vllm - ✅(Solved) Fix [Bug]: Performance degradation on MoE models with low batch sizes since vLLM v0.20.0 [1 pull requests, 1 comments, 1 participants]

vllm2026-05-19 12:15:01

ON THIS PAGE

Recommended Tools

×6

Utilities matched from this issue’s tags and category — try them while you read without losing context.

GitHub issue graph ai analysis

Paste a GitHub issue URL. We fetch that issue, discover linked issues from bodies/comments/timeline, collect linked pull requests, and produce a structured English report.

The report is written in English Markdown for sharing and archival.

GitHub issue URL

Helpful · Quick feedback

GitHub stats

vllm-project/vllm#43096•Fetched 2026-05-20 03:39:50

View on GitHub

Comments

Participants

Timeline

Reactions

Author

Wats0ns

Participants

Wats0ns

Timeline (top)

closed ×1commented ×1cross-referenced ×1labeled ×1

Root Cause

Tried to run a bisect but lot of the wheels installation fail because of cuda installation failure. Any idea or hint is appreciated !

Fix Action

Fix / Workaround

============================== 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): 224 On-line CPU(s) list: 0-223
Vendor ID: GenuineIntel Model name: Intel(R) Xeon(R) Platinum 8480CL CPU family: 6 Model: 143 Thread(s) per core: 2 Core(s) per socket: 56 Socket(s): 2 Stepping: 7 CPU max MHz: 3800.0000 CPU min MHz: 800.0000 BogoMIPS: 4000.00 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 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 ep b cat_l3 cat_l2 cdp_l3 invpcid_single intel_ppin cdp_l2 ssbd mba ibrs ibpb stibp ibrs_enhanced 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 split_lock_detect avx_vnni avx512_bf16 wbnoinvd dtherm ida arat pln pts hwp hwp_act_window hwp_epp hwp_pkg_req avx512vbmi umip pku ospke waitpkg avx512_vbmi 2 gfni vaes vpclmulqdq avx512_vnni avx512_bitalg tme avx512_vpopcntdq la57 rdpid bus_lock_detect cldemote movdiri movdir64b enqcmd fsrm md_clear serialize tsxldtrk pc onfig arch_lbr amx_bf16 avx512_fp16 amx_tile amx_int8 flush_l1d arch_capabilities L1d cache: 5.3 MiB (112 instances) L1i cache: 3.5 MiB (112 instances) L2 cache: 224 MiB (112 instances) L3 cache: 210 MiB (2 instances) NUMA node(s): 2 NUMA node0 CPU(s): 0-55,112-167 NUMA node1 CPU(s): 56-111,168-223 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 rstack overflow: 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

PR fix notes

PR #43135: [Perf][gpt-oss] Downgrade triton_kernels to v3.5.1

Repository: vllm-project/vllm
Author: mgoin
State: open | merged: False
Link: https://github.com/vllm-project/vllm/pull/43135

Description (problem / solution / changelog)

Recovers a decode-time regression on gpt-oss models after the triton_kernels v3.5.1 → v3.6.0 bump in #30525 and the routing rewrite in #38504.

Pin triton_kernels back to v3.5.1.
Activate the legacy routing path whenever SparseMatrix is missing (previously gated to ROCm only).
When no expert map is set, call the fused routing() kernel directly instead of running topk_fn + pack_bitmatrix + routing_from_bitmatrix as three separate launches.

H100, gpt-oss-20b, TP=1, 1024 in / 1024 out, concurrency=1:

version	output tok/s	ITL (ms)
v0.16.0 (3.5.1)	307.4	3.23
main (3.6.0)	230.1	4.32
this PR	307.4	3.22

GPQA (gpt-oss-20b-baseline.yaml, TP=2, low reasoning): 0.566, matches the 0.568 CI baseline.

Upstream perf bug tracked at https://github.com/triton-lang/triton/issues/9969. Alternative to #39236 with a smaller diff that keeps the v3.6.0+ code path as a compat shim.

