NVIDIA Kepler speculation thread

[pharma]

But is true for re-branded 7000 hardware, Bonaire & Malta? I think they were just pointing out that the entire 7000 series was not entirely DirectX 11.1 compliant.

 

[Dave Baumann]

And for channel products they are wrong. All the HD 7000 Series are DX11_1 feature level.

 

[DSC]

http://blogs.nvidia.com/blog/2013/0...graphics-with-less-memory-at-microsoft-build/

In a third demo illustrating another powerful use of Tiled Resources, Microsoft showed a program developed by our content technology group. It showed a power plant filled with fine shadows that dynamically shift to reveal the detailed environment.

In this example, an enormous shadow depth map is used to precisely compute the shadows in the scene.

To avoid the allocation of a prohibitive amount of memory, Tiled Resources are used to allocate only the memory required to compute the shadows for the current view, a tiny fraction of the total.

The residency map, on the left, depicts the allocation map; the black region represents high resolution and shades of gray represent progressively lower resolutions of the shadow hierarchy.

Mods, please remove the Radeon stuff, this is the Nvidia Kepler thread, if people didn't notice, it's not for discussing Radeon, go to the proper thread.

 

[AnarchX]

From link above:

There are over 90 million NVIDIA GPUs capable of supporting Tiled Resources.

Are Kepler sales such high or is Fermi supported to?

 

[pharma]

Is "Tiled Resources" a new DirectX 11.2 feature? I did see this will be very useful for terrain in games.

 

[Kaotik]

Is "Tiled Resources" a new DirectX 11.2 feature? I did see this will be very useful for terrain in games.

Yes, but it requires only D3D11_0 feature level (and the other version requires 11_1)

 

[jimbo75]

Eh? We test in a closed case. A Thermaltake Spedo, to be precise.

Your numbers don't add up to an 8% difference unless you're doing something really strange btw. I get roughly half of that.

 

[Psycho]

From link above:
Are Kepler sales such high or is Fermi supported to?

hmm.. sounds a tad high for kepler, but too low for kepler+fermi.
From John peddie it looks like nvidia is shipping around 20M per quarter, but a good deal of these are quite low end, and many of those would be fermi - especially a year ago (gtx660 didn't launch until september) and even now (gt630 and down, gt635m and down) http://en.wikipedia.org/wiki/GeForce_600_Series#GeForce_600_.286xx.29_series

...So with a bit of marketing optimism (and of course it means produced, not shipped or sold) it's likely kepler. Or maybe GF117/9 added the necessary support?

Mods, please remove the Radeon stuff, this is the Nvidia Kepler thread, if people didn't notice, it's not for discussing Radeon, go to the proper thread.

That's why you're posting microsoft dx11.2 demo stuff? Not a lot of kepler speculation there.. An 11.2 thread would hopefully be collected soon though..

 

[lanek]

http://blogs.nvidia.com/blog/2013/0...graphics-with-less-memory-at-microsoft-build/





Mods, please remove the Radeon stuff, this is the Nvidia Kepler thread, if people didn't notice, it's not for discussing Radeon, go to the proper thread.

The radeon stuff are part of a response to an article who treat both about Nvidia and AMD stuff about the same subject. ( who is the capability or not to support Tiled resources or more precisely DX11.1 features ( softwares ( or API ) in this case for one and in hardware for the second. )

So yes we should maybe move all of this on a DX11.xx thread ..

 

[Davros]

ps: incase it wasnt mentioned
Catalyst_12.11_Beta11 supports Tiled resources

 

[sebbbi]

Grr! That looks exactly like my secret tech . It seems that great minds think alike. Virtual shadow mapping will definitely become a very popular technique in the future. I wonder how they are handling the fine grained culling (or are they just brute force rendering it, because the scene is so simple).

Tiled resources has two feature levels - the first requires DX11_0 hardware feature level as the base while the second tier requires DX11_1 feature level as the base.

Could you give any more details about these feature levels, or pointers to more detailed documentation (I didn't find any from MS site)? Also Haswell supports 11_1 feature level. Does this also mean that it supports same tiled resource features as GCN?

