Seriously, fps over % as an indicator of comparative performance? :facepalm: In some titles they're about as fast, on others 580 is faster. 580 has it's downsides as well: it costs more and is an oven compared to 6970. But the real deal is that you can mod 6950 to 6970, offering a very good €/perf-ratio. :celebration:

I don't like mods (in that sense), and anyway it's also probable you can mod the 570 in order to become the 580, as it's the same chip.

And what "%" is about?

I would agree with you if this was the 4xx generation for Nvidia. Those were an oven and incredibly power hungry.

They definitely addressed that in the 5xx series with better power management and cooling. They will assuredly be cooler still when they move to a <40nm process. They may still not be as good as AMD on power/temps, but it makes them actually worth getting. I would never recommend a 4xx series, but I would a 5xx series.

Also, in terms of raw performance, I have issues with AMD;s decisions. They made a huge leap in the 6xxx series, at least, by upping their geometry throughput from 1 primitive/clock to 2 primitives/clock, but Nvidia has done 4 primitives/clock for a while now. They also don't seem to care about tessellation very much. The new architecture helps, with the geometry throughput increase, but Nvidia is still far ahead with tessellation performance.

For those reasons I align more with the way Nvidia has done things this generation, regardless of price.

It's a simple bios-mod which works always. 570 can't be modded the same way (if at all) because it has hardware modifications to prevent it, besides it's has different PCB as well. You seem like a really devoted fan.

Well, the 7xxx series should come out before Skyrim, so you might consider waiting for those.
At least I am.

edit: don't know what's up from Nvidia till then, but I'll be on the lookout.

/ignore.



Well, the thing is to buy smart. I've had always the x600 model of each Nvidia generation (6600, 7600, 8600, and my current 9600). I bought each one of them when it only costed 150 €, and I've never had any performance issues.

I'll buy the GTX 580 when it falls into the 300-350 range. It's a bit of an exception, but those are exceptional times as well



Uuuuh, ATi showing their new teeth so early?

Any news about what can we expect?

Actually, strike some parts of my last post, where I assumed you were right about this fact.

Anandtech http://www.anandtech.com/show/4061/amds-radeon-hd-6970-radeon-hd-6950/24. Except for Furmark which is synthetic, and AMD underclocks their GPUs when they detect Furmark. (Look up PowerTune which tries to keep the card TDP within range at all times)

So to be fair to Benrahir, which you called a "fan boy" (yet it got censored to "really devoted fan", you seem like you may have a biased AMD preference yourself.

Well, it's still very power-hungry compared to ATi's offerings. There's a clear difference in theorethical tesselation power, but in actual applications it doesn't seem to be that big. 6970 fares well in Uningine and Metro 2033, but gets beaten at Lost Planet 2, but most of it could be due to another factors altogether. If you wan't the fastest then 580 is the clear winner, otherwise it gets a lot more complicated.

Why did you mention <40nm? 28nm will not have anything to do with these cards.


7xxx doesn't come out before 28nm comes out from TSMC and that ain't happening this year unless hell frozes over. Same goes for nV. :shrug:

Well, for one, see my last post about you not being right about the "oven" comments.

And two, I specifically said their cards will be cooler when they move to sub-40nm. Both companies' cards. And it makes it less and less of an argument with each downsizing.

And I mentioned that because both companies will probably have new cards out around when Skyrim is released.

To be fair also, I'm an AMD fan too...in terms of CPUs. I hate how awfully expensive are Intel' ones.

So I'm a Nvidia-AMD guy. Weird combo, I know, but I'm used to those

I am a fan of neither. I go where the performance/features go. You output twice the theoretical geometry (580 vs 6970), you get one vote in your favor. Then I tally up all the things that matter to me and make a choice that generation.

I haven't preferred AMD's CPUs in many generations for that reason. I will reassess this when Bulldozer comes out, but I doubt they will take any crowns.

Yeah, sadly I know that AMD CPU's aren't top of the line. That's why the only component I know (nearly) FOR SURE that there will be in my new computer will be a GTX 580 (I was looking about the supposed Radeon HD 7xxx series and by far I haven't run into any info, so I label it as hype for now).

And about all the other components, I'll wait until October' prizes, and choose as wisely as possible. I don't discard moving to a nice Intel CPU for 200 €, if it outperforms the equivalent 200 € AMD CPU.

nVidia has similiar - albeit more primitive - measures for decreasing power comsumption under certain programs (=Furmark). The temperature difference is marginal but look at the actual power comsumption. There's huge differences. Less stress for PSU, less heat pumped to the room and smaller electricity bill and better possibilities for overclocking especially with custom coolers. 69xx wins perf/w and perf/€. Like I said earlier, 580 for teh über-performance and in most cases 69xx if other things are important as well.

