Skip to content

[Bug]: Unexpected Memory Usage in cutlass_moe_fp8() on Latest main 6bc7b57 #19923

New issue
New issue
Open
Open
[Bug]: Unexpected Memory Usage in cutlass_moe_fp8() on Latest main 6bc7b57#19923
Labels
bugSomething isn't working
@JackChuang

Description

@JackChuang
JackChuang
opened on Jun 20, 2025

Your current environment

The output of python collect_env.py 
# python3 collect_env.py
INFO 06-20 21:32:16 [__init__.py:239] Automatically detected platform cuda.
Collecting environment information...
/usr/local/lib/python3.10/dist-packages/_distutils_hack/__init__.py:30: UserWarning: Setuptools is replacing distutils. Support for replacing an already imported distutils is deprecated. In the future, this condition will fail. Register concerns at https://github.com/pypa/setuptools/issues/new?template=distutils-deprecation.yml
  warnings.warn(
==============================
        System Info
==============================
OS                           : Ubuntu 22.04.4 LTS (x86_64)
GCC version                  : (Ubuntu 11.4.0-1ubuntu1~22.04) 11.4.0
Clang version                : Could not collect
CMake version                : version 3.27.5
Libc version                 : glibc-2.35

==============================
       PyTorch Info
==============================
PyTorch version              : 2.6.0+cu124
Is debug build               : False
CUDA used to build PyTorch   : 12.4
ROCM used to build PyTorch   : N/A

==============================
      Python Environment
==============================
Python version               : 3.10.12 (main, Sep 11 2024, 15:47:36) [GCC 11.4.0] (64-bit runtime)
Python platform              : Linux-5.4.0-162-generic-x86_64-with-glibc2.35

==============================
       CUDA / GPU Info
==============================
Is CUDA available            : True
CUDA runtime version         : 12.4.131
CUDA_MODULE_LOADING set to   : LAZY
GPU models and configuration :
GPU 0: NVIDIA H20
GPU 1: NVIDIA H20
GPU 2: NVIDIA H20
GPU 3: NVIDIA H20
GPU 4: NVIDIA H20
GPU 5: NVIDIA H20
GPU 6: NVIDIA H20
GPU 7: NVIDIA H20

Nvidia driver version        : 535.161.08
cuDNN version                : Could not collect
HIP runtime version          : N/A
MIOpen runtime version       : N/A
Is XNNPACK available         : True

==============================
          CPU Info
==============================
Architecture:                       x86_64
CPU op-mode(s):                     32-bit, 64-bit
Address sizes:                      52 bits physical, 57 bits virtual
Byte Order:                         Little Endian
CPU(s):                             192
On-line CPU(s) list:                0-191
Vendor ID:                          GenuineIntel
Model name:                         Intel(R) Xeon(R) Platinum 8457C
CPU family:                         6
Model:                              143
Thread(s) per core:                 2
Core(s) per socket:                 48
Socket(s):                          2
Stepping:                           8
CPU max MHz:                        3800.0000
CPU min MHz:                        800.0000
BogoMIPS:                           5200.00
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 tsc_known_freq 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 cat_l2 cdp_l3 invpcid_single cdp_l2 ssbd mba ibrs ibpb stibp ibrs_enhanced tpr_shadow vnmi flexpriority ept vpid ept_ad fsgsbase tsc_adjust bmi1 avx2 smep bmi2 erms invpcid cqm rdt_a avx512f avx512dq rdseed adx smap avx512ifma clflushopt clwb intel_pt avx512cd sha_ni avx512bw avx512vl xsaveopt xsavec xgetbv1 xsaves cqm_llc cqm_occup_llc cqm_mbm_total cqm_mbm_local avx512_bf16 wbnoinvd dtherm ida arat pln pts hwp hwp_act_window hwp_epp hwp_pkg_req avx512vbmi umip pku ospke waitpkg avx512_vbmi2 gfni vaes vpclmulqdq avx512_vnni avx512_bitalg tme avx512_vpopcntdq rdpid cldemote movdiri movdir64b md_clear pconfig flush_l1d arch_capabilities
Virtualization:                     VT-x
L1d cache:                          4.5 MiB (96 instances)
L1i cache:                          3 MiB (96 instances)
L2 cache:                           192 MiB (96 instances)
L3 cache:                           195 MiB (2 instances)
NUMA node(s):                       2
NUMA node0 CPU(s):                  0-47,96-143
NUMA node1 CPU(s):                  48-95,144-191
Vulnerability Gather data sampling: Not affected
Vulnerability Itlb multihit:        Not affected
Vulnerability L1tf:                 Not affected
Vulnerability Mds:                  Not affected
Vulnerability Meltdown:             Not affected
Vulnerability Mmio stale data:      Not affected
Vulnerability Retbleed:             Not affected
Vulnerability Spec store bypass:    Mitigation; Speculative Store Bypass disabled via prctl and seccomp
Vulnerability Spectre v1:           Mitigation; usercopy/swapgs barriers and __user pointer sanitization
Vulnerability Spectre v2:           Mitigation; Enhanced IBRS, IBPB conditional, RSB filling, PBRSB-eIBRS SW sequence
Vulnerability Srbds:                Not affected
Vulnerability Tsx async abort:      Not affected