Changed files

cmake/external_projects/triton_kernels.cmake (modified, +1/-1)
tests/kernels/quantization/test_mxfp4_triton_ep.py (modified, +37/-19)
vllm/model_executor/layers/fused_moe/experts/gpt_oss_triton_kernels_moe.py (modified, +57/-32)

Code Example

==============================
        System Info
==============================
OS                           : Ubuntu 22.04.5 LTS (x86_64)
GCC version                  : (Ubuntu 11.4.0-1ubuntu1~22.04.3) 11.4.0
Clang version                : Could not collect
CMake version                : Could not collect
Libc version                 : glibc-2.35  

==============================
       PyTorch Info
==============================
PyTorch version              : 2.11.0+cu130
Is debug build               : False
CUDA used to build PyTorch   : 13.0
ROCM used to build PyTorch   : N/A
XPU used to build PyTorch    : N/A

==============================
      Python Environment
==============================
Python version               : 3.12.13 (main, Mar  4 2026, 09:23:07) [GCC 11.4.0] (64-bit runtime)
Python platform              : Linux-5.15.0-1053-nvidia-x86_64-with-glibc2.35

==============================
       CUDA / GPU Info
==============================
Is CUDA available            : True
CUDA runtime version         : 13.0.88
CUDA_MODULE_LOADING set to   :
GPU models and configuration :
GPU 0: NVIDIA H100 80GB HBM3
GPU 1: NVIDIA H100 80GB HBM3

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):                             224
On-line CPU(s) list:                0-223  
Vendor ID:                          GenuineIntel
Model name:                         Intel(R) Xeon(R) Platinum 8480CL
CPU family:                         6
Model:                              143
Thread(s) per core:                 2
Core(s) per socket:                 56
Socket(s):                          2
Stepping:                           7
CPU max MHz:                        3800.0000
CPU min MHz:                        800.0000
BogoMIPS:                           4000.00
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 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 ep
b cat_l3 cat_l2 cdp_l3 invpcid_single intel_ppin cdp_l2 ssbd mba ibrs ibpb stibp ibrs_enhanced 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 split_lock_detect avx_vnni avx512_bf16 wbnoinvd dtherm ida arat pln pts hwp hwp_act_window hwp_epp hwp_pkg_req avx512vbmi umip pku ospke waitpkg avx512_vbmi
2 gfni vaes vpclmulqdq avx512_vnni avx512_bitalg tme avx512_vpopcntdq la57 rdpid bus_lock_detect cldemote movdiri movdir64b enqcmd fsrm md_clear serialize tsxldtrk pc
onfig arch_lbr amx_bf16 avx512_fp16 amx_tile amx_int8 flush_l1d arch_capabilities
L1d cache:                          5.3 MiB (112 instances)
L1i cache:                          3.5 MiB (112 instances)
L2 cache:                           224 MiB (112 instances)
L3 cache:                           210 MiB (2 instances)
NUMA node(s):                       2
NUMA node0 CPU(s):                  0-55,112-167
NUMA node1 CPU(s):                  56-111,168-223
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 rstack overflow: 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] flashinfer-python==0.6.8.post1
[pip3] numpy==2.2.6
[pip3] nvidia-cublas==13.1.0.3
[pip3] nvidia-cuda-cupti==13.0.85
[pip3] nvidia-cuda-nvrtc==13.0.88
[pip3] nvidia-cuda-runtime==13.0.96
[pip3] nvidia-cudnn-cu13==9.19.0.56
[pip3] nvidia-cudnn-frontend==1.18.0
[pip3] nvidia-cufft==12.0.0.61
[pip3] nvidia-cufile==1.15.1.6
[pip3] nvidia-curand==10.4.0.35
[pip3] nvidia-cusolver==12.0.4.66
[pip3] nvidia-cusparse==12.6.3.3
[pip3] nvidia-cusparselt-cu13==0.8.0
[pip3] nvidia-cutlass-dsl==4.4.2
[pip3] nvidia-cutlass-dsl-libs-base==4.4.2 
[pip3] nvidia-ml-py==13.595.45
[pip3] nvidia-nccl-cu13==2.28.9
[pip3] nvidia-nvjitlink==13.0.88
[pip3] nvidia-nvshmem-cu13==3.4.5
[pip3] nvidia-nvtx==13.0.85
[pip3] pyzmq==27.1.0
[pip3] tokenspeed-triton==3.7.10.post20260505
[pip3] torch==2.11.0+cu130
[pip3] torch_c_dlpack_ext==0.1.5
[pip3] torchaudio==2.11.0+cu130
[pip3] torchvision==0.26.0+cu130
[pip3] transformers==5.8.1
[pip3] triton==3.6.0
[conda] Could not collect

