Version
0.3.0 - 0.9.0
Describe the bug.
Updating from 0.2.0 to 0.3.0 results in a significant performance regression for large, tiled, JP2 images. Removing the cudaStreamWaitEvents around the tile decode step and restoring the looped stream wait event processing appears to resolve this performance hit. It appears that every release after 0.3.0 suffers from the same performance regression.
In order to observe a significant latency hit the Jp2 image must be relatively large and contain a large number of tiles. I have provided a code snippet demonstrating how to convert the Hurricane Alex Geotiff to a tiled JP2 file that meet both of these criteria. This new JP2 file is utilized in the minimum reproducible example.
conda install -c conda-forge --yes glymur openjpeg python=3.12
tiff2jp2 --tilesize 256 256 'Alex_2016-01-14_1300Z_(Geotiff).tif' Alex_2016-01-14_1300Z-256-tile.jp2
Running a decode operation on this sample image results in a 7x slowdown between the 0.2.0 and the 0.3.0 release. My team and I would be interested in a solution involving an asynchronous option for decoding tiles on the GPU to restore the performance of the 0.2.0 release.
Thanks!
Minimum reproducible example
from timeit import timeit
from nvidia import nvimgcodec
with open("Alex_2016-01-14_1300Z-256-tile.jp2", "rb") as f: dataset_bytes = f.read()
decode_params = nvimgcodec.DecodeParams(color_spec=nvimgcodec.ColorSpec.UNCHANGED, allow_any_depth=True, apply_exif_orientation=False,)
decoder = nvimgcodec.Decoder(backends=[nvimgcodec.Backend(nvimgcodec.GPU_ONLY)], max_num_cpu_threads=12, device_id=0,)
print(timeit("decoder.decode(dataset_bytes, params = decode_params)", globals=globals(), number=10))Environment details
<details><summary>Click here to see environment details</summary><pre>
**git***
Not inside a git repository
***OS Information***
DISTRIB_ID=Ubuntu
DISTRIB_RELEASE=24.04
DISTRIB_CODENAME=noble
DISTRIB_DESCRIPTION="Ubuntu 24.04.4 LTS"
PRETTY_NAME="Ubuntu 24.04.4 LTS"
NAME="Ubuntu"
VERSION_ID="24.04"
VERSION="24.04.4 LTS (Noble Numbat)"
VERSION_CODENAME=noble
ID=ubuntu
ID_LIKE=debian
HOME_URL="https://www.ubuntu.com/"
SUPPORT_URL="https://help.ubuntu.com/"
BUG_REPORT_URL="https://bugs.launchpad.net/ubuntu/"
PRIVACY_POLICY_URL="https://www.ubuntu.com/legal/terms-and-policies/privacy-policy"
UBUNTU_CODENAME=noble
LOGO=ubuntu-logo
Linux ip-10-19-193-27.us-gov-west-1.compute.internal 5.10.192-183.736.amzn2.x86_64 #1 SMP Wed Sep 6 21:15:41 UTC 2023 x86_64 x86_64 x86_64 GNU/Linux
***GPU Information***
Thu Jul 2 19:07:53 2026
+---------------------------------------------------------------------------------------+
| NVIDIA-SMI 535.104.12 Driver Version: 535.104.12 CUDA Version: 12.2 |
|-----------------------------------------+----------------------+----------------------+
| GPU Name Persistence-M | Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap | Memory-Usage | GPU-Util Compute M. |
