Microsoft Launches Shader Model 6.9, DXR 1.2, and DX12 Upgrades in Agility SDK 1.619, with Extensive NVIDIA Support

Microsoft has officially launched the Agility SDK 1.619, introducing significant advancements in DirectX 12, including Shader Model 6.9 and DXR 1.2 enhancements.

Major Updates in Microsoft Agility SDK: Shader Model 6.9, DXR 1.2, & Enhanced DX12 Features for NVIDIA RTX, AMD Radeon & Intel Arc GPUs

The newly released Agility SDK 1.619 brings three pivotal updates: Shader Model 6.9, DXR 1.2, and various DX12 improvements. Below are the key highlights:

Shader Model 6.9 (via DXC 1.9.2602.16):
- Support for Long Vector operations.
- Introduction of 16-bit float Special instructions.
- Mandatory inclusion of 16-bit and 64-bit shader operations and wave operations.
- HLSL exposure for features in DXR 1.2.
DXR 1.2:
- Opacity Micromaps (previously released, with new SM 6.9 features now activated).
- Shader Execution Reordering (SER).
D3D Customer-Requested Features:
- Updated Resource View Creation APIs.
- Periodic Trim Notifications.
- Expanded Dispatch Grid Limits.
- CPU Timeline Query Resolves.

A standout element of this release is Shader Model 6.9, which introduces essential features including Long Vectors, allowing HLSL to manage vectors of lengths between 4 and 1024 elements. Furthermore, both 16-bit and 64-bit shader operations are now standard requirements.

A presentation slide titled 'DXR 1.2: Shader Execution Reordering (SER)' shows up to '2x performance gain' in path-traced scenarios.

The updates for DXR (DirectX Ray Tracing) include significant features like Opacity Micromaps and Shader Execution Reordering. The inclusion of OMM enhances how hardware manages alpha-tested geometry, potentially offering remarkable performance gains compared to traditional methods involving AnyHit shader invocations.

Performance improvement chart demonstrating the effect of Opacity Micromaps on rendering performance.

Microsoft has previously reported up to 2.3x performance boosts in path-traced titles utilizing OMM, with specific demos from NVIDIA displaying over 60% improvements. For instance, a demonstration showed a scene running at 55 FPS without OMM, while enabling this feature increased performance to 90 FPS. Currently, comprehensive support for OMM is available exclusively for NVIDIA RTX GPUs.

Performance rendering example utilizing Shader Execution Reordering and Opacity Micromaps.

One significant title that utilizes these features is Alan Wake, renowned for its heavy reliance on alpha-tested geometry. The game can handle a staggering number of triangles and vertices, with its high-ray tracing settings capable of producing up to 36.9 million rays per frame. Notably, using SER and OMM, the RTX 4090 can render a scene in approximately 10.2ms, down from 16.8ms.

Opacity Micromaps

Opacity Micromaps (OMMs) enhance the efficiency of alpha-tested geometry processing by reducing reliance on costly AnyHit shader invocations. This feature was introduced earlier, with a recent update expanding its HLSL exposure.

Shader Execution Reordering

Shader Execution Reordering (SER) enhances shader execution by allowing applications to organize shader code for improved parallel processing. This feature has now exited preview, and applications can query device support for reordering.

via Microsoft

Furthermore, the Agility SDK 1.619 delivers vital enhancements to DirectX 12 Ultimate API, addressing prior limitations in buffer views. This release allows developers to specify buffer views using byte offsets and sizes, vastly improving API usability.

As GPU architectures advance, issues have arisen with D3D12’s view-creation model, particularly regarding buffer access patterns and descriptor management. The new updates modernize this aspect of the API to ensure it meets contemporary needs.

The updated model allows for the measurement of buffer views in byte offsets and sizes, coupled with improved error handling through HRESULT rather than exclusively relying on debug layer validation.

Periodic Trim Notifications

New kernel-level trim notifications are integrated into the D3D12 runtime, facilitating applications to receive alerts on when to trim residency without requiring fresh driver support.

