Memory issues in nn.Linear due to creation of an empty tensor

[divakar-amd]

🐛 Describe the bug

Two issues:

1. Higher memory consumption of nn.Linear
2. The buffer memory is not shared i.e. the 2 linear layers allocate their own memory.

Reproducible script: basicModel.py
python basicModel.py
PYTORCH_TUNABLEOP_ENABLED=0 python dv_basicModel.py

nn.Linear takes much more memory in mi300 as compared to H100. The cause seems to be the creation of an extra empty tensor which doesn't happen on the cuda (h100) side.

This extra memory allocation eventually adds up and one of the bigger effects is the huge size of cudagraph capture. vllm-project/vllm#19579

Memory Capture:

Memory Capture (with TunableOps=0):

Chrome Trace:

Versions

The output of python collect_env.py

root@banff-cyxtera-s73-5:~# python collect_env.py
Collecting environment information...
PyTorch version: 2.7.0+gitf717b2a
Is debug build: False
CUDA used to build PyTorch: N/A
ROCM used to build PyTorch: 6.4.43483-a187df25c

OS: Ubuntu 22.04.5 LTS (x86_64)
GCC version: (Ubuntu 11.4.0-1ubuntu1~22.04) 11.4.0
Clang version: 19.0.0git (https://github.com/RadeonOpenCompute/llvm-project roc-6.4.1 25184 c87081df219c42dc27c5b6d86c0525bc7d01f727)
CMake version: version 3.31.6
Libc version: glibc-2.35

Python version: 3.12.11 (main, Jun  4 2025, 08:56:18) [GCC 11.4.0] (64-bit runtime)
Python platform: Linux-5.15.0-141-generic-x86_64-with-glibc2.35
Is CUDA available: True
CUDA runtime version: Could not collect
CUDA_MODULE_LOADING set to: LAZY
GPU models and configuration: AMD Instinct MI300X (gfx942:sramecc+:xnack-)
Nvidia driver version: Could not collect
cuDNN version: Could not collect
HIP runtime version: 6.4.43483
MIOpen runtime version: 3.4.0
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):                               224
On-line CPU(s) list:                  0-223
Vendor ID:                            GenuineIntel
Model name:                           Intel(R) Xeon(R) Platinum 8480C
CPU family:                           6
Model:                                143
Thread(s) per core:                   2
Core(s) per socket:                   56
Socket(s):                            2
Stepping:                             8
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 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 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 la57 rdpid bus_lock_detect cldemote movdiri movdir64b enqcmd fsrm md_clear serialize tsxldtrk pconfig arch_lbr amx_bf16 avx512_fp16 amx_tile amx_int8 flush_l1d arch_capabilities
Virtualization:                       VT-x
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 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; RSB filling; PBRSB-eIBRS SW sequence; BHI BHI_DIS_S
Vulnerability Srbds:                  Not affected
Vulnerability Tsx async abort:        Not affected

Versions of relevant libraries:
[pip3] numpy==2.2.6
[pip3] torch==2.7.0+gitf717b2a
[pip3] torchvision==0.21.0+7af6987
[pip3] triton==3.2.0+gite5be006a
[conda] Could not collect

[jeffdaily]

pytorch#156495 should fix this, pending CI and approval.

[divakar-amd]

Issue: Total memory occupied by multiple cudagraphs

The following python script captures multiple cudagraphs while re-using the memory pool from the largest batch-size. It is observed that the total memory occupied by all the cudagraphs is much higher in rocm as compared to cuda. It seems that we are not efficiently re-using the memory pool

https://gist.github.com/divakar-amd/7c7c3ad0fec8e57859385af0093b50f9