[Performance]: Compiled `QuantFP8.forward_native` group quantization (1, 128) slower than CUDA on H100/RTX5090

EDIT: These numbers should be collected again with torch==2.10 and torch==2.11 to check if the situation has improved.

Your current environment

Collecting environment information...
uv is set
==============================
        System Info
==============================
OS                           : Ubuntu 24.04.3 LTS (x86_64)
GCC version                  : (Ubuntu 12.4.0-2ubuntu1~24.04) 12.4.0
Clang version                : Could not collect
CMake version                : Could not collect
Libc version                 : glibc-2.39

==============================
       PyTorch Info
==============================
PyTorch version              : 2.8.0+cu128
Is debug build               : False
CUDA used to build PyTorch   : 12.8
ROCM used to build PyTorch   : N/A

==============================
      Python Environment
==============================
Python version               : 3.12.3 (main, Aug 14 2025, 17:47:21) [GCC 13.3.0] (64-bit runtime)
Python platform              : Linux-6.14.0-29-generic-x86_64-with-glibc2.39

==============================
       CUDA / GPU Info
==============================
Is CUDA available            : True
CUDA runtime version         : 12.8.61
CUDA_MODULE_LOADING set to   : LAZY
GPU models and configuration : GPU 0: NVIDIA GeForce RTX 5090
Nvidia driver version        : 580.82.07
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:                           48 bits physical, 48 bits virtual
Byte Order:                              Little Endian
CPU(s):                                  16
On-line CPU(s) list:                     0-15
Vendor ID:                               AuthenticAMD
Model name:                              AMD Ryzen 7 9800X3D 8-Core Processor
CPU family:                              26
Model:                                   68
Thread(s) per core:                      2
Core(s) per socket:                      8
Socket(s):                               1
Stepping:                                0
Frequency boost:                         enabled
CPU(s) scaling MHz:                      67%
CPU max MHz:                             5271.0000
CPU min MHz:                             600.0000
BogoMIPS:                                9399.97
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 amd_lbr_v2 nopl xtopology nonstop_tsc cpuid extd_apicid aperfmperf rapl pni pclmulqdq monitor ssse3 fma cx16 sse4_1 sse4_2 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 hw_pstate ssbd mba perfmon_v2 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 user_shstk avx_vnni avx512_bf16 clzero irperf xsaveerptr rdpru wbnoinvd cppc arat npt lbrv svm_lock nrip_save tsc_scale vmcb_clean flushbyasid decodeassists pausefilter pfthreshold avic v_vmsave_vmload vgif x2avic v_spec_ctrl vnmi avx512vbmi umip pku ospke avx512_vbmi2 gfni vaes vpclmulqdq avx512_vnni avx512_bitalg avx512_vpopcntdq rdpid bus_lock_detect movdiri movdir64b overflow_recov succor smca fsrm avx512_vp2intersect flush_l1d amd_lbr_pmc_freeze
Virtualization:                          AMD-V
L1d cache:                               384 KiB (8 instances)
L1i cache:                               256 KiB (8 instances)
L2 cache:                                8 MiB (8 instances)
L3 cache:                                96 MiB (1 instance)
NUMA node(s):                            1
NUMA node0 CPU(s):                       0-15
Vulnerability Gather data sampling:      Not affected
Vulnerability Ghostwrite:                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:      Mitigation; IBPB on VMEXIT only
Vulnerability Spec store bypass:         Mitigation; Speculative Store Bypass disabled via prctl
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 Tsx async abort:           Not affected

==============================
Versions of relevant libraries
==============================
[pip3] efficientnet-pytorch==0.7.1
[pip3] mypy==1.11.1
[pip3] mypy-extensions==1.0.0
[pip3] numpy==1.26.4
[pip3] nvidia-cublas-cu12==12.8.4.1
[pip3] nvidia-cuda-cupti-cu12==12.8.90
[pip3] nvidia-cuda-nvrtc-cu12==12.8.93
[pip3] nvidia-cuda-runtime-cu12==12.8.90
[pip3] nvidia-cudnn-cu12==9.10.2.21
[pip3] nvidia-cufft-cu12==11.3.3.83
[pip3] nvidia-cufile-cu12==1.13.1.3
[pip3] nvidia-curand-cu12==10.3.9.90
[pip3] nvidia-cusolver-cu12==11.7.3.90
[pip3] nvidia-cusparse-cu12==12.5.8.93
[pip3] nvidia-cusparselt-cu12==0.7.1
[pip3] nvidia-nccl-cu12==2.27.3
[pip3] nvidia-nvjitlink-cu12==12.8.93
[pip3] nvidia-nvtx-cu12==12.8.90
[pip3] open-clip-torch==2.32.0
[pip3] pytorch-lightning==2.5.2
[pip3] pyzmq==27.0.2
[pip3] segmentation-models-pytorch==0.4.0
[pip3] sentence-transformers==3.2.1
[pip3] terratorch==1.1rc3
[pip3] torch==2.8.0+cu128
[pip3] torchaudio==2.8.0+cu128
[pip3] torchgeo==0.7.0
[pip3] torchmetrics==1.7.4
[pip3] torchvision==0.23.0+cu128
[pip3] transformers==4.55.2
[pip3] transformers-stream-generator==0.0.5
[pip3] triton==3.4.0
[pip3] tritonclient==2.51.0
[pip3] vector-quantize-pytorch==1.21.2
[conda] Could not collect

==============================
         vLLM Info
==============================
ROCM Version                 : Could not collect
vLLM Version                 : 0.1.dev9174+gff0855ae0.d20250917 (git sha: ff0855ae0, date: 20250917)
vLLM Build Flags:
  CUDA Archs: Not Set; ROCm: Disabled
GPU Topology:
        GPU0    CPU Affinity    NUMA Affinity   GPU NUMA ID
GPU0     X      0-15    0               N/A

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

==============================
     Environment Variables
==============================
LD_LIBRARY_PATH=/usr/local/cuda-12.8/lib64:
CUDA_HOME=/usr/local/cuda-12.8
CUDA_HOME=/usr/local/cuda-12.8
PYTORCH_NVML_BASED_CUDA_CHECK=1
TORCHINDUCTOR_COMPILE_THREADS=1
CUDA_MODULE_LOADING=LAZY

🐛 Describe the bug

The recently added native torch implementation for group quantization (https://github.com/vllm-project/vllm/pull/24342) exhibits inconsistent performance across different GPUs, particularly for row-major layouts with a group shape of (1, 128). While it's faster (or on par with) CUDA on B200, it's slower on H100 and RTX 5090.

Benchmarks added below.

To reproduce, run:

python3 benchmarks/kernels/bench_per_token_quant_fp8.py --group-sizes 128

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.