GeForce GTX 460 might launch June 1st - specs

[RavenXXX2]

To be fair the 480 can push some stunning framerates in certain games and it's tessellation performance crushes ATI's at the moment " not surpring as ATI were first to the market with DX11 support and Nvidia had plenty of time to optimize for it " of course that means nothing much in current and announced titles on the way that support DX11 as tessellation will not be used that extensively for sometime to come, by then we will have ATI cards out that will do the job even better.

The big problem with fermi is the powerdraw, it is just ridiculous the amount of power it draws compared to the competition, this leads to higher temps as well, with the chip being so big and leading to poor yields, supply is currently nonexistent, Nvidia need to get on to a new process ASAP.

 

[jigger]

The problem is the card just is not ready. Really the GTX480 needed to be 200watt ~, beat a 5890 and come in around £400 just to have levelled the playing field.

The core has had to be heavily cut back and have clock speeds reduced just to increase the yields to a laughable level. Considering the time difference between it and the 5000 cards Fermi is already a dinosaur. The only success (if you can buy one) it has is it beats a card that costs £150 less.... Whoopee.

 

[Greebo]

To be fair the 480 can push some stunning framerates in certain games and it's tessellation performance crushes ATI's at the moment " not surpring as ATI were first to the market with DX11 support and Nvidia had plenty of time to optimize for it " of course that means nothing much in current and announced titles on the way that support DX11 as tessellation will not be used that extensively for sometime to come, by then we will have ATI cards out that will do the job even better.

The big problem with fermi is the powerdraw, it is just ridiculous the amount of power it draws compared to the competition, this leads to higher temps as well, with the chip being so big and leading to poor yields, supply is currently nonexistent, Nvidia need to get on to a new process ASAP.

Problem is even the gtx480 takes a massive hit with tessellation in games so when games come only in full tessellation niceness, you'll need Nvidia or ATI's next gen card to play them anyway.

As planned the gtx480 with 512 cores and 750Mhz or even 800Mhz would have being a massive boost and perhaps then worth the premium (enthusiasts would have lived with the power draw)

There are some very interesting results of gtx480 under water running at 900Mhz with good temps which gives you an indication of what the planned gtx480 would have performed at.

 

[mel d]

so the next revision of them should be better,cooler and quicker ?

 

[oweneades]

But the 480 has regained the crown. It might Not be the best bang per buck, but it's quite obviously the fastest gpu.

For it to retake the 'crown' it would have to unseat the HD5970.... hmmm..... nope didn't happen.

 

[gurusan]

so the next revision of them should be better,cooler and quicker ?

Hopefully...but ATI will have theirs first. We've just got to wait and see.

 

[Ordokai]

[snip]

[470; 480] are shockingly bad value for money, and it's not even subjective.

This.


Quite frankly, the bottom line is that to even compete these cards need not only run cooler and draw less, but also be priced reasonably ( this is green we're talking about here )

Considering potentially lowered prices of the 5770 and 5850 and more importantly their increased availability come summer proper, the 460 will have a real tough time tackling even the current AMD offerings on all fronts (drain, speed, price, heat).


To be perfectly honest I just can't see the status quo changing any time soon. The AMD refresh, or even a 'next-gen' is inevitable and at the moment it seems green will be caught with their proverbial pants down. Add the move for GloFo into the mix and its looking pretty bleak for Nvidia. But naturally, in the interest of competition I do hope green step it up a notch and roll out something really impressive as their refresh. Am I seeing a potentially lethal 460 SLI ? Hopefully...

Red and/or Green and/or Retailers have been pretty cocky with the pricing as of late...


- Ordokai

 

[drunkenmaster]

Tesselation crushes AMD's, because it wins one sythetic benchmark and drops its lead in several games when Tesselation is used, hardly conclusive proof. For instance Dirt 2, its further ahead in DX9 and less far ahead when tesselation is enabled, it takes a FAR bigger performance hit enabling tesselation than AMD.


Either way tesselation is crap for now, it "is" very good, its needed but as its only just starting to be implemented, in a not useful way, and probably not coded as efficiently as it will be with a little practice and real world experience of programmers using it, we'll be 2 generations down the line before you get anything even approaching uniengine level of IQ increases. If Nvidia is still around then, their tesselation performance won't be the same as today, and neither will AMD's.

If you buy the 480gtx for tesselation, you're dafter than if you buy it just for Physx.

The problem with it being "faster" is that it costs more than 5850's in crossfire, with performance the 480gtx can't touch to save its life.

