-
Competitor rules
Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.
GTX480 with unigine bench and fraps
- Thread starter RavenXXX2
- Start date
Caporegime
- Joined
- 18 Oct 2002
- Posts
- 33,188
It looks like the 470gtx, you know the only card Nvidia can make, that uses 60% more transistors, at 60% die size cost, and about 1000% the cost right now(to produce for Nvidia, just the cores, the memory costs more((only because there is more )), as does the pcb and power circuitry), but the 470GTX probably won't beat the 5870 in uniengine.................
So for someone who seems to want to laugh at AMD I really don't see how you can. Nvidia made a core 60% bigger, intending to be a full 60% faster in games, it looks like they can't manufacture a card that can come even remotely close.
As for your technical knowledge goating, no, no one on earth agree's with you, every single leak, from any side, from any partner, from anyone says this performance is NOT upheld in games.
I know of no one on earth except you who believes uniengine represents and accurate gaming simulation, nor do Uniengine pretend it does. Its a sythetic benchmark highlighting tesselation.
Judging by some of your posts, it would seem you don't even understand tesselation, nor the ATi hardware surrounding it
This, is 100% incorrect, on every single last level. Firstly, pumping out polygons suggests the ATi hardware somehow draws the polygons, it doesn't. The tesselation unit, really just offers hardware decoding of an compressed highly detailed bumpmap, the shaders must fill each polygon and work out the lighting and the like for each one.
Likewise, not a single person that means anything, has suggested Nvidia's implementation will be gimp its shaders, yet again showing your entire lack of understanding on the subject matter.
The tesselation work is NOT DONE ON THE SHADERS, it is done in the new multi purpose rendering cluster, or, rasterising, I forget what the called it. However, this new unit which gets smaller on lower end cards with less clusters, doesn't have dedicated tesselation hardware, or nothing apparent yet. IT would seem to be very general hardware that will likely be doing many other things in real game situations. It is that hardware, and NOT the shaders that will likely be heavily pushed by other features allowing far far less free power for tesselation to be run.
The polygon pushing thing, well, you might have worded that badly and meant something different, your evaluation of how intensive the uniengine benchmark is, is simply incorrect and your claims of what people think the problem with Nvidia's implementation of tesselation, are completely and utterly wrong.
AS for tesselation being AMD's baby, it is, they've wanted it for years, it offers faster performance for the same bumpmap quality, infact almost all the features you claim AMD hasn't had the power to run would have run fine.
You keep missing minor little things when you make your claims, a vast majority of features in dx10/10.1(moreso) and dx11 are for SPEEDING UP operations. The thing is, as you speed up operations, this allows you to add more things, be it particles, effects or more detail.
Tesselation is something that could will ALWAYS create the SAME QUALITY bump map FAR faster than can normally be done. Tesselation can very well be used that way to optimise one thing, to give dev's more power to use elsewhere. Yes a 2900xt could have sped things up with tesselation, no it couldn't have run the tesselation in the uniengine demo.
Get it straight, the uniengine demo does NOT replicate a normal bump map and do it faster with tesselation, it creates a massively more complex bumpmap which would always take more power to run, but its also a bump map that done normally would not work.
Its very easy, flick between tesselation on and off, its not the same detail level, if they did the higher detail tesselation version, without tesselation it would run at 2fps, tesselation INCREASES performance for the SAME quality level. Most dx features tend to do that, not all, but most.
You can always use faster methods to bring higher detail models into realistic performance levels, but you can always go for a very small detail increase at the same overall power cost, or go for the same detail level as with a little power saving.
Take in point Assasins Creed 1, with the SAME detail level, it wasn't slower in dx10.1, it was 25% faster. The thing is, as time moves on you get new features, rather than push a game with 100fps to 125fps, most dev's use the optimisation, which pushes the framerate up, then uses the headroom to add more detail, which brings performance down.
You can argue all you want that the 2900xt is crap, but its basic shader structure is whats in use now, with minor ratio to TMU/Rops changed, its a great card, ringbus is a fantastic thing that offers insane bandwidth, at a cost. Its not bad technology, it was just something they decided wasn't feasible while relying on TSMC. THe massive majority of things in the 2900xt are there today in the 5870, the fundamental architecture is the same.