==============================
         vLLM Info
==============================
ROCM Version                 : Could not collect
vLLM Version                 : 0.21.0
vLLM Build Flags:
  CUDA Archs: Hopper; ROCm: Disabled; XPU: Disabled
GPU Topology:
        GPU0    GPU1    NIC0    NIC1    NIC2    NIC3    NIC4    NIC5    NIC6    NIC7    NIC8    NIC9    NIC10   NIC11   CPU Affinity    NUMA Affinity   GPU NUMA ID
GPU0     X      NV18    NODE    NODE    NODE    NODE    PXB     NODE    SYS     SYS     SYS     SYS     SYS     SYS     0-55,112-167    0               N/A
GPU1    NV18     X      SYS     SYS     SYS     SYS     SYS     SYS     NODE    NODE    NODE    NODE    NODE    PXB     56-111,168-223  1               N/A
NIC0    NODE    SYS      X      NODE    NODE    NODE    NODE    NODE    SYS     SYS     SYS     SYS     SYS     SYS
NIC1    NODE    SYS     NODE     X      PIX     NODE    NODE    NODE    SYS     SYS     SYS     SYS     SYS     SYS
NIC2    NODE    SYS     NODE    PIX      X      NODE    NODE    NODE    SYS     SYS     SYS     SYS     SYS     SYS
NIC3    NODE    SYS     NODE    NODE    NODE     X      NODE    NODE    SYS     SYS     SYS     SYS     SYS     SYS
NIC4    PXB     SYS     NODE    NODE    NODE    NODE     X      NODE    SYS     SYS     SYS     SYS     SYS     SYS
NIC5    NODE    SYS     NODE    NODE    NODE    NODE    NODE     X      SYS     SYS     SYS     SYS     SYS     SYS
NIC6    SYS     NODE    SYS     SYS     SYS     SYS     SYS     SYS      X      NODE    NODE    NODE    NODE    NODE
NIC7    SYS     NODE    SYS     SYS     SYS     SYS     SYS     SYS     NODE     X      PIX     NODE    NODE    NODE
NIC8    SYS     NODE    SYS     SYS     SYS     SYS     SYS     SYS     NODE    PIX      X      NODE    NODE    NODE
NIC9    SYS     NODE    SYS     SYS     SYS     SYS     SYS     SYS     NODE    NODE    NODE     X      NODE    NODE
NIC10   SYS     NODE    SYS     SYS     SYS     SYS     SYS     SYS     NODE    NODE    NODE    NODE     X      NODE
NIC11   SYS     PXB     SYS     SYS     SYS     SYS     SYS     SYS     NODE    NODE    NODE    NODE    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_0
  NIC1: mlx5_1
  NIC2: mlx5_2
  NIC3: mlx5_3
  NIC4: mlx5_4
  NIC5: mlx5_5
  NIC6: mlx5_6
  NIC7: mlx5_7
  NIC8: mlx5_8
  NIC9: mlx5_9
  NIC10: mlx5_10
  NIC11: mlx5_11

==============================
     Environment Variables
==============================
NVIDIA_VISIBLE_DEVICES=GPU-d8ce80af-6ce3-6e64-7aef-0b74d44ac9ca,GPU-20217964-b916-91ca-11b7-fb615dd43bbb
NVIDIA_REQUIRE_CUDA=cuda
VLLM_GPT_OSS_HARMONY_SYSTEM_INSTRUCTIONS=1 
TORCH_CUDA_ARCH_LIST=Hopper
PYTORCH_ALLOC_CONF=expandable_segments:True
NVIDIA_DRIVER_CAPABILITIES=compute,utility 
VLLM_USAGE_SOURCE=production-docker-image  
CUDA_VERSION=13.0.2
VLLM_ENABLE_CUDA_COMPATIBILITY=0
LD_LIBRARY_PATH=/usr/local/nvidia/lib64:/usr/local/cuda/lib64:/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/local/cuda/lib64
PYTORCH_NVML_BASED_CUDA_CHECK=1
TORCHINDUCTOR_COMPILE_THREADS=1
TORCHINDUCTOR_CACHE_DIR=/tmp/torchinductor_user

---

MODEL=openai/gpt-oss-120b
vllm bench throughput --model $MODEL--num-prompts 200   --random-input-len 600   --random-output-len 600 --max-num-seqs=10 --tensor-parallel-size=2

