FP8 allgather throws an error "reduction support begins with sm90 capable devices." #162057
Description
🐛 Describe the bug
I am using FSDP with pytorch. Previously on 2.7 could use FP8 weights on ampere. They would be cast to correct size and inference would work OK. Since upgrading to 2.8 I run the same job again and am met with this error. If I use wan 2.2 I2V model to generate an image with the exact same code I was using before, I get an image of NaN when the weights are upcast. Proper I2V still returned something, but what's going on here?
Is this fault of pytorch or NCCL libs version? Can NCCL be downgraded, if so? What an unwelcome change.
torch.distributed.DistBackendError: NCCL error in: /pytorch/torch/csrc/distributed/c10d/ProcessGroupNCCL.cpp:3699, invalid argument (run with NCCL_DEBUG=WARN for details), NCCL version 2.27.3
ncclInvalidArgument: Invalid value for an argument.
Last error:
FP8 reduction support begins with sm90 capable devices.
Versions
PyTorch version: 2.8.0+cu126
Is debug build: False
CUDA used to build PyTorch: 12.6
ROCM used to build PyTorch: N/A
OS: Linux Mint 21.3 (x86_64)
GCC version: (conda-forge gcc 13.2.0-13) 13.2.0
Clang version: Could not collect
CMake version: version 3.31.2
Libc version: glibc-2.35
Python version: 3.11.9 (main, Apr 19 2024, 16:48:06) [GCC 11.2.0] (64-bit runtime)
Python platform: Linux-6.14.10-x64v3-xanmod1-x86_64-with-glibc2.35
Is CUDA available: True
CUDA runtime version: 12.6.20
CUDA_MODULE_LOADING set to: LAZY
GPU models and configuration:
GPU 0: NVIDIA GeForce RTX 3090
GPU 1: NVIDIA GeForce RTX 3090
GPU 2: NVIDIA GeForce RTX 3090
GPU 3: NVIDIA GeForce RTX 3090
GPU 4: NVIDIA GeForce RTX 2080 Ti
Nvidia driver version: 570.133.07
cuDNN version: Probably one of the following:
/usr/lib/x86_64-linux-gnu/libcudnn.so.8.9.3
/usr/lib/x86_64-linux-gnu/libcudnn_adv_infer.so.8.9.3
/usr/lib/x86_64-linux-gnu/libcudnn_adv_train.so.8.9.3
/usr/lib/x86_64-linux-gnu/libcudnn_cnn_infer.so.8.9.3
/usr/lib/x86_64-linux-gnu/libcudnn_cnn_train.so.8.9.3
/usr/lib/x86_64-linux-gnu/libcudnn_ops_infer.so.8.9.3
/usr/lib/x86_64-linux-gnu/libcudnn_ops_train.so.8.9.3
Is XPU available: False
HIP runtime version: N/A
MIOpen runtime version: N/A
Is XNNPACK available: True
CPU:
Architecture: x86_64
CPU op-mode(s): 32-bit, 64-bit
Address sizes: 46 bits physical, 48 bits virtual
Byte Order: Little Endian
CPU(s): 96
On-line CPU(s) list: 0-95
Vendor ID: GenuineIntel
Model name: Genuine Intel(R) CPU 0000%@
CPU family: 6
Model: 85
Thread(s) per core: 2
Core(s) per socket: 24
Socket(s): 2
Stepping: 5
CPU max MHz: 3700.0000
CPU min MHz: 1000.0000
BogoMIPS: 4400.00
