[Geeforcer]

http://www.yougamers.com/blogs/nick_r/

[AlStrong]

+151% boost to ePenis when Vista is equipped
Aura of Windows Live
Aura of Brillant Graphics X10
90% chance to blocking a crash

I'd say it doesn't matter if this is Vista only.

[trinibwoy]

I completely agree with Futuremark's decision to make this DX10 only. It's not their fault that DX10 is Vista only. We have no need for another DX9 benchmark and would not benefit from it in any way. Given that they won't share game developers' burden of creating a DX9 path they can focus on DX10 goodies and push those cards to the max.

[dizietsma]

Yes I agree, it should be DX10, or maybe one really technical DX9 game and then X number of DX10. Of course I won't be moving up to Vista any time soon so that is a shame.

I assume DX10 was never for XP purely for financial reasons rather than technical ?

[Albuquerque]

Originally Posted by dizietsma

I assume DX10 was never for XP purely for financial reasons rather than technical ?

You assume incorrectly. DX10 isn't as much about new features as it is efficiency enhancements -- efficiency brought on by major improvements in the Windows kernel and driver architecture.

[AlStrong]

Well, it could be financial in the sense that an overhaul/retrofitting of WinXP would be extremely expensive, but I take it you meant it as a strong-arm tactic to get people to buy Vista.

[Demirug]

D3D10: Yes
Vista Only: Maybe not.
I am willing to take bets that we will see it on a Windows XP system too.

[Albuquerque]

Originally Posted by Demirug

D3D10: Yes
Vista Only: Maybe not.
I am willing to take bets that we will see it on a Windows XP system too.

You might see some hobbled together "emulator" for DX10, but you'll never see it at full speed -- or anything remotely like it -- under the XP OS.

[Demirug]

Originally Posted by Albuquerque

You might see some hobbled together "emulator" for DX10, but you'll never see it at full speed -- or anything remotely like it -- under the XP OS.

I believe it would be possible to get ~90% of the native Vista D3D10 performance. A problem could be the missing “application specific driver optimizations”.

[Albuquerque]

Originally Posted by Demirug

I believe it would be possible to get ~90% of the native Vista D3D10 performance. A problem could be the missing “application specific driver optimizations”.

I'll take your bet and say that you get nothing that close on XP on equal hardware. I'd be pleasantly surprised if they could hit 70% at absolute best, and I'd wager that doesn't come for a year or more.

The only time an emulated D3D10 interface is going to go fast on XP is when using non-D3D10-specific instructions (ie texturing one screen-sized polygon). Batch up about 1,000,000 polygons and throw in about 1000 shaders and XP will grind to a halt long before Vista does.

[Demirug]

Well how long it would take to get it running is another bet. I am not sure why you throw in the poly and shader count. I believe you talk about the higher draw call overhead on Windows XP. This is more a Direct3D than a general XP problem. The draw call overhead with OpenGL is lower. With Long Evans and Mount Peaks it should reduced even more. The final sped you can gain with such an experiment would highly depend on the driver quality. But it should be good for more than 70%.

[Albuquerque]

Originally Posted by Demirug

Well how long it would take to get it running is another bet. I am not sure why you throw in the poly and shader count. I believe you talk about the higher draw call overhead on Windows XP. This is more a Direct3D than a general XP problem. The draw call overhead with OpenGL is lower. With Long Evans and Mount Peaks it should reduced even more. The final sped you can gain with such an experiment would highly depend on the driver quality. But it should be good for more than 70%.

And as OpenGL continues to reduce overhead on the XP side, Vista's driver quality will continue to improve. The problem is, you aren't always dealing with API overhead, you're also up against kernel and driver overhead -- something that will not change on XP.

You will never get 70% speede out of any DX10 XP wrapper using true DX10 featureset. And I don't understand how you think OpenGL of any flavor is going to map directly to DX10 calls...

[Demirug]

Originally Posted by Albuquerque

And as OpenGL continues to reduce overhead on the XP side, Vista's driver quality will continue to improve. The problem is, you aren't always dealing with API overhead, you're also up against kernel and driver overhead -- something that will not change on XP.

