diff --git a/crates/specialized/rtx-cfd/src/solvers/incompressible/mod.rs b/crates/specialized/rtx-cfd/src/solvers/incompressible/mod.rs
index 0d2a385..7286e06 100644
--- a/crates/specialized/rtx-cfd/src/solvers/incompressible/mod.rs
+++ b/crates/specialized/rtx-cfd/src/solvers/incompressible/mod.rs
@@ -1,3 +1,6 @@
+// PERF-2 (2026-09-16): the legacy GPU modules have not compiled since the
+// initial commit (cudarc API drift); kept out of the `cuda` build until
+// someone needs them — the batched V-cycle benchmark uses cudarc directly.
 //! Incompressible flow solvers
 //!
 //! This module implements pressure-velocity coupling algorithms for incompressible flows:
@@ -26,16 +29,16 @@ pub mod overset;
 /// PISO algorithm implementation
 pub mod piso;
 /// GPU-accelerated PISO algorithm implementation
-#[cfg(feature = "cuda")]
-pub mod piso_gpu;
+// #[cfg(feature = "cuda")]
+// pub mod piso_gpu;
 /// Five-point Poisson problems and the multigrid-preconditioned CG solver
 pub mod poisson;
 pub mod polygon_sdf;
 /// SIMPLE algorithm implementation
 pub mod simple;
 /// GPU-accelerated SIMPLE algorithm implementation
-#[cfg(feature = "cuda")]
-pub mod simple_gpu;
+// #[cfg(feature = "cuda")]
+// pub mod simple_gpu;
 /// CSR matrix + Jacobi-BiCGSTAB for the curvilinear pressure equation
 pub mod sparse_bicgstab;
 
@@ -62,8 +65,8 @@ pub use overset::{
     OversetResult, OversetSolverState, ResidualBucket, StepTimers,
 };
 pub use piso::{PisoParameters, PisoResult, PisoSolver};
-#[cfg(feature = "cuda")]
-pub use piso_gpu::PisoGpuSolver;
+// #[cfg(feature = "cuda")]
+// pub use piso_gpu::PisoGpuSolver;
 pub use poisson::{
     LevelExport, MgPrecision, MgSmoother, MultigridParameters, PcgCache, PoissonProblem,
     PoissonSolution, export_hierarchy, vcycle_f32_reference,
@@ -71,8 +74,8 @@ pub use poisson::{
 };
 pub use polygon_sdf::PolygonSdf;
 pub use simple::{ConvectionScheme, SimpleParameters, SimpleResult, SimpleSolver};
-#[cfg(feature = "cuda")]
-pub use simple_gpu::SimpleGpuSolver;
+// #[cfg(feature = "cuda")]
+// pub use simple_gpu::SimpleGpuSolver;
 
 /// Common solver parameters
 #[derive(Debug, Clone)]
diff --git a/crates/specialized/rtx-cfd/src/solvers/lbm/mod.rs b/crates/specialized/rtx-cfd/src/solvers/lbm/mod.rs
index 4a74f0b..1154314 100644
--- a/crates/specialized/rtx-cfd/src/solvers/lbm/mod.rs
+++ b/crates/specialized/rtx-cfd/src/solvers/lbm/mod.rs
@@ -1,3 +1,6 @@
+// PERF-2 (2026-09-16): the legacy GPU modules have not compiled since the
+// initial commit (cudarc API drift); kept out of the `cuda` build until
+// someone needs them — the batched V-cycle benchmark uses cudarc directly.
 //! Lattice Boltzmann Method (LBM) solvers
 //!
 //! This module implements various LBM models for fluid flow simulation:
@@ -9,23 +12,23 @@
 pub mod boundary;
 pub mod d2q9;
 /// GPU-accelerated D2Q9 LBM solver
-#[cfg(feature = "cuda")]
-pub mod d2q9_gpu;
+// #[cfg(feature = "cuda")]
+// pub mod d2q9_gpu;
 pub mod d3q19;
 /// GPU-accelerated D3Q19 LBM solver
-#[cfg(feature = "cuda")]
-pub mod d3q19_gpu;
+// #[cfg(feature = "cuda")]
+// pub mod d3q19_gpu;
 
 pub use boundary::{
     BounceBackBc, BoundaryConditionSet, BoundaryOrientation, LbmBoundaryCondition, ZouHeBc,
     ZouHeBoundaryType,
 };
 pub use d2q9::{D2Q9Parameters, D2Q9Solver};
-#[cfg(feature = "cuda")]
-pub use d2q9_gpu::D2Q9GpuSolver;
+// #[cfg(feature = "cuda")]
+// pub use d2q9_gpu::D2Q9GpuSolver;
 pub use d3q19::{D3Q19Parameters, D3Q19Solver, MacroscopicVariables3D};
-#[cfg(feature = "cuda")]
-pub use d3q19_gpu::D3Q19GpuSolver;
+// #[cfg(feature = "cuda")]
+// pub use d3q19_gpu::D3Q19GpuSolver;
 
 /// Common LBM traits and utilities
 pub mod common {
diff --git a/crates/specialized/rtx-cfd/src/turbulence/mod.rs b/crates/specialized/rtx-cfd/src/turbulence/mod.rs
index abdd8fb..a03670a 100644
--- a/crates/specialized/rtx-cfd/src/turbulence/mod.rs
+++ b/crates/specialized/rtx-cfd/src/turbulence/mod.rs
@@ -1,3 +1,6 @@
+// PERF-2 (2026-09-16): the legacy GPU modules have not compiled since the
+// initial commit (cudarc API drift); kept out of the `cuda` build until
+// someone needs them — the batched V-cycle benchmark uses cudarc directly.
 //! Turbulence modeling for CFD
 //!
 //! This module provides various turbulence models for simulating turbulent flows:
@@ -15,8 +18,8 @@ pub mod k_epsilon;
 // of the `cuda` build until someone needs it.
 // pub mod k_epsilon_gpu;
 pub mod smagorinsky;
-#[cfg(feature = "cuda")]
-pub mod smagorinsky_gpu;
+// #[cfg(feature = "cuda")]
+// pub mod smagorinsky_gpu;
 pub mod transition;
 pub mod wall_functions;