 

[HMBR]

looks like any DX11 card will support tiled resources?
http://www.youtube.com/watch?v=EswYdzsHKMc&feature=youtu.be&t=2m45s

 

[Gipsel]

looks like any DX11 card will support tiled resources?
http://www.youtube.com/watch?v=EswYdzsHKMc&feature=youtu.be&t=2m45s

That's really not clear. And even if some tiled resources are supported, there are obviously differences in the extent of the support. But details are scarce right now.

Spoiler

That's all I could find regarding the different tiers of tiled resource support:

typedef struct D3D11_FEATURE_DATA_D3D11_OPTIONS1 {
  D3D11_TILED_RESOURCES_TIER TiledResourcesTier;
  BOOL                       MinMaxFiltering;
  BOOL                       ClearViewAlsoSupportsDepthOnlyFormats;
  BOOL                       MapOnDefaultBuffers;
} D3D11_FEATURE_DATA_D3D11_OPTIONS1;

TiledResourcesTier
Specifies whether the hardware and driver support tiled resources. The runtime sets this member to a value indicating if the hardware and driver support tiled resources and at what tier level.

 

[willardjuice]

Also Haswell supports 11_1 feature level. Does this also mean that it supports same tiled resource features as GCN?

Sounds like no (still no confirmation).

I think only amd 11.1 cards are tier 2 (i.e. tier 2 = amd's prt).

 

[ltcommander.data]

Sounds like no (still no confirmation).

I think only amd 11.1 cards are tier 2 (i.e. tier 2 = amd's prt).

So if not all DX11.1 GPUs are tier 2, then can we still assume all DX11 GPUs are tier 1, particularly older GPUs like Evergreen, Fermi, and Ivy Bridge? Or is that hit or miss as well?

 

[Gipsel]

So if not all DX11.1 GPUs are tier 2, then can we still assume all DX11 GPUs are tier 1, particularly older GPUs like Evergreen, Fermi, and Ivy Bridge? Or is that hit or miss as well?

I would think, Evergreen is tier 0 = no support. But that's just my feeling.

 

[willardjuice]

So if not all DX11.1 GPUs are tier 2, then can we still assume all DX11 GPUs are tier 1, particularly older GPUs like Evergreen, Fermi, and Ivy Bridge? Or is that hit or miss as well?

I believe all feature level 11.0 gpus can at least do tier 1 (with a driver update of course). Seems tier 2 requires more than just feature level 11.1 though (atm I'm guessing it's essentially AMD's prt).

 

[Malo]

I guess the question remains that is tier 1 useful to developers or would they be interested in tier 2 only.

 

[Gipsel]

I guess the question remains that is tier 1 useful to developers or would they be interested in tier 2 only.

It is useful as evident from the demo running on a nV GPU. The question would be what the restrictions are exactly and if devs put in the effort to support also tier 2 to offer higher performance or higher texture resolution or whatever one could use it for.

 

[lanek]

like shown on the roadmap of DirectX, MS offtly ( for dont say allways ) introduce this type of feature in at min 2 phases. Like many specific DX11.1 features, you have first a software ( API + driver ) compatibility introduced, then in general an hardware requirement for support the level features.

This is for keep compatibility with older GPU's for developpement and dont break compatibility in a same "DirectX series " too fast ( between DX11.0, 11.1, 11.2 etc.. )

Like for many features on DX11.1 :

- Tiled resources first introduction " Tier1" = software DX11.1 compatibility feature level ( driver update, require DX11.1 software support on DX 11.0 GPU + DirectX 11.2)
- Tiled Resources second level "Tier 2" = Hardware DX11.1 requirement
But for Tier3, i dont know maybe
- Tiled Resources third level " Tier 3" = final level, the introduction software > hardware is completed.

Many things have happend in between, HUMA from AMD, Virtual adress memory on hardware, Nvidia have too a similar things in their roadmap, so things is maybe more convergent of what we can imagine between MS and AMD - Nvidia on this.

A second possibility is just MS have release DX11.2 + software compatibility for Tiled Resources in advance for 2 reasons:

- XBOX one games will support it, so they need it on PC too . ( for be sure developpers use it )
- A good try for make peoples move on Windows 8 ( 8.1 ), a little bit like DX10 and Vista, on his time if you wanted play the few games available on DX10. ( increase games requiring windows 8 and so imagine gamers will follow )