Impending Vista and current GPUs - is SM3.0 enough?

Guden Oden

Senior Member
Legend
To run the new GUI, I mean. Or in other words, will there be features that owners of GF6800s and similar class of cards will not be able to utilize? There's been beta releases of Vista floating around for a good while now, officially from MS and sometimes not so official, has anyone here tested a version with the full Aero GUI, or is that still held back for now?
 
To run the new GUI, I mean. Or in other words, will there be features that owners of GF6800s and similar class of cards will not be able to utilize?

I doubt it. There's nothing going on in Aero that seems all that taxing. A 5200FX for all intensive purposes should be fine. There's a reason why they specify DX9 instead of 10 for UI functionality. Since your GPU will now be a virtualized resource, alot of VRAM is more critical that actual GPU performance.

Vista outperforms XP .. just hope it helps with DX9 games.

I'd be surprised if it was noticeably faster. More of the driver is being moved in to user space from kernel space, so I'd imagine there'd be some performance hits (although that might be offset elsewhere)...
 
Well, I would expect that some cards may show what would seem to be unecessary performance issues with Vista just because they were designed before the OS's GUI. I doubt it will be show-stopping, more like, "Hrm, that's a little bit annoying."
 
archie4oz said:
I'd be surprised if it was noticeably faster. More of the driver is being moved in to user space from kernel space, so I'd imagine there'd be some performance hits (although that might be offset elsewhere)...
Moving parts into user space is one of the biggest reasons why D3D in Vista is faster.
 
Xmas said:
Moving parts into user space is one of the biggest reasons why D3D in Vista is faster.

It wouldn't matter which parts ran in user space or kernel space, so long as thenumber of boundary crossovers are limited. Given that, it's a better idea to do more in user space and less in kernel space, as this aids to limit the sort of crashes/glitches that could affect the entire system. It would be nice to not have to reboot when installing newer drivers.
 
Xmas said:
Moving parts into user space is one of the biggest reasons why D3D in Vista is faster.

I doubt it. The driver was moved into the kernel for performance reasons in Windows 2000. I suspect that they gained efficiencies through the new driver architecture by improving and revamping legacy processes.

I also discovered that under VISTA the GDI is no-longer hardware accelerated. It is software only.
 
archie4oz said:
There's nothing going on in Aero that seems all that taxing.
What about cleartype acceleration? I'd love to get hardware-AA'd fonts in vista...

rwolf said:
I also discovered that under VISTA the GDI is no-longer hardware accelerated. It is software only.
That sounds really bizarre/stupid. Lots of programs use GDI function calls, wouldn't they run at a snail's pace if GDI is done in software (as well as hogging the CPU and slowing down all other processes at the same time)?

I also wonder how java will handle vista, considering it uses opengl for 3D, and vista disables aero if opengl is used in windowed mode. I sure don't want to have all the eyecandy taken away from me just because I got a few jars running in the background... Friggen microsoft and their paranoia!
 
Guden Oden said:
I also wonder how java will handle vista, considering it uses opengl for 3D, and vista disables aero if opengl is used in windowed mode. I sure don't want to have all the eyecandy taken away from me just because I got a few jars running in the background... Friggen microsoft and their paranoia!
Umm, my understanding was that Vista was supposed to contain an OpenGL-to-direct3d wrapper, so that you can still run basic OpenGL applications in windowed mode, even with full eye-candy on the GUI itself; the problem is more that the wrapper is artificially limited to OpenGL1.4 without extensions (=roughly DirectX7-class featureset) and cannot be upgraded to expose any non-base-1.4 functionality. As such, simple Java applications should continue to work just fine in windowed mode, but state-of-the-art OpenGL games won't.
 
archie4oz said:
I'd be surprised if it was noticeably faster. More of the driver is being moved in to user space from kernel space, so I'd imagine there'd be some performance hits (although that might be offset elsewhere)...

I have it on good authority that Vista is faster .. and my source is very reliable.

US
 
We know that Vista is supposed to be faster, but posts on these very boards from users of the Beta have not found it to be so.
 
rwolf said:
I doubt it. The driver was moved into the kernel for performance reasons in Windows 2000. I suspect that they gained efficiencies through the new driver architecture by improving and revamping legacy processes.

It was moved from the Win32 subsystem to the kernel. The Win32 Subsystem is a separate process. Every time a program want to send graphic commands it have to transfer them to the Win32 Subsystem first to the user mode driver and this driver send the raw hardware commandos to the kernel driver. Transfers between different Processes require two mode switches. Because of this Windows NT 4.0 need 6 switches for every block of graphic commands.

Windows 2000 move the whole driver to the kernel an reduced the number of switches to two.