Increased 1D Dispatch Limit

This update raises the maximum 1-Dimensional Dispatch/Dispatch Mesh size beyond the previous limit of 65535, accommodating the capabilities of contemporary hardware.

D3D12_FEATURE_DATA_D3D12_OPTIONS22. Max1DDispatchSize

D3D12_FEATURE_DATA_D3D12_OPTIONS22. Max1DDispatchMeshSize

CPU Timeline Query Resolves

A newly introduced Query Heap allows resolution on the CPU timeline, minimizing unnecessary overhead during GPU operations. This feature introduces ID3D12Device15::CreateQueryHeap1 and ID3D12Device15::ResolveQueryData.

via Microsoft

GPU Vendor Support for Microsoft Agility SDK 1.619

Support for the Agility SDK 1.619 is now available across all three major GPU vendors, albeit with varying levels of compatibility based on product architecture.

Hardware Support Overview

Independently Held Vendor (IHV)	Driver Links
AMD	AMD Software: Adrenalin Edition 26.2.1 AMD Software: AgilitySDK Developer Preview Edition 25.30.21.01
Intel	Intel Arc Graphics – Windows
NVIDIA	Official NVIDIA Drivers \| NVIDIA or via the NVIDIA App for automatic updates.

NVIDIA reports that its entire line of RTX GPUs is compatible with the SDK’s latest enhancements. For DXR 1.2, the hardware acceleration of OMM applies to RTX 40 series and newer, whereas SER is supported on the same generation. Older GPUs only provide software emulation for OMM features.

On the AMD front, support encompasses RX 7000, RX 900, and Ryzen AI 300/400 series, pointing to compatibility with RDNA architectures. However, full feature support for DXR 1.2 and Shader Model 6.9 is specific to RDNA 4 (Radeon RX 9000) GPUs.

Intel claims that its Arc B-Series (Battlemage) will support all SDK features. Moreover, certain D3D functionalities, like VPblit 3DLUT, apply to Lunar Lake CPUs.

Notably, while OMMs are exclusive to RTX 40 series and higher, SER exhibits reordering capabilities on NVIDIA RTX 40+ and Intel Arc B-Series, with no reordering functionality available on RX 9000 series GPUs.

AMD	Intel	NVIDIA
Long Vector	AMD Radeon RX 9000 series	Intel Arc B-Series Graphics	All RTX hardware
16-bit Float Specials	AMD Radeon RX 9000 series	Intel Arc B-Series Graphics	All RTX hardware
Opacity Micromaps (OMM)	N/A	N/A	All RTX hardware. Hardware-accelerated on RTX 4xxx+, emulated on older models.
Shader Execution Reordering (SER)	AMD Radeon RX 9000 series supports API but doesn’t implement reordering.	RTX 4xxx+ GPUs support the API and perform reordering.	Intel Arc B-Series Graphics. Existing 64k limit, set to increase with future drivers.
Revised Resource View Creation APIs	AMD Radeon RX 7000 and 9000 series	Intel Arc B-Series Graphics	All RTX hardware
Periodic Trim Notifications	N/A	Intel Arc B-Series Graphics	All RTX hardware
Increased Dispatch Grid Limit	AMD Radeon RX 7000 and 9000 series. UINT_MAX compute, 64k mesh.	Intel Arc B-Series Graphics. Existing 64k limit, planned increases in future drivers.	All RTX hardware. Existing 64k limit, planned increases in future drivers.
CPU Timeline Query Resolves	AMD Radeon RX 7000 and 9000 series	Intel Arc B-Series Graphics	All RTX hardware
Fence Barriers (preview)	AMD Radeon RX 7000 and 9000 series	Intel Arc B-Series Graphics	Contact developer relations for access to in-development drivers.
VPblit 3DLUT (preview)	AMD Radeon RX 7000 series graphics cards and Ryzen AI 300/400 series processors with integrated graphics	Intel Core Ultra processors, Lunar Lake and Panther Lake platforms	Contact developer relations for detailed access.