The OpenGL overhead should be reduced with Long Peaks and Mount Evans later this year on Windows XP, too.

Originally Posted by Albuquerque

You will never get 70% speede out of any DX10 XP wrapper using true DX10 featureset. And I don't understand how you think OpenGL of any flavor is going to map directly to DX10 calls...

There would be no direct mapping but beside of some fixed function render states the D3D10 feature set is mostly defined through shader model 4.0. Converting this 4.0 shaders to OpenGL shaders would cause some higher setup costs but you have to pay this only once until they are compiled and linked.

Anyway I am sure that we will see 3DMark 07 on XP and then we can check who is right.

[Albuquerque]

Originally Posted by Demirug

The OpenGL overhead should be reduced with Long Peaks and Mount Evans later this year on Windows XP, too.

Yes, I understand that. You missed the part about driver and kernel overhead. Care to get me up to date on how OpenGL is going to fix the XP driver model and ring-level-zero kernel functions? Maybe I missed that part?

Originally Posted by Demirug

There would be no direct mapping but beside of some fixed function render states the D3D10 feature set is mostly defined through shader model 4.0.

Ok, so what about stuff that doesn't map over? You can't just reissue a "different" instruction from DX10 to OpenGL; recompiling a shader might solve a few cases, but certainly far from all of them. And in those far-greater cases, you're going to be stuck issuing multiple commands and/or processing it entirely through the CPU.

Originally Posted by Demirug

Anyway I am sure that we will see 3DMark 07 on XP and then we can check who is right.

We shall see

[Demirug]

Originally Posted by Albuquerque

Yes, I understand that. You missed the part about driver and kernel overhead. Care to get me up to date on how OpenGL is going to fix the XP driver model and ring-level-zero kernel functions? Maybe I missed that part?

OpenGL ICDs already use nearly the same kernel driver model on XP that Vista use for Direct3D driver too. That’s the reason why the overhead is already much lower for OpenGL.

Originally Posted by Albuquerque

Ok, so what about stuff that doesn't map over? You can't just reissue a "different" instruction from DX10 to OpenGL; recompiling a shader might solve a few cases, but certainly far from all of them. And in those far-greater cases, you're going to be stuck issuing multiple commands and/or processing it entirely through the CPU.

EXT_GPU_Shader4 looks very promising for me. Remember that even Shader Model 4 compiles to a binary shader language. This language is assembler like and supports general basis operations. Not that hard to convert after you have understand the binary encoding system.

I agree that you will never get 100% perfection with an API to API Wrapper. Therefore I am already saying that such solutions would not bring Direct3D 10 to another OS. They only provide a way that could allow D3D10 applications run on another host system than Vista.

[Entropy]

Originally Posted by Demirug

I believe it would be possible to get ~90% of the native Vista D3D10 performance. A problem could be the missing “application specific driver optimizations”.

[Albuquerque]

Originally Posted by Demirug

OpenGL ICDs already use nearly the same kernel driver model on XP that Vista use for Direct3D driver too. That’s the reason why the overhead is already much lower for OpenGL.

Obviously I bolded the wrong part for you, so let me try once more:

Yes, I understand that. You missed the part about driver and kernel overhead. Care to get me up to date on how OpenGL is going to fix the XP driver model and ring-level-zero kernel functions? Maybe I missed that part?

Your ICD solves nothing about how the XP kernel is going to force you to access the video driver. XP's kernel interaction with drivers is far different than Vista's, and not in a good way.

Originally Posted by Demirug

EXT_GPU_Shader4 looks very promising for me. Remember that even Shader Model 4 compiles to a binary shader language.

Again, this only solves problems when you're talking about shader power. What about virtualized resources? What about creating buffers for render targets? Sure, both API's have this functionality, but they provide it in different manners and handle them differently too. There are things you're blatently skipping over...

Originally Posted by Demirug

I agree that you will never get 100% perfection with an API to API Wrapper. Therefore I am already saying that such solutions would not bring Direct3D 10 to another OS. They only provide a way that could allow D3D10 applications run on another host system than Vista.

