Yes, CUDA 7.0 dropped a number of aspects of 32-bit support, including the NPP library in 32-bit mode. Excerpting from the release notes that came with the CUDA 7.0 toolkit:
The following features are officially unsupported in the current release. Developers must
employ alternative solutions to these features in their software.
Support for 32-bit x86 Linux Systems
The CUDA Toolkit and CUDA Driver no longer support developing and running
CUDA and OpenCL Applications on 32-bit x86 Linux operating systems.
Note 1: Developing and running 32-bit applications on 64-bit (x86_64) Linux
operating systems is still supported, but that functionality is marked as deprecated
and may be dropped in a future release. 64-bit applications are not impacted.
Note 2: This notice applies to x86 architectures only; 32-bit application support on the
ARM architecture remains officially supported.
Red Hat Enterprise Linux 5 and CentOS 5
CUDA no longer supports the RHEL 5 and CentOS 5 Linux distributions. Please note
that RHEL 6, RHEL 7, CentOS 6, and CentOS 7 are all supported.
CUDA Toolkit and CUDA Driver Support for Tesla Architecture
The CUDA Toolkit and CUDA Driver no longer supports the sm_10, sm_11, sm_12,
and sm_13 architectures. As a consequence, CU_TARGET_COMPUTE_1x enum values
have been removed from the CUDA headers.
Certain CUDA Features on 32-bit and 32-on-64-bit Windows Systems
The CUDA Toolkit no longer supports 32-bit Windows operating systems.
Furthermore, the Windows CUDA Driver no longer supports Tesla and Quadro
products on 32-bit Windows operating systems. Additionally, on 64-bit Windows
operating systems, the following features are no longer supported by the CUDA
driver or CUDA toolkit:
‣ Running 32-bit applications on Tesla and Quadro products
‣ Using the Thrust library from 32-bit applications
‣ 32-bit versions of the CUDA Toolkit scientific libraries, including cuBLAS,
cuSPARSE, cuFFT, cuRAND, and NPP
‣ 32-bit versions of the CUDA samples
Note the above list doesn’t impact any 64-bit components on Windows.