[Bug]: gpt-oss-120b fails for ROCm on gfx950 with Triton 3.7.1

[rjrock]

Your current environment

The output of python collect_env.py

Collecting environment information...

    System Info

==============================
OS : Ubuntu 24.04.4 LTS (x86_64)
GCC version : (Ubuntu 13.3.0-6ubuntu2~24.04.1) 13.3.0
Clang version : 23.0.0git (https://github.com/ROCm/llvm-project.git 46fcb339fb61119b337f973c7ca9e710a319fdd0+PATCHED:440716f8b87be9d8e20ed910e10e5b6d14d57cf6)
CMake version : version 3.31.10
Libc version : glibc-2.39

==============================
PyTorch Info

PyTorch version : 2.12.0+rocm7.14.0
Is debug build : False
CUDA used to build PyTorch : N/A
ROCM used to build PyTorch : 7.14.60850
XPU used to build PyTorch : N/A

==============================
Python Environment

Python version : 3.14.6 (main, Jun 23 2026, 15:18:23) [Clang 22.1.3 ] (64-bit runtime)
Python platform : Linux-5.15.0-157-generic-x86_64-with-glibc2.39

==============================
CUDA / GPU Info

Is CUDA available : True
CUDA runtime version : Could not collect
CUDA_MODULE_LOADING set to :
GPU models and configuration : AMD Instinct MI355X (gfx950:sramecc+:xnack-)
Nvidia driver version : Could not collect
cuDNN version : Could not collect
HIP runtime version : 7.14.60850
MIOpen runtime version : 3.5.2
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): 256
On-line CPU(s) list: 0-255
Vendor ID: AuthenticAMD
Model name: AMD EPYC 9575F 64-Core Processor
CPU family: 26
Model: 2
Thread(s) per core: 2
Core(s) per socket: 64
Socket(s): 2
Stepping: 1
Frequency boost: enabled
CPU(s) scaling MHz: 91%
CPU max MHz: 5008.0068
CPU min MHz: 1500.0000
BogoMIPS: 6600.23
Flags: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ht syscall nx mmxext fxsr_opt pdpe1gb rdtscp lm constant_tsc rep_good nopl nonstop_tsc cpuid extd_apicid aperfmperf rapl pni pclmulqdq monitor ssse3 fma cx16 pcid sse4_1 sse4_2 x2apic movbe popcnt aes xsave avx f16c rdrand lahf_lm cmp_legacy svm extapic cr8_legacy abm sse4a misalignsse 3dnowprefetch osvw ibs skinit wdt tce topoext perfctr_core perfctr_nb bpext perfctr_llc mwaitx cpb cat_l3 cdp_l3 invpcid_single hw_pstate ssbd mba ibrs ibpb stibp ibrs_enhanced vmmcall fsgsbase tsc_adjust bmi1 avx2 smep bmi2 erms invpcid cqm rdt_a avx512f avx512dq rdseed adx smap avx512ifma clflushopt clwb avx512cd sha_ni avx512bw avx512vl xsaveopt xsavec xgetbv1 xsaves cqm_llc cqm_occup_llc cqm_mbm_total cqm_mbm_local avx_vnni avx512_bf16 clzero irperf xsaveerptr rdpru wbnoinvd amd_ppin cppc arat npt lbrv svm_lock nrip_save tsc_scale vmcb_clean flushbyasid decodeassists pausefilter pfthreshold avic v_vmsave_vmload vgif v_spec_ctrl avx512vbmi umip pku ospke avx512_vbmi2 gfni vaes vpclmulqdq avx512_vnni avx512_bitalg avx512_vpopcntdq la57 rdpid bus_lock_detect movdiri movdir64b overflow_recov succor smca fsrm avx512_vp2intersect flush_l1d
Virtualization: AMD-V
L1d cache: 6 MiB (128 instances)
L1i cache: 4 MiB (128 instances)
L2 cache: 128 MiB (128 instances)
L3 cache: 512 MiB (16 instances)
NUMA node(s): 2
NUMA node0 CPU(s): 0-63,128-191
NUMA node1 CPU(s): 64-127,192-255
Vulnerability Gather data sampling: Not affected
Vulnerability Indirect target selection: 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 Reg file data sampling: 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 / Automatic IBRS; IBPB conditional; STIBP always-on; PBRSB-eIBRS Not affected; BHI Not affected
Vulnerability Srbds: Not affected
Vulnerability Tsa: Not affected
Vulnerability Tsx async abort: Not affected

==============================
Versions of relevant libraries

[pip3] amd-torch-device-gfx90a==2.12.0+rocm7.14.0
[pip3] amd-torch-device-gfx942==2.12.0+rocm7.14.0
[pip3] amd-torch-device-gfx950==2.12.0+rocm7.14.0
[pip3] amd-torchvision-device-gfx90a==0.27.0+rocm7.14.0
[pip3] amd-torchvision-device-gfx942==0.27.0+rocm7.14.0
[pip3] amd-torchvision-device-gfx950==0.27.0+rocm7.14.0
[pip3] conch-triton-kernels==1.2.1
[pip3] mypy_extensions==1.1.0
[pip3] numpy==2.2.6
[pip3] onnx==1.22.0
[pip3] onnx-ir==0.2.1
[pip3] onnxscript==0.7.1
[pip3] onnxslim==0.1.94
[pip3] open_clip_torch==2.32.0
[pip3] pyzmq==27.1.0
[pip3] segmentation_models_pytorch==0.5.0
[pip3] sentence-transformers==5.3.0
[pip3] torch==2.12.0+rocm7.14.0
[pip3] torchaudio==2.11.0+rocm7.14.0
[pip3] torchcodec==0.10.0a0
[pip3] torchvision==0.27.0+rocm7.14.0
[pip3] transformers==5.13.1
[pip3] transformers-stream-generator==0.0.5
[pip3] triton==3.7.1+git0263a6a6.rocm7.14.0
[pip3] tritonclient==2.66.0
[pip3] vector-quantize-pytorch==1.28.0
[conda] Could not collect

