Vista delays = D3D10 GPU delays ?

[Pete]

Seems to me the earlier ATI/NV/anyone gets their DX10 GPUs out, the faster they can win OEM deals. Assuming DX10 GPUs can work in XP with DX9 drivers, I see no real reason for any hesitation on ATI's or NV's part. Well, unless their subsequent parts are delayed.

More importantly, where does this put Halo 2 for PC? Who will bundle it first?!

 

[Geo]

Seems to me the earlier ATI/NV/anyone gets their DX10 GPUs out, the faster they can win OEM deals. Assuming DX10 GPUs can work in XP with DX9 drivers, I see no real reason for any hesitation on ATI's or NV's part. Well, unless their subsequent parts are delayed.

More importantly, where does this put Halo 2 for PC? Who will bundle it first?!

I wonder how gddr4 plays into all this as well. That might be a driver for one more round of DX9 rather than going earlier with D3D10 (do I have to use that <he whined>? It's one character longer than "DX10"!) If the current gen can be shown to be somewhat bandwidth starved. . .

Process availability in general could play in too. I don't see ATI doing R6 flagship at anything bigger than 80nm given the current size of R580. Less willing to predict that for NV, tho it might be true --on the one hand they would appear to have more room to play with, on the other I'm not terribly willing to try to extrapolate G80 die size from G71 anyway.

Edit: Tho come to think of it, there's the yowza-wowza factor to consider. If Vista native parts are the thing you want people to get excited about, you might want to save some performance delta rather than eat it up with yet another refresh.

 

[Ghost of D3D]

I don't think this Vista delay is significant, to both MS as well as, possibly of course, the IHVs.

Will probably screw up significantly more for some ISVs' plans (and their publishers') if any of them have the Christmas target dates.

 

[chavvdarrr]

Microsoft is on track to complete the product this year, with business availability in November 2006 and broad consumer availability in January 2007."

http://www.microsoft.com/presspass/press/2006/mar06/03-21WindowsVistaDeliveryPR.mspx
So, what does "business availability" means ? Big OEM's able to bundle new PCs with Vista?
So its not IHV's who are screwed, but enthusiasts - in order to get Vista before 2006 end one will have to buy PC bundled with it.
Who will buy Vista-bundled PC without full-spec D10 hardware?!

 

[SugarCoat]

http://www.microsoft.com/presspass/press/2006/mar06/03-21WindowsVistaDeliveryPR.mspx
So, what does "business availability" means ? Big OEM's able to bundle new PCs with Vista?
So its not IHV's who are screwed, but enthusiasts - in order to get Vista before 2006 end one will have to buy PC bundled with it.
Who will buy Vista-bundled PC without full-spec D10 hardware?!

Well thats not technically true, while you're not suppose to be able to get them many places sell OEM versions of Windows. XP64 was never suppose to be for retail sale for example, yet you can buy it. Personally i dont buy the release schedule being seperated by a notable margine, wouldnt make any sense.

As noted before ATI did bring out the Radeon 9700 Pro that was DX9 and ahead of MS' release date.

Yes but the R300 wasnt using a key function and technology only exploitable by DX9. It worked will with DX8.x. The R600 series on the other hand would need some form of driver emulation i would imagine to give current DX9 computers the idea that it has seperated pipelines in order to work with current Direct X versions. Otherwise you would just get a bunch of errors every time you try to run something. I dont see what the rush is anyway. Of course we all want to know what the next great thing is as i think most are sick of DX9, but reality is DX9 based games will dominate well into 2007 and problably into 2008. ATI, Nvidia and MS have very little reason to rush just to add support for the hell of it. Most people here are just as if not more excited about Unified architecture then they are about DX10 and what that will bring. Only reason i plan on buying an R600 flagship card will be because its the first Unified PC part, whether its further DX9 extention or DX10, doesnt matter.

We are long over due for some new effects in DX but with the amount of lackluster titles in the recent year i'v lost a bit of want to rush into DX10.

