[release/2.9] llama_v2_7b_16h amp inference performance got Segmentation fault

[mengfei25]

🐛 Describe the bug

llama_v2_7b_16h got segmentation fault with release/2.9, but passed with release/2.8

xpu  eval  llama_v2_7b_16h
Segmentation fault from GPU at 0xff00000688c74000, ctx_id: 1 (CCS) type: 0 (NotPresent), level: 1 (PDE), access: 1 (Write), banned: 1, aborting.
Segmentation fault from GPU at 0xff00000688c74000, ctx_id: 1 (CCS) type: 0 (NotPresent), level: 1 (PDE), access: 1 (Write), banned: 1, aborting.
Abort was called at 288 line in file:
./shared/source/os_interface/linux/drm_neo.cpp
Run failed with return code:  -6
Output:  None
Error:  None

Reproducer

python benchmarks/dynamo/torchbench.py --performance --amp --amp-dtype float16 -d xpu -n10 --inference  --only llama_v2_7b_16h --cold-start-latency --backend=inductor
python benchmarks/dynamo/torchbench.py --performance --amp --amp-dtype bfloat16 -d xpu -n10 --inference  --only llama_v2_7b_16h --cold-start-latency --backend=inductor

Versions

Collecting environment information...
PyTorch version: 2.9.0+xpu
Is debug build: False
CUDA used to build PyTorch: None
ROCM used to build PyTorch: N/A
OS: Ubuntu 22.04.5 LTS (x86_64)
GCC version: (Ubuntu 11.4.0-1ubuntu1~22.04.2) 11.4.0
Clang version: Could not collect
CMake version: version 3.22.1
Libc version: glibc-2.35
Python version: 3.10.18 (main, Jun 4 2025, 04:05:45) [GCC 11.4.0] (64-bit runtime)
Python platform: Linux-5.15.0-153-generic-x86_64-with-glibc2.35
Is CUDA available: False
CUDA runtime version: No CUDA
CUDA_MODULE_LOADING set to: N/A
GPU models and configuration: No CUDA
Nvidia driver version: No CUDA
cuDNN version: No CUDA
Is XPU available: True
XPU used to build PyTorch: 20250201
Intel GPU driver version:

