Issues

• [FEA] Compute float32 POW, LOG_BASE and ATAN2 in float32 when the output is FLOAT32 (4-9x faster on low-FP64 GPUs)

  NVIDIA/cudf#25299 · Filter by author maxwbuckley opened on Nov 15, 2026

• [FEA] Overlap staging windows in compressed experimental pack_into

  ? - Needs Triagefeature requestlibcudfPerformance
  NVIDIA/cudf#25267 · Filter by author vuule opened on Nov 13, 2026

• [FEA] Rank-local CPU fallback for IR nodes in cudf-polars

  NVIDIA/cudf#25256 · Filter by author madsbk opened on Nov 13, 2026
  

• [BUG] strings::find with a targets column: all-null input loses nulls, all-empty targets return -1

  NVIDIA/cudf#25231 · Filter by author dan13bauer opened on Nov 12, 2026
  1

• [FEA] Java: Support IS_NULL on LIST columns in AST

  NVIDIA/cudf#25230 · Filter by author findepi opened on Nov 12, 2026

• [FEA] API for unique key JOIN followed by GROUP BY

  feature requestlibcudfPerformance
  NVIDIA/cudf#25228 · Filter by author joosthooz opened on Nov 12, 2026
  5

  

• [FEA] Fallible variants of integer narrowing casts

  NVIDIA/cudf#25223 · Filter by author findepi opened on Nov 12, 2026

• [BUG] INT96 Parquet output through Spark emits 8-byte timestamp statistics and fails DuckDB readback

  ? - Needs Triagebug
  NVIDIA/cudf#25222 · Filter by author wjxiz1992 opened on Nov 12, 2026

• [BUG] isin matches integers against datetime and timedelta values by their raw nanoseconds

  NVIDIA/cudf#25221 · Filter by author Satya-0023 opened on Nov 12, 2026

• Benchmark hash-based equality-group lookup for as-of joins

  NVIDIA/cudf#25193 · Filter by author shrshi opened on Nov 10, 2026

• delete deprecated fixed_point_scalar_device_view::set_value method

  NVIDIA/cudf#25182 · Filter by author jayshrivastava opened on Nov 10, 2026

• [BUG]: pylibcudf bit_cast copies while libcudf returns a view

  bugcudf-polarsPerformancepylibcudf
  NVIDIA/cudf#25170 · Filter by author TomAugspurger opened on Nov 10, 2026
  4

• [FEA] Expose filter_join_indices_jit (AST overload) in Java

  NVIDIA/cudf#25168 · Filter by author findepi opened on Nov 10, 2026

• [FEA]: Also run the Narwhals test suite against Narwhals main in the nightly GPU tests

  ? - Needs Triagefeature request
  NVIDIA/cudf#25164 · Filter by author FBruzzesi opened on Nov 9, 2026
  

• [FEA] Share the cuCascade cudf::io::datasource with other libcudf-based engines

  NVIDIA/cudf#25160 · Filter by author Matt711 opened on Nov 9, 2026
  

• [BUG] cudf-polars: decimal division by zero returns null on the GPU but raises on the CPU

  bugcudf-polarsPython
  NVIDIA/cudf#25159 · Filter by author Matt711 opened on Nov 9, 2026
  

• [FEA] Java: slice a HostColumnVector by row range on the host

  ? - Needs Triagefeature requestJavaSpark
  NVIDIA/cudf#25156 · Filter by author thirtiseven opened on Nov 9, 2026

• [FEA] Remove numba-cuda as a test dependency

  feature requestPython
  NVIDIA/cudf#25144 · Filter by author mroeschke opened on Nov 6, 2026

• [FEA]: cudf-polars: remote I/O benchmark validation across scale factors and S3 / NIC limit tooling

  cudf-polarsfeature request
  NVIDIA/cudf#25137 · Filter by author Matt711 opened on Nov 6, 2026
  

• [FEA]: cudf-polars: remote Parquet I/O on multiple GPUs with the cuCascade cache

  cudf-polarsfeature request
  NVIDIA/cudf#25136 · Filter by author Matt711 opened on Nov 6, 2026
  

• [PERF]: cudf-polars: one host-memory budget for the cuCascade cache, rapidsmpf spill and staging buffers

  cudf-polarsPerformance
  NVIDIA/cudf#25135 · Filter by author Matt711 opened on Nov 6, 2026
  1

  

• [PERF]: cudf-polars: reduce spill of scan output (Q18, Q1 warm, Q21, Q9)

  cudf-polarsPerformance
  NVIDIA/cudf#25134 · Filter by author Matt711 opened on Nov 6, 2026
  

• [PERF]: cudf-polars: decode columns as their bytes arrive inside one read_parquet

  cudf-polarscuIOPerformance
  NVIDIA/cudf#25133 · Filter by author Matt711 opened on Nov 6, 2026
  

• [PERF]: cudf-polars: separate fetching a scan task's bytes from decoding them

  cudf-polarscuIOPerformance
  NVIDIA/cudf#25132 · Filter by author Matt711 opened on Nov 6, 2026
  

• [PERF]: cudf-polars: run Parquet scan reads on their own thread pool

  cudf-polarsPerformance
  NVIDIA/cudf#25131 · Filter by author Matt711 opened on Nov 6, 2026