==============================
vLLM Info

ROCM Version : 7.14.60850-0000000
vLLM Version : 0.1.dev19023+g5342131d0 (git sha: 5342131)
vLLM Build Flags:
CUDA Archs: Not Set; ROCm: Enabled; XPU: Disabled
GPU Topology:
============================ ROCm System Management Interface ============================
================================ Weight between two GPUs =================================
GPU0 GPU1 GPU2 GPU3 GPU4 GPU5 GPU6 GPU7
GPU0 0 15 15 15 15 15 15 15
GPU1 15 0 15 15 15 15 15 15
GPU2 15 15 0 15 15 15 15 15
GPU3 15 15 15 0 15 15 15 15
GPU4 15 15 15 15 0 15 15 15
GPU5 15 15 15 15 15 0 15 15
GPU6 15 15 15 15 15 15 0 15
GPU7 15 15 15 15 15 15 15 0

================================= Hops between two GPUs ==================================
GPU0 GPU1 GPU2 GPU3 GPU4 GPU5 GPU6 GPU7
GPU0 0 1 1 1 1 1 1 1
GPU1 1 0 1 1 1 1 1 1
GPU2 1 1 0 1 1 1 1 1
GPU3 1 1 1 0 1 1 1 1
GPU4 1 1 1 1 0 1 1 1
GPU5 1 1 1 1 1 0 1 1
GPU6 1 1 1 1 1 1 0 1
GPU7 1 1 1 1 1 1 1 0

=============================== Link Type between two GPUs ===============================
GPU0 GPU1 GPU2 GPU3 GPU4 GPU5 GPU6 GPU7
GPU0 0 XGMI XGMI XGMI XGMI XGMI XGMI XGMI
GPU1 XGMI 0 XGMI XGMI XGMI XGMI XGMI XGMI
GPU2 XGMI XGMI 0 XGMI XGMI XGMI XGMI XGMI
GPU3 XGMI XGMI XGMI 0 XGMI XGMI XGMI XGMI
GPU4 XGMI XGMI XGMI XGMI 0 XGMI XGMI XGMI
GPU5 XGMI XGMI XGMI XGMI XGMI 0 XGMI XGMI
GPU6 XGMI XGMI XGMI XGMI XGMI XGMI 0 XGMI
GPU7 XGMI XGMI XGMI XGMI XGMI XGMI XGMI 0

======================================= Numa Nodes =======================================
GPU[0] : (Topology) Numa Node: 0
GPU[0] : (Topology) Numa Affinity: 0
GPU[1] : (Topology) Numa Node: 0
GPU[1] : (Topology) Numa Affinity: 0
GPU[2] : (Topology) Numa Node: 0
GPU[2] : (Topology) Numa Affinity: 0
GPU[3] : (Topology) Numa Node: 0
GPU[3] : (Topology) Numa Affinity: 0
GPU[4] : (Topology) Numa Node: 1
GPU[4] : (Topology) Numa Affinity: 1
GPU[5] : (Topology) Numa Node: 1
GPU[5] : (Topology) Numa Affinity: 1
GPU[6] : (Topology) Numa Node: 1
GPU[6] : (Topology) Numa Affinity: 1
GPU[7] : (Topology) Numa Node: 1
GPU[7] : (Topology) Numa Affinity: 1
================================== End of ROCm SMI Log ===================================

==============================
Environment Variables

PYTORCH_ROCM_ARCH=gfx90a;gfx942;gfx950
MAX_JOBS=64
LD_LIBRARY_PATH=/opt/python/lib/python3.14/site-packages/_rocm_sdk_devel/lib:/opt/python/lib/python3.14/site-packages/_rocm_sdk_devel/lib/rocm_sysdeps/lib:/opt/python/lib/python3.14/site-packages/_rocm_sdk_core/lib:/opt/python/lib/python3.14/site-packages/_rocm_sdk_core/lib/rocm_sysdeps/lib
PYTORCH_NVML_BASED_CUDA_CHECK=1
TORCHINDUCTOR_COMPILE_THREADS=1
TORCHINDUCTOR_CACHE_DIR=/tmp/torchinductor_root

🐛 Describe the bug

Running:

vllm serve openai/gpt-oss-120b \
  --enable-expert-parallel \
  --tensor-parallel-size 4

on gfx950 with Triton 3.7.1 fails with:

triton.runtime.errors.OutOfResources: out of resource: shared memory,
Required: 165888, Hardware limit: 163840.
Reducing block sizes or `num_stages` may help.

This configuration works with Triton 3.6 but fails on 3.7.1 due to exceeding the available shared memory on gfx950.

ROCm will soon update to 7.14, which ships with Triton 3.7.1.

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.

[github-actions]

CC @hongxiayang @tjtanaa @vllmellm @giuseppegrossi for ROCm-related issue

[github-actions]

Hi @rjrock, thanks for reporting this ROCm issue!

To help us investigate, please make sure the following information is included:

• A minimal reproducer (code snippet or script that triggers the issue)
• The full error message or traceback
• How you installed vLLM (Docker image name, pip install, or build from source steps)
• The command you used to launch vLLM (e.g., vllm serve ... or the Python script)
• Your GPU model and GFX architecture (e.g., MI300X / gfx942) — run rocminfo | grep gfx

Please provide any unchecked items above. This will help us reproduce and resolve the issue faster. Thank you!