| | | MIG M. |
|=========================================+======================+======================|
| 0 Tesla T4 On | 00000000:00:1E.0 Off | 0 |
| N/A 30C P8 14W / 70W | 0MiB / 15360MiB | 0% Default |
| | | N/A |
+-----------------------------------------+----------------------+----------------------+
+---------------------------------------------------------------------------------------+
| Processes: |
| GPU GI CI PID Type Process name GPU Memory |
| ID ID Usage |
|=======================================================================================|
| No running processes found |
+---------------------------------------------------------------------------------------+
***CPU***
Architecture: x86_64
CPU op-mode(s): 32-bit, 64-bit
Address sizes: 46 bits physical, 48 bits virtual
Byte Order: Little Endian
CPU(s): 16
On-line CPU(s) list: 0-15
Vendor ID: GenuineIntel
Model name: Intel(R) Xeon(R) Platinum 8259CL CPU @ 2.50GHz
CPU family: 6
Model: 85
Thread(s) per core: 2
Core(s) per socket: 8
Socket(s): 1
Stepping: 7
BogoMIPS: 4999.99
Flags: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ss ht syscall nx pdpe1gb rdtscp lm constant_tsc rep_good nopl xtopology nonstop_tsc cpuid aperfmperf tsc_known_freq pni pclmulqdq ssse3 fma cx16 pcid sse4_1 sse4_2 x2apic movbe popcnt tsc_deadline_timer aes xsave avx f16c rdrand hypervisor lahf_lm abm 3dnowprefetch invpcid_single pti fsgsbase tsc_adjust bmi1 avx2 smep bmi2 erms invpcid mpx avx512f avx512dq rdseed adx smap clflushopt clwb avx512cd avx512bw avx512vl xsaveopt xsavec xgetbv1 xsaves ida arat pku ospke avx512_vnni
Hypervisor vendor: KVM
Virtualization type: full
L1d cache: 256 KiB (8 instances)
L1i cache: 256 KiB (8 instances)
L2 cache: 8 MiB (8 instances)
L3 cache: 35.8 MiB (1 instance)
NUMA node(s): 1
NUMA node0 CPU(s): 0-15
Vulnerability Gather data sampling: Unknown: Dependent on hypervisor status
Vulnerability Itlb multihit: KVM: Mitigation: VMX unsupported
Vulnerability L1tf: Mitigation; PTE Inversion
Vulnerability Mds: Vulnerable: Clear CPU buffers attempted, no microcode; SMT Host state unknown
Vulnerability Meltdown: Mitigation; PTI
Vulnerability Mmio stale data: Vulnerable: Clear CPU buffers attempted, no microcode; SMT Host state unknown
Vulnerability Retbleed: Vulnerable
Vulnerability Spec rstack overflow: Not affected
Vulnerability Spec store bypass: Vulnerable
Vulnerability Spectre v1: Mitigation; usercopy/swapgs barriers and __user pointer sanitization
Vulnerability Spectre v2: Mitigation; Retpolines, STIBP disabled, RSB filling
Vulnerability Srbds: Not affected
Vulnerability Tsx async abort: Not affected
***CMake***
***g++***
***nvcc***
***Python***
/opt/conda/bin/python
Python 3.12.13
***Environment Variables***
PATH : /opt/conda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin
LD_LIBRARY_PATH :
NUMBAPRO_NVVM :
NUMBAPRO_LIBDEVICE :
CONDA_PREFIX :
PYTHON_PATH :