---

Throughput: 1.88 requests/s, 2252.27 total tokens/s, 1126.14 output tokens/s
Total num prompt tokens:  120000
Total num output tokens:  120000

---

Throughput: 1.46 requests/s, 1750.12 total tokens/s, 875.06 output tokens/s
Total num prompt tokens:  120000
Total num output tokens:  120000

---

Throughput: 1.46 requests/s, 1750.30 total tokens/s, 875.15 output tokens/s
Total num prompt tokens:  120000
Total num output tokens:  120000

RAW_BUFFERClick to expand / collapse

Your current environment

<details> <summary>The output of <code>python collect_env.py</code></summary>

==============================
        System Info
==============================
OS                           : Ubuntu 22.04.5 LTS (x86_64)
GCC version                  : (Ubuntu 11.4.0-1ubuntu1~22.04.3) 11.4.0
Clang version                : Could not collect
CMake version                : Could not collect
Libc version                 : glibc-2.35  

==============================
       PyTorch Info
==============================
PyTorch version              : 2.11.0+cu130
Is debug build               : False
CUDA used to build PyTorch   : 13.0
ROCM used to build PyTorch   : N/A
XPU used to build PyTorch    : N/A

==============================
      Python Environment
==============================
Python version               : 3.12.13 (main, Mar  4 2026, 09:23:07) [GCC 11.4.0] (64-bit runtime)
Python platform              : Linux-5.15.0-1053-nvidia-x86_64-with-glibc2.35

==============================
       CUDA / GPU Info
==============================
Is CUDA available            : True
CUDA runtime version         : 13.0.88
CUDA_MODULE_LOADING set to   :
GPU models and configuration :
GPU 0: NVIDIA H100 80GB HBM3
GPU 1: NVIDIA H100 80GB HBM3

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):                             224
On-line CPU(s) list:                0-223  
Vendor ID:                          GenuineIntel
Model name:                         Intel(R) Xeon(R) Platinum 8480CL
CPU family:                         6
Model:                              143
Thread(s) per core:                 2
Core(s) per socket:                 56
Socket(s):                          2
Stepping:                           7
CPU max MHz:                        3800.0000
CPU min MHz:                        800.0000
BogoMIPS:                           4000.00
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 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 ep
b cat_l3 cat_l2 cdp_l3 invpcid_single intel_ppin cdp_l2 ssbd mba ibrs ibpb stibp ibrs_enhanced 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 split_lock_detect avx_vnni avx512_bf16 wbnoinvd dtherm ida arat pln pts hwp hwp_act_window hwp_epp hwp_pkg_req avx512vbmi umip pku ospke waitpkg avx512_vbmi
2 gfni vaes vpclmulqdq avx512_vnni avx512_bitalg tme avx512_vpopcntdq la57 rdpid bus_lock_detect cldemote movdiri movdir64b enqcmd fsrm md_clear serialize tsxldtrk pc
onfig arch_lbr amx_bf16 avx512_fp16 amx_tile amx_int8 flush_l1d arch_capabilities
L1d cache:                          5.3 MiB (112 instances)
L1i cache:                          3.5 MiB (112 instances)
L2 cache:                           224 MiB (112 instances)
L3 cache:                           210 MiB (2 instances)
NUMA node(s):                       2
NUMA node0 CPU(s):                  0-55,112-167
NUMA node1 CPU(s):                  56-111,168-223
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 rstack overflow: 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] flashinfer-python==0.6.8.post1
[pip3] numpy==2.2.6
[pip3] nvidia-cublas==13.1.0.3
[pip3] nvidia-cuda-cupti==13.0.85
[pip3] nvidia-cuda-nvrtc==13.0.88
[pip3] nvidia-cuda-runtime==13.0.96
[pip3] nvidia-cudnn-cu13==9.19.0.56
[pip3] nvidia-cudnn-frontend==1.18.0
[pip3] nvidia-cufft==12.0.0.61
[pip3] nvidia-cufile==1.15.1.6
[pip3] nvidia-curand==10.4.0.35
[pip3] nvidia-cusolver==12.0.4.66
[pip3] nvidia-cusparse==12.6.3.3
[pip3] nvidia-cusparselt-cu13==0.8.0
[pip3] nvidia-cutlass-dsl==4.4.2
[pip3] nvidia-cutlass-dsl-libs-base==4.4.2 
[pip3] nvidia-ml-py==13.595.45
[pip3] nvidia-nccl-cu13==2.28.9
[pip3] nvidia-nvjitlink==13.0.88
[pip3] nvidia-nvshmem-cu13==3.4.5
[pip3] nvidia-nvtx==13.0.85
[pip3] pyzmq==27.1.0
[pip3] tokenspeed-triton==3.7.10.post20260505
[pip3] torch==2.11.0+cu130
[pip3] torch_c_dlpack_ext==0.1.5
[pip3] torchaudio==2.11.0+cu130
[pip3] torchvision==0.26.0+cu130
[pip3] transformers==5.8.1
[pip3] triton==3.6.0
[conda] Could not collect