 

[DmitryKo]

Untill I get some facts, I still think a fully spec D3D10 card will not work in Windows XP/2003/2000. From the changes D3D10 has its not a upgrade like we have seen with DX5 up to DX9.0C but a total diffent way of doing things. Why do you think theres a D3D10 and D3D9E? Older cards like DX5 to DX8.1 work with DX9.0C so you would think older cards would (DX9.0C based cards) work with D3D10 and not need a D3D9E. From the info thats out now I think a full speced Direct3D10 card will not work with Windows XP and will only work with Vista.

I can hardly see major vendors offering Direct3D10-only GPUs that cannot run WindowsXP, at least not in the next 5 or so years

Pre-Direct3D9 level cards work with DirectX 9.0c runtime because it still supports old DDI (Display Driver Interface), and resource management is handled by the driver. This is stll the case in Vista, but Direct3D9Ex (and Direct3D10) require new DDI that comes with WDDM (aka Basic model) which requires D3D9 and just maps older DDI calls to new APIs.

Considering that WDDM is compatible with any D3D9-level card from ATI or NVidia, it's only natural to make D3D10 features work as extension to D3D9 functionality, just like it was before. Also consider that all current DX9 GPUs can woek with a completely different pipeline, that is OpenGL, despite compelely different programming model and incompatible data structures.

 

[DmitryKo]

So seems that the general consensus is that NV/ATI will get on with their own thing and release the hardware whenever suits them - be that on time or early. Interesting!

I'm willing to bet that D3D10 parts will not ship until Spring 2007 or later

D3D10 is a basic component of Windows Vista. I think there is a refresh for WDDM post-release, but thats more low-level and won't really affect the features that the end-user sees (probably just performance).

It will provide for validation and memory virtualization in hardware (something like AGP/GART), so I guess it will affect performance (apart from improvements already in Basic model), but we'll have to wait until WinHEC 2006 to hear more details on WDDM and WDDM v2 (previously known as Advanced model).
www.microsoft.com/whdc/winhec/trackdetail06.mspx?track=11

WDDM v2 and Beyond

Graphics processors have become extremely powerful in computational capability and flexible in their degree of programmability. Until recently, only a limited number of applications (primarily games and workstation applications) fully used this power. However, increasingly broader classes of applications are becoming interested in using the power of the graphics processor. The Windows Display Driver Model (WDDM) takes the first steps to enabling the graphics processing unit (GPU) to be shared across multiple applications. Improvements are planned for Version 2 of WDDM that significantly increase the ability to virtualize the graphics processor and its memory. This enables true preemptive multitasking of the GPU with full virtual memory support. It will allow many applications to use the GPU simultaneously without sacrificing process isolation or performance.

This session describes the hardware advances that are necessary to enable these new scenarios, including an overview of graphics hardware infrastructure requirements and information on how the operating system and hardware will work together. It also describes the benefits of this design and discusses scenarios that are enabled or improved.

 

[trinibwoy]

Yes but the R300 wasnt using a key function and technology only exploitable by DX9. It worked will with DX8.x. The R600 series on the other hand would need some form of driver emulation i would imagine to give current DX9 computers the idea that it has seperated pipelines in order to work with current Direct X versions.

Your computer doesn't know anything about separate or unified pipelines. All it knows is an API. The driver sits between that API and the hardware. All they would need is a DX9 driver, just like we have today.

It is possible to build a unified DX9 part (evidenced by Xenos), and a non-unified DX10 part. It's the API that's unified, not the hardware.

 

[Maintank]

Untill I get some facts, I still think a fully spec D3D10 card will not work in Windows XP/2003/2000. From the changes D3D10 has its not a upgrade like we have seen with DX5 up to DX9.0C but a total diffent way of doing things. Why do you think theres a D3D10 and D3D9E? Older cards like DX5 to DX8.1 work with DX9.0C so you would think older cards would (DX9.0C based cards) work with D3D10 and not need a D3D9E. From the info thats out now I think a full speced D3D10 card will not work with Windows XP and will only work with Vista.

This should be taken care of at the driver level. DX9E is DX9 for Vista afaik.
I dont see why DX10 cards wont work with DX9, and if they was a limitation it would be suicide for Nvidia and ATI as their sales would slump in a major way.