• intel-opencl-icd: 25.18.33578.38-1146~22.04
• libze1: 1.21.9.0-1136~22.04
  Intel GPU models onboard:
  N/A
  Intel GPU models detected:
• [0] _XpuDeviceProperties(name='Intel(R) Data Center GPU Max 1100', platform_name='Intel(R) oneAPI Unified Runtime over Level-Zero', type='gpu', device_id=0xBDA, uuid=8680da0b-2f00-0000-0f00-000000000000, driver_version='1.6.33578+38', total_memory=49136MB, max_compute_units=448, gpu_eu_count=448, gpu_subslice_count=56, max_work_group_size=1024, max_num_sub_groups=64, sub_group_sizes=[16 32], has_fp16=1, has_fp64=1, has_atomic64=1)
• [1] _XpuDeviceProperties(name='Intel(R) Data Center GPU Max 1100', platform_name='Intel(R) oneAPI Unified Runtime over Level-Zero', type='gpu', device_id=0xBDA, uuid=8680da0b-2f00-0000-1600-000000000000, driver_version='1.6.33578+38', total_memory=49136MB, max_compute_units=448, gpu_eu_count=448, gpu_subslice_count=56, max_work_group_size=1024, max_num_sub_groups=64, sub_group_sizes=[16 32], has_fp16=1, has_fp64=1, has_atomic64=1)
• [2] _XpuDeviceProperties(name='Intel(R) Data Center GPU Max 1100', platform_name='Intel(R) oneAPI Unified Runtime over Level-Zero', type='gpu', device_id=0xBDA, uuid=8680da0b-2f00-0000-1a00-000000000000, driver_version='1.6.33578+38', total_memory=49136MB, max_compute_units=448, gpu_eu_count=448, gpu_subslice_count=56, max_work_group_size=1024, max_num_sub_groups=64, sub_group_sizes=[16 32], has_fp16=1, has_fp64=1, has_atomic64=1)
• [3] _XpuDeviceProperties(name='Intel(R) Data Center GPU Max 1100', platform_name='Intel(R) oneAPI Unified Runtime over Level-Zero', type='gpu', device_id=0xBDA, uuid=8680da0b-2f00-0000-1e00-000000000000, driver_version='1.6.33578+38', total_memory=49136MB, max_compute_units=448, gpu_eu_count=448, gpu_subslice_count=56, max_work_group_size=1024, max_num_sub_groups=64, sub_group_sizes=[16 32], has_fp16=1, has_fp64=1, has_atomic64=1)
• [4] _XpuDeviceProperties(name='Intel(R) Data Center GPU Max 1100', platform_name='Intel(R) oneAPI Unified Runtime over Level-Zero', type='gpu', device_id=0xBDA, uuid=8680da0b-2f00-0000-af00-000000000000, driver_version='1.6.33578+38', total_memory=49136MB, max_compute_units=448, gpu_eu_count=448, gpu_subslice_count=56, max_work_group_size=1024, max_num_sub_groups=64, sub_group_sizes=[16 32], has_fp16=1, has_fp64=1, has_atomic64=1)
• [5] _XpuDeviceProperties(name='Intel(R) Data Center GPU Max 1100', platform_name='Intel(R) oneAPI Unified Runtime over Level-Zero', type='gpu', device_id=0xBDA, uuid=8680da0b-2f00-0000-b300-000000000000, driver_version='1.6.33578+38', total_memory=49136MB, max_compute_units=448, gpu_eu_count=448, gpu_subslice_count=56, max_work_group_size=1024, max_num_sub_groups=64, sub_group_sizes=[16 32], has_fp16=1, has_fp64=1, has_atomic64=1)
• [6] _XpuDeviceProperties(name='Intel(R) Data Center GPU Max 1100', platform_name='Intel(R) oneAPI Unified Runtime over Level-Zero', type='gpu', device_id=0xBDA, uuid=8680da0b-2f00-0000-bf00-000000000000, driver_version='1.6.33578+38', total_memory=49136MB, max_compute_units=448, gpu_eu_count=448, gpu_subslice_count=56, max_work_group_size=1024, max_num_sub_groups=64, sub_group_sizes=[16 32], has_fp16=1, has_fp64=1, has_atomic64=1)
• [7] _XpuDeviceProperties(name='Intel(R) Data Center GPU Max 1100', platform_name='Intel(R) oneAPI Unified Runtime over Level-Zero', type='gpu', device_id=0xBDA, uuid=8680da0b-2f00-0000-c300-000000000000, driver_version='1.6.33578+38', total_memory=49136MB, max_compute_units=448, gpu_eu_count=448, gpu_subslice_count=56, max_work_group_size=1024, max_num_sub_groups=64, sub_group_sizes=[16 32], has_fp16=1, has_fp64=1, has_atomic64=1)
  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, 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 8468V
  CPU family: 6
  Model: 143
  Thread(s) per core: 2
  Core(s) per socket: 48
  Socket(s): 2
  Stepping: 8
  Frequency boost: enabled
  CPU max MHz: 2401.0000
  CPU min MHz: 800.0000
  BogoMIPS: 4800.00
  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 intel_ppin 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 split_lock_detect avx_vnni avx512_bf16 wbnoinvd dtherm ida arat pln pts avx512vbmi umip pku ospke waitpkg avx512_vbmi2 gfni vaes v
  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 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 disabled; PBRSB-eIBRS SW sequence; BHI BHI_DIS_S
  Vulnerability Srbds: Not affected
  Vulnerability Tsx async abort: Not affected
  Versions of relevant libraries:
  [pip3] dpcpp-cpp-rt==2025.2.1
  [pip3] impi-rt==2021.16.1
  [pip3] intel-cmplr-lib-rt==2025.2.1
  [pip3] intel-cmplr-lib-ur==2025.2.1
  [pip3] intel-cmplr-lic-rt==2025.2.1
  [pip3] intel-opencl-rt==2025.2.1
  [pip3] intel-openmp==2025.2.1
  [pip3] intel-pti==0.13.1
  [pip3] intel-sycl-rt==2025.2.1
  [pip3] mkl==2025.2.0
  [pip3] numpy==2.1.2
  [pip3] oneccl==2021.16.1
  [pip3] oneccl-devel==2021.16.1
  [pip3] onemkl-sycl-blas==2025.2.0
  [pip3] onemkl-sycl-dft==2025.2.0
  [pip3] onemkl-sycl-lapack==2025.2.0
  [pip3] onemkl-sycl-rng==2025.2.0
  [pip3] onemkl-sycl-sparse==2025.2.0
  [pip3] pytorch-triton-xpu==3.5.0
  [pip3] tbb==2022.2.0
  [pip3] tcmlib==1.4.0
  [pip3] torch==2.9.0+xpu
  [pip3] torchao==0.13.0+xpu
  [pip3] torchaudio==2.9.0+xpu
  [pip3] torchvision==0.24.0+xpu
  [pip3] umf==0.11.0
  [conda] Could not collect

[jianyizh]

@mengfei25 it fails after torchbench turns of kv cache, can pytorch 2.8 pass with kc cache off? I will keep investigating this issue. pytorch/benchmark@9ebf2e2#diff-5460c7b984164250a2fb3ea078e0501855a8ec63220ec2bd0543dbeb01e37ec4R88 self.model.config.use_cache = False

[jianyizh]

@chuanqi129 do you think current oob/dynamo performance test should turn kv cache on? we can also apply pytorch/benchmark#2587 to load this model as fp16, which uses less memory and model will pass

[jianyizh]

The seg fault is actually trigged by OOM issue, we may again allocate a host pointer on GPU. It can pass on 1550 because of it's larger memory. It can also pass if use cache=True or load as fp16.
About OOM issues in torchbenchmark, here's something we can do:

1. merge the torchbench pr
2. avoid deep copy. We can use deep_copy(model_cpu).to("xpu") instead of always keep the original model on xpu. I can pass on 1100 if I remove deep copy in warmpup https://github.com/pytorch/pytorch/blob/release/2.9/benchmarks/dynamo/common.py#L2700
3. cuda has a feature to set PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation to save some memory, we can have similar settings. But this may not solve OOM issues for most OOB cases. https://github.com/intel/llvm/blob/5a8fa3e97bcc1902d97af1efab950f05b1d432b6/sycl/doc/extensions/experimental/sycl_ext_oneapi_virtual_mem.asciidoc#L69