AnandTech on WGF

[Geo]

This is a few days old now, but can one of our resident experts take a look and see if this article is accurate? There are some interesting points in there that I didn't know, but I'm not quite confident enough (no insult intended) in AT to get it right at this level of detail with graphics.

http://www.anandtech.com/tradeshows/showdoc.aspx?i=2403&p=4

For instance, is this bit accurate re the extensions for WGF1.0 beyond 9.0c?:

WGF 1.0 is also known as DirectX 9.L and will add a couple new features beyond DirectX 9.0c. These new features include: cross-process shared surfaces, managed graphics memory (and virtualized graphics memory), prioritization of reasources, text antialiasing, advanced gamma functions, and device removed (in order to gracefully recover from a failure the hardware can be "removed" and then "added").

 

[Blazkowicz]

All these things sounds to me like they are about drivers features, with no hardware feature really needed?

mainly software stuff meant for the window manager to work well with multiple applications and windows.


the last bit reminds of ATI "VPU recover" though. and I wonder about virtual memory (is something doable with AGP texturing?)

 

[Geo]

All these things sounds to me like they are about drivers features, with no hardware feature really needed?

mainly software stuff meant for the window manager to work well with multiple applications and windows.


the last bit reminds of ATI "VPU recover" though. and I wonder about virtual memory (is something doable with AGP texturing?)

Yes, it does become interesting if virtual memory is a WGF1.0 feature.

Does this mean current 9.0 cards will be shut out, and the difference between 9.0c and 9.0l support is a bigger step than we (well, at least I) have been thinking?

Or does it mean that they'll have virtual memory support, but at the cost of a lot of performance as the driver struggles to provide support that the hardware wasn't designed around?

And if the latter, then my thinking on the logic of this next gen (R520) starting the mosey towards WGF in hardware design makes more sense again.

 

[Khronus]

As I understand it, the biggest difference between WGF 1.0 (DX9L) and WGF 2.0 will be hardware support for scheduling/context switching and greater flexibility to work on/with data already in the graphics pipeline. Hopefully I'm not too off base, if I am I'm sure some of this boards smarter people will correct me

 

[Fodder]

ExtremeTech are claiming WGF1.0 is equivalent to DX9.0c, which sounds a little off the mark, as it would cut out 90% of the DX9 chipsets out there for seemingly little gain.

 

[K.I.L.E.R]

Managed graphics memory means that DX will have control over VRAM, programs can only request.

Virtual graphics memory is virtual memory for graphics.
Would work better under PCI-X because AGP isn't up for the task.

 

[Nappe1]

Managed graphics memory means that DX will have control over VRAM, programs can only request.

Virtual graphics memory is virtual memory for graphics.
Would work better under PCI-X because AGP isn't up for the task.

can I ask why AGP isn't up to task?
whenever we have global poll of memory, it needs to be controlled somehow and afaik, even with PCI-X there isn't memory bus access on the card. (so that main memory controller on motherboard could also be responsible for controlling graphics memory.

..SO, the graphics chip has to be the memory controller. (and afaik, 3DLabs P10, which was released just before Matrox Parhelia about 2 (or 3?) years ago, already had full virtual graphics memory support.)

I don't see why AGP would not be up to task here, though it might work a bit slower, while swpping data. If parts of the graphics memory data is held in HDD, the most limiting factor will be HDD transfer speeds in any case. (SATA 2.0 gives theoretical 300MB/s transfer speed, which ain't much when we are talking speeds needed to fill 64-128 MB of graphics memory.)

This actually reminds me another topic that I was about to ask here... so thanks K.I.L.E.R. otherwise I would have forgot it.

 

[Rys]

With AGP you only have access to the memory space as a single chunk, rather than with the fine granularity you'd need for a completely virtualised memory setup.

PCI Express lets you write into off-card memory in (almost) arbitrary sizes, anywhere in the memory map.

 

[Xmas]

ExtremeTech are claiming WGF1.0 is equivalent to DX9.0c, which sounds a little off the mark, as it would cut out 90% of the DX9 chipsets out there for seemingly little gain.

DX9.0c doesn't work with 90% of the current DX9 chipsets? I didn't know that...

 

[Charmaka]

Compatibility vs compliance? I mean, a GeForce 2 will work under DX9, but it doesn't support all the features therein. If what's actually being suggested is that WGF1.0 will require hardware to implement everything in 9.0c, ie SM3.0 or whatever, the only stuff that'll work is 6x00 hardware, no?

 

[Demirug]

WGF 1.0 adds only some new features to DX9.

But you will need a card with some features if you want to use Aero or Aero Glass.

 

[Dave B(TotalVR)]

can I ask why AGP isn't up to task?

Becuase it suxx0rs, its single direcitonal so to change from read to write takes a large turnaround time.

 

[Sandwich]

What's the advantage of Virtual graphics memory?
So windows can swap textures back and forth as aggressively as it does now with the swap file?