 

[aaronbond]

http://www.gamespot.com/features/6143883/p-4.html

ummm in the video he clearly states that he "thinks" DX10 titles will be shipping on Vista release day. Let's hold him to it! lol

 

[Tahir2]

http://www.microsoft.com/presspass/press/2006/mar06/03-21WindowsVistaDeliveryPR.mspx
So, what does "business availability" means ? Big OEM's able to bundle new PCs with Vista?
So its not IHV's who are screwed, but enthusiasts - in order to get Vista before 2006 end one will have to buy PC bundled with it.
Who will buy Vista-bundled PC without full-spec D10 hardware?!

Actually the smaller companies that build OEM machines are screwed the most, the likes of Dell and HP are sorted and the enthusiast crowd probably has BT.

 

[Love_In_Rio]

Any true in this ? http://www.theinquirer.net/?article=30516

 

[Geo]

Any true in this ? http://www.theinquirer.net/?article=30516

Maybe, just maybe, I could believe that 60% of some high level metric grouping of the various pieces (however defined) are still considered not-ready-for-release and still require some tweaking. Maybe, since stuff seems to come together into final form at the end anyway on major software projects. But going to "rewrite" 60% of the code this late? No f'in way.

 

[Tim]

Yes but the R300 wasnt using a key function and technology only exploitable by DX9. It worked will with DX8.x. The R600 series on the other hand would need some form of driver emulation i would imagine to give current DX9 computers the idea that it has seperated pipelines in order to work with current Direct X versions.

Repeat this 100 times:

"Direct 3D 10 does not unify shaders"

 

[Tim]

Your computer doesn't know anything about separate or unified pipelines. All it knows is an API. The driver sits between that API and the hardware. All they would need is a DX9 driver, just like we have today.

It is possible to build a unified DX9 part (evidenced by Xenos), and a non-unified DX10 part. It's the API that's unified, not the hardware.

The API is not unified, It still has seperate vertex and pixel shaders. It is the requirements that are identical for both vertex and pixel shaders, this makes unification on the hardware side make more sense than before.

 

[trinibwoy]

The API is not unified, It still has seperate vertex and pixel shaders. It is the requirements that are identical for both vertex and pixel shaders, this makes unification on the hardware side make more sense than before.

So the actual method signatures/names in DX10 for dealing with vertex and pixel shaders are different but the instruction sets for both are equivalent?

 

[Demirug]

So the actual method signatures/names in DX10 for dealing with vertex and pixel shaders are different but the instruction sets for both are equivalent?

There is only one universal Shadercompiler function (D3D10CompileShader)but you have to specified the profile (vs_4_0, gs_4_0, ps_4_0).

After this it goes different ways. There are different methods to create the shaders. CreateVertexShader, CreateGeometryShader and CreatePixelShader.

For every of the three shaders there are 8 functions to set and get the current shader, the samplers, the views and the constants. The only differ in the first two chars. VS, GS, PS.

 

[trinibwoy]

Thanks a lot Demirug. That clears it up a bit.

 

[rwolf]

Maybe, just maybe, I could believe that 60% of some high level metric grouping of the various pieces (however defined) are still considered not-ready-for-release and still require some tweaking. Maybe, since stuff seems to come together into final form at the end anyway on major software projects. But going to "rewrite" 60% of the code this late? No f'in way.

I doubt that Microsoft can rewrite 60% of the code and release next year. I think the inquirer has misquoted someone.

 

[DmitryKo]

http://blogs.msdn.com/larryosterman/archive/2006/03/24/560147.aspx

Btw, on the /. article (for those that click on the link), I have NO idea where that 60% thingy came from, as best as I can figure, the site that published the article pulled that information totally out of their hat. In addition, if you think about it, it's a nonsensical comment. According to the wikipedia, Windows contains 40 million lines of code (I have no idea if that's accurate or not). But assuming that it is, and assuming that Vista had the same amount of code that XP had, that means that Microsoft would be re-writing 24 MILLION lines of code. In two months (Vista's only slipped for 2 months according to this press release). Now Microsoft programmers are good, but they aren't THAT good. Anyone who's ever worked on a project that involves more than a thousand or so lines of code understands how utterly laughable that is.