V0.13.0 Release Plan

[yukirora]

Release Manager

@cp5555

Endgame

• Code freeze: Oct, 2025
• Bug Bash date: TBD
• Release date: TBD

Main Features

SuperBench Improvement

1. • Add cuda13.0.dockerfile support (Dockerfile - add cuda13.0.dockerfile #739)
2. • Add nsys and pytorch profiler debug trace support (Enhancement: Add nsys and pytorch profiler debug trace support #744)

Micro-benchmark Improvement

1. • Collect per-snapshot per-GPU flops/temp in gpu burn (Benchmarks: Micro benchmark - collect per-snapshot per-GPU flops/temp in gpu burn #735)
2. • Add simultanneously all-to-host / host-to-all bandwidth testcases to nvbandwidth (Benchmarks: Micro benchmark - Add simultanneously all-to-host / host-to-all bandwidth testcases to nvbandwidth #736)
3. • Add ncu profile support in cublaslt-gemm (Benchmarks: Micro benchmark - add ncu profile support in cublaslt-gemm #740)
4. • Support verification and parallel run for disk performance benchmark (Benchmarks: Micro benchmark - Support verification and parallel run for disk performance benchmark #741)
5. • Add numa support for nvbandwidth (Benchmarks: Micro benchmark - Add numa support for nvbandwidth #742)
6. • Change cublasLtMatmulDescCreate scaleType from CUDA_R_32F to CUDA_R_16F in FP16 dist inference (Benchmarks: micro benchmarks - change cublasLtMatmulDescCreate scaleType from CUDA_R_32F to CUDA_R_16F in FP16 dist inference  #732)
7. • Support gemm correctness check in cublaslt-gemm
8. • Multi node nccl validation enhancement
9. • mscclpp support
10. • Add new busbw metrics for NCCL/MSCCL testing with specific algorithm
11. • Fix NVBandwidth benchmark results parsing bug
12. • Support FP4 kernels for cutlass benchmark

Model Benchmark Improvement

1. • Add option to exclude data copy time in model benchmarks (Benchmark: Model benchmark - add option to exclude data copy time in model benchmarks #734)
2. • Support state-of-art LLM model training perf including Deepseek, qwen
3. • Support state-of-art LLM model inference perf including Deepseek, qwen
4. • Support state-of-art LLM module and model correctness benchmark
5. • Deterministic training support (Benchmark: Model benchmark - deterministic training support #731)

Bug fix

1. • dist-inference raise cublaslt error
2. • Add --set_ib_devices option to auto-select IB device by MPI local rank in ib validation (Benchmarks: micro benchmarks - add --set_ib_devices option to auto-select IB device by MPI local rank in ib validation #733)
3. • NVBandwidth benchmark results parsing bug (NVBandwidth benchmark results parsing bug #748)
4. • CI/CD - Fix image merge in GitHub Action (CI/CD - Fix image merge in GitHub Action. #749)
5. • Fix pipelines - Update mlc version in dockerfiles from v3.11 to v3.12 (Fix pipelines - Update mlc version in dockerfiles from v3.11 to v3.12 #752)
6. • CI/CD - Fix python3.10 pipeline (CI/CD - Fix python3.10 pipeline #753)
7. • CI/CD - Fix Azure test pipeline (CI/CD - Fix Azure test pipeline #754)

Tools

1. • System info enhancement