Its not at all difficult to have a board with 2x8 pci-e at the moment, if you go AMD you only need to be spending £70 on a mobo to get great overclocks and crossfire.

Raw performance is just about the worst reason to buy the 480gtx, because at its price, its incredibly slow. For £450 you can have well over 50% higher performance and about the same power usage, lower temps, better overclocking and you can GET IT TOMORROW if you want.

At £450 theres not a single reason to get it, frankly as I've said in another thread, 2x5850 + a 240gts gives you FAR more performance, physx, cuda, at the same cost and you can get it tomorrow.

At this point there will likely be in the 100's available worldwide a month now, maybe under 100, the 480gtx isn't even available which makes the argument for it even more ridiculous.

If you want raw performance quite literally 5850/70/5970's are the answer, because you can buy them. A 480gtx you can't buy, won't perform very well as an e-mail confirming your pre-order sitting in your mail box for months. Even if you could get one, its quite significantly slower than 5850's for the same price.

The one and only situation to buy 480gtx's is to run 4 of them because AMD can't match the performance(though can get very close), but theres probably 3 people on these forums who can afford to buy 4, and wants to and none of them can get 4 anyway.

However for me, if I was building the ultimate gaming rig, a jet engine under the desk wouldn't be part of it.

 

[Ordokai]

^^^^


The above puts in words -exactly- my feelings when it comes to Nvidia's current offerings.

Very well said mate.


- Ordokai

 

[Rroff]

Tesselation crushes AMD's, because it wins one sythetic benchmark and drops its lead in several games when Tesselation is used, hardly conclusive proof. For instance Dirt 2, its further ahead in DX9 and less far ahead when tesselation is enabled, it takes a FAR bigger performance hit enabling tesselation than AMD.

Its more than just one synthetic benchmark - you have stone giant, heaven, couple of other purely tessellation benchmarks and a couple of videogames - all showing between 60 and 90+% performance advantage to the GTX480 over a 5870. ATI is going to be on the backfoot tessellation wise for awhile too which is kinda ironic as it was their "baby" - doubling tessellation performance on ATI requires quite a bit more die space and really needs a reworking of the engine (which is coming in the 6 series) whereas nVidia can keep beefing up their polymorph engine without significant reworking, its actually cutback atm to shoe horn the design onto 40nm.

Either way tesselation is crap for now, it "is" very good, its needed but as its only just starting to be implemented, in a not useful way, and probably not coded as efficiently as it will be with a little practice and real world experience of programmers using it, we'll be 2 generations down the line before you get anything even approaching uniengine level of IQ increases. If Nvidia is still around then, their tesselation performance won't be the same as today, and neither will AMD's.

Theres not really that much too it - its more or less an extension of your displacement mapping shader, feed in the low poly surface + displacement map stage to the hardware, tessellated results pop out.

If games do start coming out with extensive use of tessellation stages in their shader pipeline people with the GTX4x0 series are laughing, it is actually one of the easiest "DX11" effects to implement.

 

[kylew]

Its more than just one synthetic benchmark - you have stone giant, heaven, couple of other purely tessellation benchmarks and a couple of videogames - all showing between 60 and 90+% performance advantage to the GTX480 over a 5870. ATI is going to be on the backfoot tessellation wise for awhile too which is kinda ironic as it was their "baby" - doubling tessellation performance on ATI requires quite a bit more die space and really needs a reworking of the engine (which is coming in the 6 series) whereas nVidia can keep beefing up their polymorph engine without significant reworking, its actually cutback atm to shoe horn the design onto 40nm.



Theres not really that much too it - its more or less an extension of your displacement mapping shader, feed in the low poly surface + displacement map stage to the hardware, tessellated results pop out.

If games do start coming out with extensive use of tessellation stages in their shader pipeline people with the GTX4x0 series are laughing, it is actually one of the easiest "DX11" effects to implement.

To be honest, most examples of tesselation adds very little real detail.

Currently, meshes seem to just be sub-divided between existing polygons, upping the poly count but not upping the details where it's needed.

 

[Rroff]

Its a bit hit and miss currently agreed. Theres some nice examples where it works well tho such as the roof tiles in heaven benchmark, whereas some of the other areas it just looks weird.

I'm not really a big fan of tessellation anyhow.

 

[Ordokai]

To be honest, most examples of tesselation adds very little real detail.

Currently, meshes seem to just be sub-divided between existing polygons, upping the poly count but not upping the details where it's needed.

Indeed, its my observation as well.

