Direct3D feature levels discussion

I have updated the Wikipedia article with my own version of a basic Direct3D 12 feature matrix table (and some revisions to the main feature level table):
https://en.wikipedia.org/wiki/Direct3D#Direct3D_12_levels

I see you've added Asynchronous Shaders to the feature level table now but I'm not sure if it's correct? According to this post from @Andrew Lauritzen :

Direct3D feature levels discussion
All DX12 (API) compatible GPU's support async compute.

However if the table is implying the capability to use both graphics and compute queues at the same time, then according to @Ryan Smith in this Anandtech article:
[/URL]
http://www.anandtech.com/show/9124/amd-dives-deep-on-asynchronous-shading

Maxwell 2 is capable of that.
 
Last edited:
What's up with GCN's not supporting Asynchronous Shaders according to Wiki-table?
It's one of the big things AMD has been touting, they definitely support it

edit:
asyncshaders.jpg
 
Asynchronous shaders is not a cap-bit features.

D3D12 API allows asynchronous copy and computes command along with "normal" graphics compute. Depending on hardware, async-copy and async-compute could be serialized by driver or run in parallel with the graphics engine. All this is related to multi-engine https://msdn.microsoft.com/en-us/library/dn899217.aspx

As far I knwo, all D3D12 GPUs should take advantage of async-copy (AMD calls it "async-DMA"), while only GCN (all generations) and Maxwell 2.0 I am sure they have built-in hardware support for async-compute (ie: there is no driver serialization). I am not sure about Skylake iGPu.
 
By the way, the number of compute units is not directly comparable between different architectures, eg: Maxwell 2.0 has 31 compute unit while GCN 1.1 and 1.2 have 8 compute unit, but it doesn't mean that Maxwell 2.0 implementation is ~4 time better, they are different architectures, it is not like CPU where you can try to compare an AMD quad-core with an intel quad-core.
 
Sigh. As I've explained earlier in the thread, the notion isn't even well-defined in terms of hardware. *Which* shaders with *which* state can run simultaneously in *which* situations exactly? Depends entirely on the architecture and even the SKU. You can say that on some hardware compute and 3D shaders will *never* run simultaneously, but you cannot ever guarantee simultaneous execution on any architecture. Which, incidentally is why the API does not differentiate - it must work on all implementations and the rest is highly hardware dependent and not really a "feature" any more than ability to pipeline different 3D state changes is a "feature".
 
Last edited:
That page should be locked. To much bad PRs (not everyone!) and to much bad "tech" sites added to much imaginary things to the API and their favourites hardware... I am just curious to see what will the first GPU able to travel the galaxy via warp tunnelling XD
 
Taking this matter to the Wikipedia administrative action.
[EDIT] User blocked for edit warring.
 
Last edited:
I wonder where ExtremeTech got that outdated/erroneous table which they claim is a "Microsoft slide"? Specifically, according to our latest optional feature reports,
  • Maxwell-1 does not seem to support Tiled Resources tier 2 and UAV loads for additional formats to qualify for level 12_0;
  • Kepler does not support UAVs in every stage to qualify for level 11_1, and does not support Tiled Resources tier 2 either;
  • Maxwell-2 only supports Conservative Rasterization tier 1; and
  • UAV Tier 2 is not a part of public Windows SDK.
Moreover it doesn't resemble any recent Direct3D 12 presentation by Max McMullen, it rather looks like some piece of internal documentation or another fan-made feature matrix table.

Ryan, we need your test setup once again please!
 
Last edited:
Back
Top