GPU: Clarify toolkit baseline and device packet traits

Eigen/src/Core/util/Macros.h

diff --git a/Eigen/src/Core/util/Macros.h b/Eigen/src/Core/util/Macros.h
index 2928ab7..98cdde7 100644
--- a/Eigen/src/Core/util/Macros.h
+++ b/Eigen/src/Core/util/Macros.h
@@ -578,6 +578,7 @@
 #define EIGEN_CUDA_SDK_VER 0
 #endif
 
+// CUDA 11.8 is the oldest toolkit in GPU CI (it supports the sm_89 runners), not a new packet-intrinsic requirement.
 #if defined(EIGEN_CUDACC) && EIGEN_CUDA_SDK_VER > 0 && EIGEN_CUDA_SDK_VER < 110800
 #error "Eigen requires CUDA 11.8 or later."
 #endif

test/packetmath_gpu.cu

diff --git a/test/packetmath_gpu.cu b/test/packetmath_gpu.cu
index 2efa60c..73d7231 100644
--- a/test/packetmath_gpu.cu
+++ b/test/packetmath_gpu.cu
@@ -275,7 +275,7 @@
 };
 
 // The device pass's packet traits, written into an int array: the flags the host selects operations by are the
-// device's, not the host pass's, which differ under nvcc (EIGEN_HAS_GPU_DEVICE_FUNCTIONS).
+// device's, not the host pass's, where the half packets and their traits do not exist.
 #define EIGEN_GPU_TEST_TRAIT_FLAGS(X) \
   X(Vectorizable)                     \
   X(size)                             \