==============================
Versions of relevant libraries
==============================
[pip3] numpy==1.26.4
[pip3] nvidia-cublas-cu12==12.4.5.8
[pip3] nvidia-cuda-cupti-cu12==12.4.127
[pip3] nvidia-cuda-nvrtc-cu12==12.4.127
[pip3] nvidia-cuda-runtime-cu12==12.4.127
[pip3] nvidia-cudnn-cu12==9.1.0.70
[pip3] nvidia-cufft-cu12==11.2.1.3
[pip3] nvidia-cufile-cu12==1.11.1.6
[pip3] nvidia-curand-cu12==10.3.5.147
[pip3] nvidia-cusolver-cu12==11.6.1.9
[pip3] nvidia-cusparse-cu12==12.3.1.170
[pip3] nvidia-cusparselt-cu12==0.6.2
[pip3] nvidia-ml-py==12.560.30
[pip3] nvidia-modelopt==0.25.0
[pip3] nvidia-modelopt-core==0.25.0
[pip3] nvidia-nccl-cu12==2.21.5
[pip3] nvidia-nvjitlink-cu12==12.4.127
[pip3] nvidia-nvtx-cu12==12.4.127
[pip3] onnx==1.17.0
[pip3] onnx_graphsurgeon==0.5.8
[pip3] pynvml==12.0.0
[pip3] pyzmq==26.2.0
[pip3] torch==2.6.0
[pip3] torchaudio==2.6.0
[pip3] torchprofile==0.0.4
[pip3] torchvision==0.21.0
[pip3] transformers==4.51.2
[pip3] triton==3.2.0
[conda] Could not collect

==============================
         vLLM Info
==============================
ROCM Version                 : Could not collect
Neuron SDK Version           : N/A
vLLM Version                 : 0.8.5.post2.dev1+g35a4376c9.d20250618 (git sha: 35a4376c9, date: 20250618)
vLLM Build Flags:
  CUDA Archs: Not Set; ROCm: Disabled; Neuron: Disabled
GPU Topology:
        GPU0    GPU1    GPU2    GPU3    GPU4    GPU5    GPU6    GPU7    NIC0    NIC1    NIC2    NIC3    CPU Affinity    NUMA Affinity   GPU NUMA ID
GPU0     X      NV18    NV18    NV18    NV18    NV18    NV18    NV18    PIX     NODE    SYS     SYS     0-47,96-143     0               N/A
GPU1    NV18     X      NV18    NV18    NV18    NV18    NV18    NV18    NODE    NODE    SYS     SYS     0-47,96-143     0               N/A
GPU2    NV18    NV18     X      NV18    NV18    NV18    NV18    NV18    NODE    PIX     SYS     SYS     0-47,96-143     0               N/A
GPU3    NV18    NV18    NV18     X      NV18    NV18    NV18    NV18    NODE    NODE    SYS     SYS     0-47,96-143     0               N/A
GPU4    NV18    NV18    NV18    NV18     X      NV18    NV18    NV18    SYS     SYS     PIX     NODE    48-95,144-191   1               N/A
GPU5    NV18    NV18    NV18    NV18    NV18     X      NV18    NV18    SYS     SYS     NODE    NODE    48-95,144-191   1               N/A
GPU6    NV18    NV18    NV18    NV18    NV18    NV18     X      NV18    SYS     SYS     NODE    PIX     48-95,144-191   1               N/A
GPU7    NV18    NV18    NV18    NV18    NV18    NV18    NV18     X      SYS     SYS     NODE    NODE    48-95,144-191   1               N/A
NIC0    PIX     NODE    NODE    NODE    SYS     SYS     SYS     SYS      X      NODE    SYS     SYS
NIC1    NODE    NODE    PIX     NODE    SYS     SYS     SYS     SYS     NODE     X      SYS     SYS
NIC2    SYS     SYS     SYS     SYS     PIX     NODE    NODE    NODE    SYS     SYS      X      NODE
NIC3    SYS     SYS     SYS     SYS     NODE    NODE    PIX     NODE    SYS     SYS     NODE     X