What? You're saying that this solution wouldn't bring DX10 to another OS, but would allow DX10 apps to run ... on.. another...OS?? Huh? You're confusing the hell out of me. What are you trying to say?

[Demirug]

Originally Posted by Albuquerque

Your ICD solves nothing about how the XP kernel is going to force you to access the video driver. XP's kernel interaction with drivers is far different than Vista's, and not in a good way.

It’s very different for Direct3D drivers. In the case of OpenGL ICDs the changes are less significant. The even loss some freedom as they have to give more control over the memory and the execution time to the graphics subsystem in the kernel. The kernel overhead was never a big problem for OpenGL on Windows XP. OpenGL have an another overhead problem that is caused by multiple extensions for the same GPU part.

Originally Posted by Albuquerque

Again, this only solves problems when you're talking about shader power. What about virtualized resources? What about creating buffers for render targets? Sure, both API's have this functionality, but they provide it in different manners and handle them differently too. There are things you're blatently skipping over...

OpenGL resources are already mostly virtualized as OpenGL was designed as a client server model. FBOs solve another big part of this problem. I agree that there are some black holes but without some real world D3D10 applications it is hardly to say how intensive this parts of the D3D10 applications will make use of this area.

Originally Posted by Albuquerque

What? You're saying that this solution wouldn't bring DX10 to another OS, but would allow DX10 apps to run ... on.. another...OS?? Huh? You're confusing the hell out of me. What are you trying to say?

Direct3D 10 is more than just the API. There is a DDI, too. Getting D3D10 work on another OS then Vista means that you have to make sure that the drivers would work too. I am sure we agree that this is hardly possible for most OS. Maybe ReactOS will handle this task. On the other Hand Direct3D 10 applications would not care about the DDI. They need only the API and most would not even need everything. Therefore it would be easier to get them running on different host systems. It is only a fine difference when I am talking about “getting D3D10 running” and “getting D3D10 applications” running but it is significant when it come to possibilities and compatibilities questions. To make sure that you understand me right. I am never claimed that it is possible to provide a 100% compatible Direct3D 10 version for Windows XP for anyone else then Microsoft. I even believe that if they had decide to go this way they could not do it but game developers have to test against an official version anyway. I am only claimed that it would be possible to get applications (possible not everyone) that need Direct3D 10 running on Windows XP.

[Albuquerque]

Yeah, I suppose that does make more sense; thanks for the deeper explanation on DX10 vs other OSes thing.

I'm still not anywhere near agreement with your performance expectations for DX10-ish stuff on OpenGL under WinXP, but only time will tell.

[ANova]

Originally Posted by Albuquerque

You assume incorrectly. DX10 isn't as much about new features as it is efficiency enhancements -- efficiency brought on by major improvements in the Windows kernel and driver architecture.

The efficiency improvements are done through the updated directx api, the kernel or driver architecture should have little to do with that aside from maybe small improvements. I simply don't buy the "well it would have been impossible to port to XP" argument. Nothing is impossible, though it may have cost them more and I'm sure having it as an exclusive feature to Vista was a major incentive as well.

[Albuquerque]

Originally Posted by ANova

The efficiency improvements are done through the updated directx api, the kernel or driver architecture should have little to do with that aside from maybe small improvements. I simply don't buy the "well it would have been impossible to port to XP" argument. Nothing is impossible, though it may have cost them more and I'm sure having it as an exclusive feature to Vista was a major incentive as well.

There are more than certainly things in the Vista kernel and driver model that are direct results of DX10's ability to do things in a more efficient manner. Just like the thread you dropped into over in Hardware and Software -- just because you don't understand it doesn't mean it doesn't exist.

Even OGL has to go through the windows kernel and the video driver, which means even OGL is going to be hampered by the inability to quee and batch-issue IO requests and intelligently handle context switching and load balancing on CPU resources.

Could DX10 have been "native" on XP? Sure. If the next service pack included an entirely new kernel, and required a video driver update to even function correctly. You think people are chastizing them now about Vista? Imagine the fallout from SP3 when it broke nearly as many machines that were previously working...