Flags: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe syscall nx pdpe1gb rdtscp lm constant_tsc art arch_perfmon pebs bts rep_good nopl xtopology nonstop_tsc cpuid aperfmperf pni pclmulqdq dtes64 monitor ds_cpl vmx smx est tm2 ssse3 sdbg fma cx16 xtpr pdcm pcid dca sse4_1 sse4_2 x2apic movbe popcnt tsc_deadline_timer aes xsave avx f16c rdrand lahf_lm abm 3dnowprefetch cpuid_fault epb cat_l3 cdp_l3 intel_ppin ssbd mba ibrs ibpb stibp tpr_shadow flexpriority ept vpid ept_ad fsgsbase tsc_adjust bmi1 hle avx2 smep bmi2 erms invpcid rtm cqm mpx rdt_a avx512f avx512dq rdseed adx smap clflushopt clwb intel_pt avx512cd avx512bw avx512vl xsaveopt xsavec xgetbv1 xsaves cqm_llc cqm_occup_llc cqm_mbm_total cqm_mbm_local dtherm ida arat pln pts hwp hwp_act_window hwp_epp hwp_pkg_req vnmi pku ospke md_clear flush_l1d arch_capabilities
Virtualization: VT-x
L1d cache: 1.5 MiB (48 instances)
L1i cache: 1.5 MiB (48 instances)
L2 cache: 48 MiB (48 instances)
L3 cache: 66 MiB (2 instances)
NUMA node(s): 2
NUMA node0 CPU(s): 0-23,48-71
NUMA node1 CPU(s): 24-47,72-95
Vulnerability Gather data sampling: Vulnerable: No microcode
Vulnerability Ghostwrite: Not affected
Vulnerability Indirect target selection: Not affected
Vulnerability Itlb multihit: KVM: Vulnerable
Vulnerability L1tf: Not affected
Vulnerability Mds: Vulnerable; SMT vulnerable
Vulnerability Meltdown: Not affected
Vulnerability Mmio stale data: Vulnerable
Vulnerability Reg file data sampling: Not affected
Vulnerability Retbleed: Vulnerable
Vulnerability Spec rstack overflow: Not affected
Vulnerability Spec store bypass: Vulnerable
Vulnerability Spectre v1: Vulnerable: __user pointer sanitization and usercopy barriers only; no swapgs barriers
Vulnerability Spectre v2: Vulnerable; IBPB: disabled; STIBP: disabled; PBRSB-eIBRS: Not affected; BHI: Not affected
Vulnerability Srbds: Not affected
Vulnerability Tsx async abort: Vulnerable
Versions of relevant libraries:
[pip3] mkl-fft==1.3.8
[pip3] mkl-random==1.2.4
[pip3] mkl-service==2.4.0
[pip3] nexfort==0.1.dev309+torch241cu121
[pip3] numpy==1.26.4
[pip3] numpy==2.3.1
[pip3] nunchaku==0.2.0+torch2.6
[pip3] nunchaku==0.2.0+torch2.6
[pip3] nvidia-cublas-cu12==12.6.4.1
[pip3] nvidia-cuda-cupti-cu12==12.6.80
[pip3] nvidia-cuda-nvrtc-cu12==12.6.77
[pip3] nvidia-cuda-runtime-cu12==12.6.77
[pip3] nvidia-cudnn-cu12==9.10.2.21
[pip3] nvidia-cufft-cu12==11.3.0.4
[pip3] nvidia-curand-cu12==10.3.7.77
[pip3] nvidia-cusolver-cu12==11.7.1.2
[pip3] nvidia-cusparse-cu12==12.5.4.2
[pip3] nvidia-cusparselt-cu12==0.7.1
[pip3] nvidia-nccl-cu12==2.27.3
[pip3] nvidia-nvjitlink-cu12==12.6.85
[pip3] nvidia-nvtx-cu12==12.6.77
[pip3] onnx==1.17.0
[pip3] pytorch-lightning==2.3.3
[pip3] torch==2.8.0+cu126
[pip3] torchaudio==2.8.0+cu126
[pip3] torchmetrics==1.4.0.post0
[pip3] torchsde==0.2.6
[pip3] torchvision==0.23.0+cu126
[pip3] triton==3.4.0
[conda] blas 1.0 mkl
[conda] cuda-cudart 12.6.37 0 nvidia/label/cuda-12.6.0
[conda] cuda-cudart-dev 12.6.37 0 nvidia/label/cuda-12.6.0
[conda] cuda-cudart-dev_linux-64 12.6.37 0 nvidia/label/cuda-12.6.0
[conda] cuda-cudart-static 12.6.37 0 nvidia/label/cuda-12.6.0