==============================
         vLLM Info
==============================
ROCM Version                 : Could not collect
vLLM Version                 : 0.21.0
vLLM Build Flags:
  CUDA Archs: Hopper; ROCm: Disabled; XPU: Disabled
GPU Topology:
        GPU0    GPU1    NIC0    NIC1    NIC2    NIC3    NIC4    NIC5    NIC6    NIC7    NIC8    NIC9    NIC10   NIC11   CPU Affinity    NUMA Affinity   GPU NUMA ID
GPU0     X      NV18    NODE    NODE    NODE    NODE    PXB     NODE    SYS     SYS     SYS     SYS     SYS     SYS     0-55,112-167    0               N/A
GPU1    NV18     X      SYS     SYS     SYS     SYS     SYS     SYS     NODE    NODE    NODE    NODE    NODE    PXB     56-111,168-223  1               N/A
NIC0    NODE    SYS      X      NODE    NODE    NODE    NODE    NODE    SYS     SYS     SYS     SYS     SYS     SYS
NIC1    NODE    SYS     NODE     X      PIX     NODE    NODE    NODE    SYS     SYS     SYS     SYS     SYS     SYS
NIC2    NODE    SYS     NODE    PIX      X      NODE    NODE    NODE    SYS     SYS     SYS     SYS     SYS     SYS
NIC3    NODE    SYS     NODE    NODE    NODE     X      NODE    NODE    SYS     SYS     SYS     SYS     SYS     SYS
NIC4    PXB     SYS     NODE    NODE    NODE    NODE     X      NODE    SYS     SYS     SYS     SYS     SYS     SYS
NIC5    NODE    SYS     NODE    NODE    NODE    NODE    NODE     X      SYS     SYS     SYS     SYS     SYS     SYS
NIC6    SYS     NODE    SYS     SYS     SYS     SYS     SYS     SYS      X      NODE    NODE    NODE    NODE    NODE
NIC7    SYS     NODE    SYS     SYS     SYS     SYS     SYS     SYS     NODE     X      PIX     NODE    NODE    NODE
NIC8    SYS     NODE    SYS     SYS     SYS     SYS     SYS     SYS     NODE    PIX      X      NODE    NODE    NODE
NIC9    SYS     NODE    SYS     SYS     SYS     SYS     SYS     SYS     NODE    NODE    NODE     X      NODE    NODE
NIC10   SYS     NODE    SYS     SYS     SYS     SYS     SYS     SYS     NODE    NODE    NODE    NODE     X      NODE
NIC11   SYS     PXB     SYS     SYS     SYS     SYS     SYS     SYS     NODE    NODE    NODE    NODE    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_0
  NIC1: mlx5_1
  NIC2: mlx5_2
  NIC3: mlx5_3
  NIC4: mlx5_4
  NIC5: mlx5_5
  NIC6: mlx5_6
  NIC7: mlx5_7
  NIC8: mlx5_8
  NIC9: mlx5_9
  NIC10: mlx5_10
  NIC11: mlx5_11

==============================
     Environment Variables
==============================
NVIDIA_VISIBLE_DEVICES=GPU-d8ce80af-6ce3-6e64-7aef-0b74d44ac9ca,GPU-20217964-b916-91ca-11b7-fb615dd43bbb
NVIDIA_REQUIRE_CUDA=cuda
VLLM_GPT_OSS_HARMONY_SYSTEM_INSTRUCTIONS=1 
TORCH_CUDA_ARCH_LIST=Hopper
PYTORCH_ALLOC_CONF=expandable_segments:True
NVIDIA_DRIVER_CAPABILITIES=compute,utility 
VLLM_USAGE_SOURCE=production-docker-image  
CUDA_VERSION=13.0.2
VLLM_ENABLE_CUDA_COMPATIBILITY=0
LD_LIBRARY_PATH=/usr/local/nvidia/lib64:/usr/local/cuda/lib64:/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/local/cuda/lib64
PYTORCH_NVML_BASED_CUDA_CHECK=1
TORCHINDUCTOR_COMPILE_THREADS=1
TORCHINDUCTOR_CACHE_DIR=/tmp/torchinductor_user

