On November 17, 2026, Natalie Vock submitted a v2 series of 10 AMDGPU kernel patches proposing to batch virtual-memory mapping updates. The change aims to address stalls that can happen under heavy sparse-binding workloads; VideoCardz connected that problem to some games played through Proton. The proposal has no measured game-performance result. Vock’s patch series
What problem does the AMDGPU proposal target?
Vock’s patch description says that heavy sparse-binding work can stall graphics and compute submissions when PRT mappings are synchronized with other submissions. Sparse-binding updates also require separate ioctl calls—requests from user-space software to the kernel—which can mean repeated switches between user and kernel mode. The proposal aims to change how those updates are submitted and synchronized, addressing this specific source of potential stutter rather than stutter in general. The patch thread
How would AMDGPU_GEM_VA2 change memory updates?
The proposed AMDGPU_GEM_VA2 interface would let software submit multiple virtual-memory mapping changes in one ioctl, rather than sending each update separately. In the v2 design, an update context can hold pending mapping changes while they are validated, before applying them together. Batching is intended to reduce the work involved in frequent updates; the patch series does not quantify any resulting change in game performance. The patch description
The synchronization rules also matter. Where the proposed explicit synchronization is supported, earlier submissions are not necessarily waited on as they are with AMDGPU_GEM_VA. For unmap operations, Vock’s description limits explicit synchronization to always-valid or PRT mappings. Those details define the proposal’s scope; they do not describe a general change to every kind of mapping.
A proposed reduction in update overhead is a design goal; smoother gameplay still depends on the changes being used and their effect being measured.
What would have to happen before players benefit?
A kernel patch submission alone does not put the interface into a player’s graphics stack. VideoCardz reported that a user-space driver such as Mesa RADV would also need to adopt the interface before players could benefit. The patch series therefore describes a proposed driver change, not a measured improvement for Proton games.
What did the automated review flag?
A Sashiko automated review in the mailing-list thread flagged 15 potential issues in the proposed ioctl. The thread’s review discussion does not characterize those flags as confirmed defects. The patch thread