Furthermore, and this might just be my ignorance speaking, but the main difference upon comparing dx10 and dx11 versions of heaven [visually] seems to be nothing much more then the addition of some pretty extreme bump-mapping (most clearly evident in the pavement and roof-titles). Which I'm pretty sure we used all over the place way before dx11 hit. Note that heaven is the only dx11 benchie I've tried. What about stone giant ?

Of course, eager to stand corrected.


- Ordokai

 

[Rroff]

With bump mapping tho the surface still looks flat if you view it from any angle other than face on. With tessellation if you view it from the side, etc. it still looks "bumpy".

 

[kylew]

Indeed, its my observation as well. Furthermore, this might just be my ignorance speaking, but the main difference upon comparing dx10 and dx11 versions of heaven [visually] seems to be nothing much more then the addition of some pretty extreme bump-mapping (most clearly evident in the pavement). Which I'm pretty sure we used all over the place way before dx11 hit. Note that heaven is the only dx11 benchie I've tried. What about stone giant ?

Of course, eager to stand corrected.


- Ordokai

Well I've noticed that the heaven benchmark is a bit dodgy, things look bad (like they've made them that way on purpose to make tessellation look better).

The stair section, there aren't even real stairs when it's untessellated.

It's just a sloping plane, which is bizarre.

The stones on the ground look really nice, but they're quite basic tessellation really.

Roof tiles, again they've just used a sloping plane, the tessellated roof looks much better of course, but they look dodgy with it off.

The ship is the most questionable section though as no detail seems to be introduced in to the mesh, it's just the subdividing of the current polygons which is pointless.

It's like tessellating planes that are going to be a flat floor surface (concrete screed for example), there would be no benefit at all in tessellating such an object because it's generally going to be flat, a bumpmap would be as much as you'd need (for the detail in the concrete such as air bubbles, lines/cracks).

Which is exactly what the current implementations of tessellation are currently using.

I dunno, maybe it's just been ATi trying to get devs using tessellation because they were the first to the market with DX11 cards.

That's probably the most likely scenario.

 

[kylew]

With bump mapping tho the surface still looks flat if you view it from any angle other than face on. With tessellation if you view it from the side, etc. it still looks "bumpy".

That's not entirely correct, take parallax occlusion mapping for example (which is a form of bump-mapping) makes objects "bumpy".

Though the main difference is that tessellation alters the object's mesh, whereas bump-mapping and so on, don't.

 

[Rroff]

Yeah the "slopes" suddenly becoming stairs irritates me... a LOT.

The ship had me wondering what they were playing at too - it looks 90% identical yet has tessellation passes in the materials shader - tho I think some bits like the engine and the serrated decor do benefit.

 

[Ordokai]

With bump mapping tho the surface still looks flat if you view it from any angle other than face on. With tessellation if you view it from the side, etc. it still looks "bumpy".

I think we have a slight terminology misunderstanding here. I do the occasional interior design render using raytracing and the bump-maps I apply to surfaces do, very much, add depth to complex surfaces when viewed from any angle, that's the reason I add them. Its much, much easier to overlay a simple bump-map based off lighter/darker areas of the same texture rather than creating a custom geometry for the surface separately.

Also yeah, It does very much seem like the heaven devs made the difference stark on purpose, which completely defeats the purpose of the demo IMO and makes me suspicious of the benchie in general, I mean - Really ? We haven't had simple-slope stairs in AAA tites for years now.

Also, merely adding more 'triangles' without actually upping the detail just hogs up resources without contributing meaningfully to IQ besides making round surfaces 'rounder'. Which granted is something but nowhere near what last year presentations promised.


Downloading stone giant now.


- Ordokai

 

[kylew]

Yeah the "slopes" suddenly becoming stairs irritates me... a LOT.

The ship had me wondering what they were playing at too - it looks 90% identical yet has tessellation passes in the materials shader - tho I think some bits like the engine and the serrated decor do benefit.

They definitely do, but they've placed less visual emphasis on them the way the benchmark runs.

I can't understand what the stair slope is about though, really a simple stair case has to be what? 20 polys for a 10 step staircase? 40 triangles total?

I can only see it as being suspect and intentional considering the effort they've put in everywhere else.

 

[Rroff]

That's not entirely correct, take parallax occlusion mapping for example (which is a form of bump-mapping) makes objects "bumpy".

Though the main difference is that tessellation alters the object's mesh, whereas bump-mapping and so on, don't.

Parallax occlusion mapping will still make texels inside the surface appear to have parallax (and occlude) compared to other texels within that surface but won't have anything extruded above the plane so compared to anything else in the world it looks flat - if that makes sense without a diagram heh. Whereas plain old normals based bumpmapping won't occlude other texels.