Legend:

  X    = Self
  SYS  = Connection traversing PCIe as well as the SMP interconnect between NUMA nodes (e.g., QPI/UPI)
  NODE = Connection traversing PCIe as well as the interconnect between PCIe Host Bridges within a NUMA node
  PHB  = Connection traversing PCIe as well as a PCIe Host Bridge (typically the CPU)
  PXB  = Connection traversing multiple PCIe bridges (without traversing the PCIe Host Bridge)
  PIX  = Connection traversing at most a single PCIe bridge
  NV#  = Connection traversing a bonded set of # NVLinks

NIC Legend:

  NIC0: mlx5_1
  NIC1: mlx5_2
  NIC2: mlx5_3
  NIC3: mlx5_4

==============================
     Environment Variables
==============================
NVIDIA_VISIBLE_DEVICES=all
NVIDIA_REQUIRE_CUDA=cuda>=12.4 brand=tesla,driver>=470,driver<471 brand=unknown,driver>=470,driver<471 brand=nvidia,driver>=470,driver<471 brand=nvidiartx,driver>=470,driver<471 brand=geforce,driver>=470,driver<471 brand=geforcertx,driver>=470,driver<471 brand=quadro,driver>=470,driver<471 brand=quadrortx,driver>=470,driver<471 brand=titan,driver>=470,driver<471 brand=titanrtx,driver>=470,driver<471 brand=tesla,driver>=525,driver<526 brand=unknown,driver>=525,driver<526 brand=nvidia,driver>=525,driver<526 brand=nvidiartx,driver>=525,driver<526 brand=geforce,driver>=525,driver<526 brand=geforcertx,driver>=525,driver<526 brand=quadro,driver>=525,driver<526 brand=quadrortx,driver>=525,driver<526 brand=titan,driver>=525,driver<526 brand=titanrtx,driver>=525,driver<526 brand=tesla,driver>=535,driver<536 brand=unknown,driver>=535,driver<536 brand=nvidia,driver>=535,driver<536 brand=nvidiartx,driver>=535,driver<536 brand=geforce,driver>=535,driver<536 brand=geforcertx,driver>=535,driver<536 brand=quadro,driver>=535,driver<536 brand=quadrortx,driver>=535,driver<536 brand=titan,driver>=535,driver<536 brand=titanrtx,driver>=535,driver<536
NCCL_MIN_NCHANNELS=24
NCCL_VERSION=2.20.5-1
NVIDIA_DRIVER_CAPABILITIES=compute,utility
NCCL_DEBUG=INFO
VLLM_WORKER_MULTIPROC_METHOD=spawn
NVIDIA_PRODUCT_NAME=CUDA
CUDA_VERSION=12.4.0
CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7
CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7
NCCL_IB_QPS_PER_CONNECTION=8
LD_LIBRARY_PATH=/usr/local/nvidia/lib:/usr/local/nvidia/lib64
VLLM_HOST_IP=192.168.12.2
NCCL_CUMEM_ENABLE=0
PYTORCH_NVML_BASED_CUDA_CHECK=1
TORCHINDUCTOR_COMPILE_THREADS=1
CUDA_MODULE_LOADING=LAZY```

</details>


### 🐛 Describe the bug

### Summary

We discovered a significant memory issue with the latest `cutlass_moe_fp8()` implementation in the `main` branch of vLLM. The function now allocates **at least 24.5 GB** of GPU memory per rank, which appears **abnormally high** and prevents us from rebasing our [PR #19843](https://github.com/vllm-project/vllm/pull/19843) onto `main`.

This issue did **not** exist in `v0.8.5.post1`, where the memory footprint of the same function was much smaller under otherwise identical conditions.