That "really devoted fan" was actually quite funny. :blush:


Oven is an oven even if it's better cooled.

Why do you think they would shrink current designs to 28nm? Both are bringing at least evolutionary architectures over present models and it makes comparing them based on present cards somewhat useless.


I take it that you're not familiar with overclocking?

Dude you're thinking I am writing things that I haven't. And are just over complicating whatever you are saying with semantics. In my original reply, myabe I should have clarified bump mapping rather than writing bump map. All I am saying is that you won't find bump maps apart from parallax maps in games. As you basically implied that all assets already do have bump map, while I pointed out they do have normal maps, but this is not a greyscale height map. they apparently do not convert to height/bump/displacement map that works with tessellation well. Apparently it actually matters having an actual bump map. Unless that is incorrect. I am only going on this dev talk http://www.slideboom.com/presentations/159293/TGC-2010-Practical-Triangle-Tessellation and info on polycount.

The relevent points:
"Slide 20
3- Deriving Displacement Maps from already constructed assets proved tricky. CrazyBump helped derive Displacement from Normal Maps but still not the solution for various reasons. Approach failed for 3 reasons: Using Existing Assets"
"Conclusions Tessellation of triangles can add to existing pipelines, but it’s not a free change. Reusing existing art assets is harder than most anticipate."

Obviously, I don't know how tessellation is going to work on all engines or if better solutions for converting normal maps to displacement maps will happen, Just saying apparently converting tangent space normal maps to displacement maps for tessellation produces less than ideal results.

tl;dr

It's best to have tessellation in mind when creating assets, it's a lot of work to prepare existing assets to tessellation, you need actual displacement maps as apparently converting normal maps to fulfill this role is crap in some way. And the only assets usually found with bump maps that aren't normal maps, would be parallax maps. And those are few and far between.

Ugh. I've already seen those very same slides. Note they don't even explain why converting the normal map "failed". Especially in regard to my Oblivion example having existing assets for use in tessellation, I never said anything about it looking pretty. But converting into a height map from normal map is more than possible, as the slide shows. But they fail to explain why it "failed".

Secondly, I never said anything about adding tessellation to Skyrim after the fact. They would be building their meshes to support all renderers, including DX11, and their normal maps could serve as height maps just fine. If they wanted to however they could bake displacement maps from their original sculpts in zBrush and easily add those assets into the PC version. And on top of that tessellation can use a "density map" to focus on specific areas to tessellate, if they wanted to make those.


Doesn't mean it can never work for anybody. And in response to their "UVs were set up incorrectly for displacement"... those UVs are UGLY and not artist friendly. Yes, artists use them but you should use nicer UV mapping. We don't know what type they are using on Skyrim. If they use zBrush for a lot of their character modeling, they should also be using UV Master to create artist friendly unwraps. The "issues" the slide brings up (which is from the Gamebryo makers, btw, hah!) are more for complex meshes. If we're talking about simple walls/terrain/etc. there is going to be little of those issues. Simply making the flat surfaces of objects tessellated would improve the look so much.

EDIT:

Oh, and their constraints for the project were hilarious. 1 month, 2 programmers, only 1 of them programming for tessellation. No wonder they couldn't get everything done.

I'll also clarify my statements above by saying that when I said "pretty" I mean "perfect". All forms of bump mapping have their artifacts. With bump, normal, and parallax occlusion, they all have silhouettes/edges to deal with. I'd take some seams in tessellation on non-optimal meshes over parallax occlusion mapping. Not to mention POM is hideously expensive for the effect. Tessellation is cheaper.

Edit #2:

They should also have no issue implementing PN Triangles tessellation or Phong tessellation. That could at least smooth out some of the very low poly geometry.

I still think you are making it sound too easy. Well to go the whole character + everything and the dog being tessellated wouldn't be considered easy.

But then loads of things can just be done no sweat, water, anything they were going to use parallax maps on, all manner of walls and surfaces. That could be very doable. UV maps and hard edges would be a non issue for those kinds of assets.

I still doubt any efforts being made to that end for some reason. :shrug:

Well I am ready to drop in on DX11 card when the game is released
hopefully get 5x0/69x0 on discount when 600/7000 series are released