NVIDIA G80 & ATi R600 Info

[kinggost]

nVidia must be working on getting the upper hand back again by bringing out a new card.

Yeh because nvidia want to do that to be just ahead of ati all the time. When ati seem to launch a new card nvidia seem to do one little bit faster. I think it just to catch the top end of the market where people sepnd loads.

When it comes to the g80 i think they wil sit back and wait to see what ati got. Then if its little faster do summot better by bumping the speeds little. Then they can do stupid prices. That where ati gain some of the market back by cheap cards but are really fast.

 

[Goksly]

nVidia must be working on getting the upper hand back again by bringing out a new card. Maybe nVidia want the G80 to come out before the R600 does. Who knows really. We dont. Only the people in thier labs at ATi and nVidia know what really is going on.

Yeah its called the G71 which is out in a few weeks After that nvidia have the upper hand and there are rumours that ATI cannot release the R600 till November due to an agreement they signed with MS over the Xbox360. If thats true, then nvidia have no need to bring out a new product so quickly.

 

[D.P.]

:\ G80 eh, that's twice as many numbers as my NV40 series card.

:Edit: silly me, just realised that's they're totally different numbers althogether.

G70 is NV47, G80 is NV50

 

[D.P.]

I dont see why June/July would be the most logical.... most logical is going by past release dates which is the only 6 monthly cycles: March -> August... but thats with cards being updated... few pipelines here and there etc, the G80/R600 are being built from scratch, so there are going to be problems... So Sept->December is probably more realistic. Also around the launch date of vista

The G80 wont be that much different to the G70 really. Pipelines updates to be DX10 shader model, and there iwll be more shader pipes and ALUs. Maybe they'll sepaarte TMU and have soemthing like 48 Shaders, 32 TMUs, 16 ROPs and 12 VS.

The R600 will be unified so will be more different.

 

[Goksly]

Really? I thought the G80 was unified as well?

 

[D.P.]

Almost definitely the G80 wont be unified, NV has stated many times that it is just not the right time to unify hardware yet. Unified shaders although being efficient use up lots of transistors in the load balancing and are only as fast as the balancing logic. NV view it as too much of a compromise at the moment.

DX10 requires unified shaders in the API but says nothing about the hardware. So to developers the G80 will appear unified, but the GPU will have special vertex and pixel shaders.

Evtaully NV will gou unified, but probably not to G90, or maybe later, when the efficincy gains outweigh the drawbacks.

 

[Goksly]

Ya learn something new every day - and nice to see ATI pushing forward technology for once :}

 

[Antx777]

Here's me, all geared up now to purchase a X1900-XTX - 1001 reviews read/compared - 4 AM bedtimes.., and then rumours of this come along!. So X1900 won't be DX10 compatable?? - also sorry for being a bit dumb, but whats vistas??

 

[Goksly]

Vista is the next operating system from Microsoft (aka Longhorn)... it will be the only DX10 platform (ie support for DX10 wont be extended to WinXP etc).
I thought DX10 wouldnt be supported by any of the current graphics cards, but if what DP is saying is true, then maybe support can be provided by a new API?

 

[Subtlesnake]

I can see where your coming from but however, each compnay try and out do the other. The GTX 512MB cards had the upper hand till the X1900 XT's arrived, now everyone is getting one of them.

nVidia must be working on getting the upper hand back again by bringing out a new card. Maybe nVidia want the G80 to come out before the R600 does. Who knows really. We dont. Only the people in thier labs at ATi and nVidia know what really is going on.

We would just have to wait and see.

Graphics cards are developed several years in advance, according to their own project schedules, so ATI or Nvidia can't just 'make a better card' if they don't already have one in the pipeline. I doubt there's much they can do to speed up projects already in development, either.

The X1900 seems like a direct response to the 7800 GTX 512, but really it was always scheduled to ship in January as a product refresh. This makes sense when you consider that the X1800 line was going to be released in Spring 2005, and ATI did have working cards then, but it took them six months to work out a bug in some third party silicon, massively delaying the launch.

The 7800 GTX 512 isn't a new GPU - it's just a G70 (7800 GTX core) clocked higher and released in limited quantities, so again we see that Nvidia couldn't release a new chip to fight ATI, even several months after the launch of the 7800 GTX.

 

[Duff-Man]

Unified shaders although being efficient use up lots of transistors in the load balancing and are only as fast as the balancing logic. NV view it as too much of a compromise at the moment.

Kind of. Basically the tradeoff with a unified architecture is increased flexibility for decreased performance (relative to transistor count, which is the real limiting factor on a new high-end part).

Like with almost all things, when you move from something designed for a specific purpose to something with increased flexibility you lose the efficiency of the specialisation. I'm still skeptical about a unified architecture, but as always I'll wait and see with an open mind.

 

[D.P.]

Vista is the next operating system from Microsoft (aka Longhorn)... it will be the only DX10 platform (ie support for DX10 wont be extended to WinXP etc).
I thought DX10 wouldnt be supported by any of the current graphics cards, but if what DP is saying is true, then maybe support can be provided by a new API?

No, the way current gfx carda handle resources is not compatible with DX0 at all.
DX10 allow multiple instances of D3D to be run at the same time- e.g. the AERo glas GUI instance and a computer game\modelling package. With DX 9 you can only have one instance running, similalry only OGL or DX- not both. Otherwise the mutliple instances will fight each other for resorces and the GPU has no way of do that kind of multitasking. This requires important OS changes to be properly supported, and in any case MS just aren't interested in supporting DX10 on an OS they no longer support (the end of XP updates ar nigh!).

DX10 will not run on anything less than a DX10 GPU with Vista. However, the major benefits of DX10 are mainly for the developer and not the user. SM4.0 doesn't add too much functionality, just a unification oin the way the shaders are coded. There will also be higher level geomtery shaders, it doesn't directly add anything that can't be done with DX9.C, especially on Nvidia hardware with VTF support.

 

[Subtlesnake]

Kind of. Basically the tradeoff with a unified architecture is increased flexibility for decreased performance (relative to transistor count, which is the real limiting factor on a new high-end part).

Well, we don't know that. Unifying the pipelines has the potential to increase performance, because pixel units aren't stalled waiting for vertex data, and vertex units aren't stalled waiting for pixel data (as is the case in nearly all modern games, which are pixel shader limited).

Now yes, there may be a reduction in performance because each ALU is less specialised, but without quantifying that, you can't make the comparison.