### Error Log
<details>
<summary>Click to expand</summary>

```python
...
  File "/nvme0n1/jack/vllm-w8a8-cutlass/vllm/worker/model_runner.py", line 1426, in _dummy_run
    self.execute_model(model_input, kv_caches, intermediate_tensors)
  File "/usr/local/lib/python3.10/dist-packages/torch/utils/_contextlib.py", line 116, in decorate_context
    return func(*args, **kwargs)
  File "/nvme0n1/jack/vllm-w8a8-cutlass/vllm/worker/model_runner.py", line 1844, in execute_model
    hidden_or_intermediate_states = model_executable(
  File "/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py", line 1751, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
  File "/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py", line 1762, in _call_impl
    return forward_call(*args, **kwargs)
  File "/nvme0n1/jack/vllm-w8a8-cutlass/vllm/model_executor/models/deepseek_v2.py", line 714, in forward
    hidden_states = self.model(input_ids, positions, intermediate_tensors,
  File "/nvme0n1/jack/vllm-w8a8-cutlass/vllm/compilation/decorators.py", line 173, in __call__
    return self.forward(*args, **kwargs)
  File "/nvme0n1/jack/vllm-w8a8-cutlass/vllm/model_executor/models/deepseek_v2.py", line 672, in forward
    hidden_states, residual = layer(positions, hidden_states, residual)
  File "/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py", line 1751, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
  File "/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py", line 1762, in _call_impl
    return forward_call(*args, **kwargs)
  File "/nvme0n1/jack/vllm-w8a8-cutlass/vllm/model_executor/models/deepseek_v2.py", line 592, in forward
    hidden_states = self.mlp(hidden_states)
  File "/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py", line 1751, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
  File "/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py", line 1762, in _call_impl
    return forward_call(*args, **kwargs)
  File "/nvme0n1/jack/vllm-w8a8-cutlass/vllm/model_executor/models/deepseek_v2.py", line 160, in forward
    final_hidden_states = self.experts(
  File "/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py", line 1751, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
  File "/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py", line 1762, in _call_impl
    return forward_call(*args, **kwargs)
  File "/nvme0n1/jack/vllm-w8a8-cutlass/vllm/model_executor/layers/fused_moe/modular_kernel.py", line 426, in forward
    fused_out = self._do_fused_experts(
  File "/nvme0n1/jack/vllm-w8a8-cutlass/vllm/model_executor/layers/fused_moe/modular_kernel.py", line 318, in _do_fused_experts
    workspace13 = torch.zeros(workspace13_shape,
torch.OutOfMemoryError: CUDA out of memory. Tried to allocate 24.50 GiB. GPU 0 has a total capacity of 95.22 GiB of which 13.66 GiB is free. Process 4078568 has 81.55 GiB memory in use. Of the allocated memory 79.79 GiB is allocated by PyTorch, and 108.60 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)
[rank0]:[W618 00:18:45.548079750 ProcessGroupNCCL.cpp:1476] Warning: WARNING: destroy_process_group() was not called before program exit, which can leak resources. For more info, please see https://pytorch.org/docs/stable/distributed.html#shutdown (function operator())

💡 Context

  • Model: DeepSeek-R1
  • GPUs: 8x NVIDIA H20
  • Tensor Parallelism (TP): 8
  • Quantization: --quantization fp8
  • vLLM Version (working): v0.8.5.post1
  • vLLM Version (issue): main branch (latest: 6bc7b57)
  • Backend: Cutlass MoE FP8 (cutlass_moe_fp8())

🔁 Reproduction Steps

It's a bit hard but

  1. Check out the latest main branch of vLLM (6bc7b57).
  2. Apply changes PR #19843
  3. Follow the setup procedure described in PR #19843 to run the test. (VLLM_USE_CUTLASS_MOE_FP8=1)
  4. Hit the error

🧨 Observed Behavior

  • GPU memory usage increases by at least 24.5 GB per rank due to a tensor allocation inside cutlass_moe_fp8(). (See Error Log)
  • The same workload on v0.8.5.post1 uses much less memory, with no functional degradation.

✅ Expected Behavior

  • Memory usage should remain similar to previous versions.

🔗 Related PR

This issue blocks the validation and merging of our contribution:

  • PR: #19843 — Add Cutlass backend for MoE FP8 workloads
    • Summary: We introduced a Cutlass kernel for MoE FP8 workloads with strong throughput gains. It is currently tested and validated under v0.8.5.post1, but we are unable to rebase it onto main due to the unexpected memory increase.

🙏 Request

We would really appreciate:

  • A review or fix of the memory allocation behavior in the latest cutlass_moe_fp8() implementation.
  • Any guidance that would help us rebase and verify our PR (Add Cutlass integration for MoE FP8 #19843) against main.

Thanks!

Before submitting a new issue...

  • Make sure you already searched for relevant issues, and asked the chatbot living at the bottom right corner of the documentation page, which can answer lots of frequently asked questions.

Metadata

Metadata

Assignees

No one assigned

    Labels

    bugSomething isn't working

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions