but more expensive card to turn on a gimmicky feature is not the definition of a value. My point still stands.

 

[BenSkywalker]

So instead of using tessellation which is an open standard, they use CUDA.

And offer it to FAR more gamers by doing so.

People shouldn't be encouraging that crap, no matter which side you're on.

If they offered it with heavy tesselation for water instead ATi would have been smoked in the benchmarks and people would have bashed them for nV biased code. The way they did it offered the largest benefit to the most gamers. Or, the more likely solution, nV could have not paid them and noone would have gotten it.

 

[ugaboga232]

I doubt any of those extremely heavy loads will be apart of games anytime soon because only Fermi can do it and Fermi doesn't seem likely to attain any significant market share soon.

And these are just rumors. Its supposed to be a whole new architecture.
BTW That 8800gt can only do that at relatively low resolutions (1280 x 1080). A 5450 could do that at low enough resolutions as well.

The point is, no game will use that level of tessellation in the next year or so. Those are just extremely artificial benchmarks that test ONLY tessellation. I doubt a 480 can pull 200+ fps on something like unigine. So while raw power is quite different now, Fermi does not receive 8X fps in unigine. This is why I don't like drawing conclusions on gaming from single-faceted benchmarks.

Edit for BenSkywalker's newest post: It would offer it to far more gamers, but only a very very small fraction could run it at acceptable settings. Offering != Being able to use it in real-life settings.

Also, ATI doesn't get that smoked with real-world levels of tessellation. You can't just say that turning feature Y that uses performance stastic X makes ATi soo much faster than Nvidia because its 8X faster in synthetic benchmarks that test that exact statistic.

In conclusion, just because Fermi has a large theoretical tessellation performance advantage doesn't mean it will show that same advantage in games with tessellation because:

1. Fermi can't do max tessellation + effects that well
2. Very few (10,000 or less) cards can do that level of tessellatoin
3. The difference becomes much less when you do more than just tessellation

 

[BenSkywalker]

BTW That 8800gt can only do that at relatively low resolutions (1280 x 1080). A 5450 could do that at low enough resolutions as well.

5450 runs CUDA real well does it? Also, what exactly does the resolution have to do with it?

The point is, no game will use that level of tessellation in the next year or so.

Exactly, for right now it is CUDA or nothing.

The second and third pictures show absolutely zero difference in IQ

Fired up the shots figuring they weren't ones suited for showing the differnces, you may want to get your eyes or monitor checked if you don't see the differences.

Edit for BenSkywalker's newest post: It would offer it to far more gamers, but only a very very small fraction could run it at acceptable settings. Offering != Being able to use it in real-life settings.

Did Orton divine what 'real life' settings were for all gamers on the face of the Earth? What did he tell you acceptable settings were?

Also, ATI doesn't get that smoked with real-world levels of tessellation.

Real world in game level of tesselation isn't close to what is needed to mimic the CUDA effect seen in JC2. Go ahead and check out how fast ATi runs that water demo you are thinking of(it isn't pretty).

1. Fermi can't do max tessellation + effects that well
2. Very few (10,000 or less) cards can do that level of tessellatoin
3. The difference becomes much less when you do more than just tessellation

In this discussion our options are using the incredible level of tesselation you are saying isn't viable, using CUDA, or not doing it at all.