Its laughable, can you imagine Nvidia making Fermi on 55nm, neither can i, thats what AMD did with the 2900XT. Just like Fermi has dropped 25% clock speeds on the process they designed it for, the 2900xt had a massive drop in clock speeds, on a higher process than it was designed for, with a shorter delay than Fermi has had, and lower power.
Polymorph engine - one sits inline for ever shader cluster... this unit does not get smaller on lower end cards you just have less clusters so the tessellation power is reduced overall.
At the simplest level tessellation works by taking a low LOD mesh, and an associated heightmap, subdividing the faces on that mesh depending on distance and deforming the vertices based on the heightmap.
Tessellation does NOT create bumpmaps - you accuse me of not knowing what I'm talking about but then go on to show you know absolutely nothing about tessellation.
Infact almost everything you've said in that post is completely and utterly wrong and demonstrating a distinct lack of technical knowledge... its going to be a bit hard to address all the points concisely but I'll have a go...
By gimping the shaders I'm referring to the claims you, charlie and a few other parties have made about how the method nVidia use for handling tessellation will reduce non-tessellated shader processing output... which I claim shows a lack of understanding about how nVidia are actually going about it and completely ignorant of the load balancing aspect - and I think the heaven benchmark backs me up on that point.
Tessellation does not work by producing bump maps, it doesn't use some "compressed highly detailed bumpmap" seperate from the non-tessellated shader effect, it takes the same normal/heightmap data thats used for the displacement map or dot3 shader in game but produces extruded polygon data instead of rendering per-texel psuedo depth effects.
Rendering 1 millions polygons via any method has the same rendering time penalty no matter if its the output from tessellation or an original run time loaded high detail mesh - tessellation does not give you faster performance at the same quality as such - it just makes it easier to implement (and as such gives better performance) high detail on the fly with seamless dynamic LOD adjustment. Which results in better performance unless someone writes some complex visibility algorithms and doesn't have the popin/out effects that alternative methods would suffer from.
AC1 and DX10.1 was a unique situation where the deferred shader pipeline made it possible to absorb some of the antialiasing hit. DX10.1 does have some performance increases such as being able to render 8 lights in the same pass compared to only 1 light in DX10 deferred pipeline but thats a whole different story.
Last edited:
Its not unusual for developers to have pre-release versions of software... although a little dubious to use it for public released data like this.
nVidia is probably working with them to provide upto date multi GPU support and I suspect AMD is also.
nVidia is probably working with them to provide upto date multi GPU support and I suspect AMD is also.
OMFG, NV have got the unigine developers in their pockets, how come they are running 1.1 of the bench, when the only one publicly available is 1.0, fermi optimized bench..
If true that says it all
Associate
- Joined
- 3 Sep 2005
- Posts
- 1,670
- Location
- Glasgow, Scotland
What are you talking about? Every implementation I've seen so far has used a displacement map. A dot3 bumpmap is just light normals (hence why its called a dot product '3' map) encoded into a texture in the RGB channels, of course it can't be used for tesselation because it doesn't store gemoetric data, just image space light collision approximations.
Soldato
- Joined
- 22 Nov 2007
- Posts
- 3,763
- Location
- Ayr, Scotland
Jesus, which card is the best bang for buck? Does anything else matter.
Associate
- Joined
- 5 Jan 2007
- Posts
- 550
You know what, I don't think I have ever seen so much blood, sweat and other stuff (cant remember the proper term lol) over the release of a new card.
Also, just how intense the ATI V Nvidia war has got, sure it never used to be like this.
Like others have said, release the card already to hopefully settle things down a bit
Also, just how intense the ATI V Nvidia war has got, sure it never used to be like this.
Like others have said, release the card already to hopefully settle things down a bit
I'm just saying its not some super special data set, its the same info as would be used with older effects. Dot3 normal maps are often baked from the same kinda heightmap data as displacement maps.
Last edited:
Agreed, they only showed one benchmark. We need many different benchmarks. Also, keep in mind that you can replace the 5870 in the graph with an overclocked 5850 and get the same result.
Is that the best they can do? And the ATI 6000 series is on the horizon.