Vista now will move the driver in the process that needs it. You will still need two switches every time this driver sends a block of commands to the kernel. The big difference is that in the Windows 2000/XP model the Direct3D runtime builds this block in a device independent format. The kernel driver has to unpack this buffer, parse everything and build a device depended command buffer. In the Vista model the user mode driver will build the command block already device depended. This reduces the work of the CPU because the command buffer encoding and decoding is gone. Additional the new model let the driver send the block if it is full. The Windows 2000 model requires sending the buffer even if there is room left. This reduces the number of mode switches on Vista. With OpenGL they already use a very similar model since the beginning and can execute more than twice number of commands at the same hardware.

The natural questions why they don’t use the OpenGL model for Direct3D too have a simple answer. The Direct3D model is based on the Windows95 model to make it easier to port drivers. With it less secure kernel architecture the Windows95 kernel was less sensitive about all this problems that we see today with this model on Windows XP.
 
arjan de lumens said:
Umm, my understanding was that Vista was supposed to contain an OpenGL-to-direct3d wrapper, so that you can still run basic OpenGL applications in windowed mode, even with full eye-candy on the GUI itself; the problem is more that the wrapper is artificially limited to OpenGL1.4 without extensions (=roughly DirectX7-class featureset) and cannot be upgraded to expose any non-base-1.4 functionality. As such, simple Java applications should continue to work just fine in windowed mode, but state-of-the-art OpenGL games won't.

Vista contains:
- The OpenGL 1.1 software renderer
- An OpenGL 1.4 to Direct3D 9.0 wrapper
- Support for OpenGL ICDs

All will run at the same time and you can even write programs that use all three in different OpenGL contexts at the same time.
 
Chalnoth said:
We know that Vista is supposed to be faster, but posts on these very boards from users of the Beta have not found it to be so.

My source is a developer on Vista, so have no reason not to believe him.

US
 
I'd imagine SM3.0 would be pleanty for vista. All you really need for what they're doing is a basic pixel shader. It's just GPUs are a lot faster at resizing textures and alpha blending than a CPU would be.

As for performance I'd imagine it would have to do more with the amount of GPU memory then how fast the card is. Since typically nothing on the desktop changes unless you open/close/move/resize a window etc there is no need to completely redraw the scene. I'd imagine each window is saved as a render target and drawn as needed.

This might also be why there are 512MB X1300's on the market. At 1600x1200 with a typical 32 bit desktop that's ~8mb for a full screen window. So assuming the ram is being used for nothing but caching window render targets a 256MB card can store roughly 32 windows before it has to worry about swapping to system ram. So don't open more than 32 windows at any given time.
 
Chalnoth said:
We know that Vista is supposed to be faster, but posts on these very boards from users of the Beta have not found it to be so.
That's not even remotely surprising. Until I broke it yesterday, I was dual-booting 5270 and XP - XP runs circles around Vista when it comes to performance! BUT, and this is the important part, one is a stable mature 5 year old OS, one is only just feature-complete :LOL:

I'd imagine that the true performance of Vista won't start to be known for a while yet - 5308 is feature complete, which means they're now going to be working on bug fixes and optimizations rather than new stuff.

Jack
 
What about cleartype acceleration? I'd love to get hardware-AA'd fonts in vista...
I'm not 100% sure about this, but my understanding is the ClearType is a related but different algorithm. It makes use of particular properties of how text is rendered/displayed, the monitor and how people read text. Lots of special-cases in there compared with current HW-MSAA that's more general.

Anarchist4000 said:
It's just GPUs are a lot faster at resizing textures and alpha blending than a CPU would be
There's also the simple fact of concurrency in here. If you offload the graphics processing to a dedicated but seperate co-processor then you're going to leave the CPU to do more CPU-like stuff.

Anarchist4000 said:
At 1600x1200 with a typical 32 bit desktop that's ~8mb for a full screen window.
Where'd you pull out 8mb from? I can get 14.64mb with a conservative calculation ;)

Jack
 
Demirug said:
Vista contains:
- The OpenGL 1.1 software renderer
- An OpenGL 1.4 to Direct3D 9.0 wrapper
- Support for OpenGL ICDs

All will run at the same time and you can even write programs that use all three in different OpenGL contexts at the same time.
Can you use the OpenGL ICD in an AeroGlass window yet?
 
JHoxley said:
That's not even remotely surprising. Until I broke it yesterday, I was dual-booting 5270 and XP - XP runs circles around Vista when it comes to performance! BUT, and this is the important part, one is a stable mature 5 year old OS, one is only just feature-complete :LOL:

I'd imagine that the true performance of Vista won't start to be known for a while yet - 5308 is feature complete, which means they're now going to be working on bug fixes and optimizations rather than new stuff.

Jack
5270 has alot of debugging stuff going on if I'm not mistaken. As well as 5308. And no reason to doubt the April CTP wont be any different. Probably wont see what Vista is capable of until RTM.
 
Back
Top