</details>

🐛 Describe the bug

Hello, I have observed a performance decrease in MoE models, notably gpt-oss-120b since vLLM v0.20.0

I use this bench command to measure the performance, on 2xH100 (same performance drop with and without NvLink)

MODEL=openai/gpt-oss-120b
vllm bench throughput --model $MODEL--num-prompts 200   --random-input-len 600   --random-output-len 600 --max-num-seqs=10 --tensor-parallel-size=2

The performance drops happens only with --max-num-seqs=10, if I don't set it (so batch sizes of 1000), no performance drop is observed

On docker image vllm/vllm-openai:v0.19.1:

Throughput: 1.88 requests/s, 2252.27 total tokens/s, 1126.14 output tokens/s
Total num prompt tokens:  120000
Total num output tokens:  120000

On vllm/vllm-openai:v0.20.0:

Throughput: 1.46 requests/s, 1750.12 total tokens/s, 875.06 output tokens/s
Total num prompt tokens:  120000
Total num output tokens:  120000

On vllm/vllm-openai:v0.20.0-cu129

Throughput: 1.46 requests/s, 1750.30 total tokens/s, 875.15 output tokens/s
Total num prompt tokens:  120000
Total num output tokens:  120000

No configuration changes in the log output of vLLM startup, only flags that change are:

fast_moe_cold_start (from True to False in v0.20.0), changing it has no impact
quantization=mxfp4 changed to quantization=gpt_oss_mxfp4, but from my investigations this seems to be a simple renaming

I tried the following:

used VLLM_MXFP4_USE_MARLIN=1 VLLM_MXFP4_USE_TRITON=1, same issue
set VLLM_ATTENTION_BACKEND=TRITON_ATTN, same issue
benched on meta-llama/Llama-3.1-8B-Instruct, difference is minimal (2218 tok/s vs 2090 tok/s)

Tried to run a bisect but lot of the wheels installation fail because of cuda installation failure. Any idea or hint is appreciated !

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.

Vote matrix · Quick signals

Works

Did the solution work? Tap to confirm.

Easy Fix

Was it a quick fix?

Time Saver

Did it save you time?

Blocking

Was it severely blocking?

Common Issue

Are others likely hitting this too?

Flaky / Intermittent

Is it intermittent?

Verified / Reproducible

Can you reproduce it reliably?

#api #installation #environment variable #network issue #logging issue

Still need to ship something?

×6

Another batch ranked right after the header list — different links, same matching logic.

Data

Security

Network

Code

UI/UX

Text

System

Multimedia

Protocol

API

Engineering

vllm - ✅(Solved) Fix [Bug]: Performance degradation on MoE models with low batch sizes since vLLM v0.20.0 [1 pull requests, 1 comments, 1 participants]

Recommended Tools

GitHub issue graph ai analysis

Root Cause

Fix Action

Fix / Workaround

============================== CPU Info

PR fix notes

PR #43135: [Perf][gpt-oss] Downgrade triton_kernels to v3.5.1

Description (problem / solution / changelog)

Changed files

Code Example

Your current environment

🐛 Describe the bug

Before submitting a new issue...

Still need to ship something?

TRENDING

vllm - ✅(Solved) Fix [Bug]: Performance degradation on MoE models with low batch sizes since vLLM v0.20.0 [1 pull requests, 1 comments, 1 participants]

Recommended Tools

GitHub issue graph ai analysis

Root Cause

Fix Action

Fix / Workaround

============================== CPU Info

PR fix notes

PR #43135: [Perf][gpt-oss] Downgrade triton_kernels to v3.5.1

Description (problem / solution / changelog)

Changed files

Code Example

Your current environment

🐛 Describe the bug

Before submitting a new issue...

Still need to ship something?

RELATED_DISCOVERY

TRENDING