[conda] cuda-cudart-static_linux-64 12.6.37 0 nvidia/label/cuda-12.6.0
[conda] cuda-cudart_linux-64 12.6.37 0 nvidia/label/cuda-12.6.0
[conda] cuda-cupti 12.6.37 0 nvidia/label/cuda-12.6.0
[conda] cuda-cupti-dev 12.6.37 0 nvidia/label/cuda-12.6.0
[conda] cuda-libraries 12.6.0 0 nvidia/label/cuda-12.6.0
[conda] cuda-libraries-dev 12.6.0 0 nvidia/label/cuda-12.6.0
[conda] cuda-nvrtc 12.6.20 0 nvidia/label/cuda-12.6.0
[conda] cuda-nvrtc-dev 12.6.20 0 nvidia/label/cuda-12.6.0
[conda] cuda-nvtx 12.6.37 0 nvidia/label/cuda-12.6.0
[conda] cuda-nvtx-dev 12.6.37 0 nvidia/label/cuda-12.6.0
[conda] cuda-opencl 12.6.37 0 nvidia/label/cuda-12.6.0
[conda] cuda-opencl-dev 12.6.37 0 nvidia/label/cuda-12.6.0
[conda] intel-openmp 2023.1.0 hdb19cb5_46306
[conda] libcublas 12.6.0.22 0 nvidia/label/cuda-12.6.0
[conda] libcublas-dev 12.6.0.22 0 nvidia/label/cuda-12.6.0
[conda] libcufft 11.2.6.28 0 nvidia/label/cuda-12.6.0
[conda] libcufft-dev 11.2.6.28 0 nvidia/label/cuda-12.6.0
[conda] libcurand 10.3.7.37 0 nvidia/label/cuda-12.6.0
[conda] libcurand-dev 10.3.7.37 0 nvidia/label/cuda-12.6.0
[conda] libcusolver 11.6.4.38 0 nvidia/label/cuda-12.6.0
[conda] libcusolver-dev 11.6.4.38 0 nvidia/label/cuda-12.6.0
[conda] libcusparse 12.5.2.23 0 nvidia/label/cuda-12.6.0
[conda] libcusparse-dev 12.5.2.23 0 nvidia/label/cuda-12.6.0
[conda] libjpeg-turbo 2.0.0 h9bf148f_0 pytorch
[conda] libnvjitlink 12.6.20 0 nvidia/label/cuda-12.6.0
[conda] libnvjitlink-dev 12.6.20 0 nvidia/label/cuda-12.6.0
[conda] mkl 2023.1.0 h213fc3f_46344
[conda] mkl-service 2.4.0 py311h5eee18b_1
[conda] mkl_fft 1.3.8 py311h5eee18b_0
[conda] mkl_random 1.2.4 py311hdb19cb5_0
[conda] nexfort 0.1.dev309+torch241cu121 pypi_0 pypi
[conda] numpy 1.26.4 pypi_0 pypi
[conda] nunchaku 0.2.0+torch2.6 pypi_0 pypi
[conda] nvidia-cublas-cu12 12.6.4.1 pypi_0 pypi
[conda] nvidia-cuda-cupti-cu12 12.6.80 pypi_0 pypi
[conda] nvidia-cuda-nvrtc-cu12 12.6.77 pypi_0 pypi
[conda] nvidia-cuda-runtime-cu12 12.6.77 pypi_0 pypi
[conda] nvidia-cudnn-cu12 9.10.2.21 pypi_0 pypi
[conda] nvidia-cufft-cu12 11.3.0.4 pypi_0 pypi
[conda] nvidia-curand-cu12 10.3.7.77 pypi_0 pypi
[conda] nvidia-cusolver-cu12 11.7.1.2 pypi_0 pypi
[conda] nvidia-cusparse-cu12 12.5.4.2 pypi_0 pypi
[conda] nvidia-cusparselt-cu12 0.7.1 pypi_0 pypi
[conda] nvidia-nccl-cu12 2.27.3 pypi_0 pypi
[conda] nvidia-nvjitlink-cu12 12.6.85 pypi_0 pypi
[conda] nvidia-nvtx-cu12 12.6.77 pypi_0 pypi
[conda] pytorch-lightning 2.3.3 pypi_0 pypi
[conda] pytorch-mutex 1.0 cuda pytorch
[conda] tbb 2021.8.0 hdb19cb5_0
[conda] torch 2.8.0+cu126 pypi_0 pypi
[conda] torchaudio 2.8.0+cu126 pypi_0 pypi
[conda] torchmetrics 1.4.0.post0 pypi_0 pypi
[conda] torchsde 0.2.6 pypi_0 pypi
[conda] torchvision 0.23.0+cu126 pypi_0 pypi
[conda] triton 3.4.0 pypi_0 pypi
cc @H-Huang @awgu @wanchaol @fegin @fduwjj @wz337 @wconstab @d4l3k @pragupta @ezyang