***conda packages***
/opt/conda/bin/conda
# packages in environment at /opt/conda:
#
# Name Version Build Channel
_openmp_mutex 4.5 20_gnu conda-forge
_python_abi3_support 1.0 hd8ed1ab_2 conda-forge
annotated-types 0.7.0 pyhd8ed1ab_1 conda-forge
anyio 4.14.1 pyhcf101f3_0 conda-forge
archspec 0.2.5 pyhd8ed1ab_0 conda-forge
backports.zstd 1.6.0 py312h90b7ffd_0 conda-forge
boltons 25.0.0 pyhd8ed1ab_0 conda-forge
brotli-python 1.2.0 py312hdb49522_1 conda-forge
bzip2 1.0.8 hda65f42_9 conda-forge
c-ares 1.34.6 hb03c661_0 conda-forge
ca-certificates 2026.6.17 hbd8a1cb_0 conda-forge
cached-property 1.5.2 hd8ed1ab_1 conda-forge
cached_property 1.5.2 pyha770c72_1 conda-forge
certifi 2026.6.17 pyhd8ed1ab_0 conda-forge
cffi 2.0.0 py312h460c074_1 conda-forge
charset-normalizer 3.4.7 pyhd8ed1ab_0 conda-forge
colorama 0.4.6 pyhd8ed1ab_1 conda-forge
conda 26.5.3 py312h20c3967_2 conda-forge
conda-index 0.12.1 pyhd8ed1ab_0 conda-forge
conda-libmamba-solver 26.4.2 pyhd8ed1ab_0 conda-forge
conda-lockfiles 0.2.0 pyhd8ed1ab_0 conda-forge
conda-package-handling 2.4.0 pyh7900ff3_2 conda-forge
conda-package-streaming 0.12.0 pyhd8ed1ab_0 conda-forge
conda-pypi 0.10.1 py312h20c3967_2 conda-forge
conda-rattler-solver 0.1.1 pyhcf101f3_0 conda-forge
conda-self 0.2.0 pyhd8ed1ab_0 conda-forge
cpp-expected 1.3.1 h171cf75_0 conda-forge
cpython 3.12.13 py312hd8ed1ab_0 conda-forge
distro 1.9.0 pyhd8ed1ab_1 conda-forge
exceptiongroup 1.3.1 pyhd8ed1ab_0 conda-forge
filelock 3.29.4 pyhd8ed1ab_0 conda-forge
fmt 12.1.0 hff5e90c_0 conda-forge
frozendict 2.4.7 pyh851646a_2 conda-forge
glymur 0.14.8 pyhd8ed1ab_0 conda-forge
h11 0.16.0 pyhcf101f3_1 conda-forge
h2 4.3.0 pyhcf101f3_0 conda-forge
hpack 4.1.0 pyhd8ed1ab_0 conda-forge
httpcore 1.0.9 pyh29332c3_0 conda-forge
httpx 0.28.1 pyhd8ed1ab_0 conda-forge
hyperframe 6.1.0 pyhd8ed1ab_0 conda-forge
icu 78.3 h33c6efd_0 conda-forge
idna 3.17 pyhcf101f3_0 conda-forge
imageio 2.37.0 pyhfb79c49_0 conda-forge
importlib-metadata 9.0.0 pyhcf101f3_0 conda-forge
jinja2 3.1.6 pyhcf101f3_1 conda-forge
jsonpatch 1.33 pyhd8ed1ab_1 conda-forge
jsonpointer 3.1.1 pyhcf101f3_0 conda-forge
keyutils 1.6.3 hb9d3cd8_0 conda-forge
krb5 1.22.2 ha1258a1_0 conda-forge
lcms2 2.19.1 h0c24ade_1 conda-forge
ld_impl_linux-64 2.45.1 default_hbd61a6d_102 conda-forge
lerc 4.1.0 hdb68285_0 conda-forge
libarchive 3.8.7 gpl_hc2c16d8_101 conda-forge
libblas 3.11.0 8_h4a7cf45_openblas conda-forge
libcblas 3.11.0 8_h0358290_openblas conda-forge
libcurl 8.20.0 hcf29cc6_0 conda-forge
libdeflate 1.25 h17f619e_0 conda-forge
libedit 3.1.20250104 pl5321h7949ede_0 conda-forge
libev 4.33 hd590300_2 conda-forge
libexpat 2.8.1 hecca717_0 conda-forge
libffi 3.5.2 h3435931_0 conda-forge
libfreetype 2.14.3 ha770c72_0 conda-forge
libfreetype6 2.14.3 h73754d4_0 conda-forge
libgcc 15.2.0 he0feb66_19 conda-forge
libgcc-ng 15.2.0 h69a702a_19 conda-forge
libgfortran 15.2.0 h69a702a_19 conda-forge
libgfortran5 15.2.0 h68bc16d_19 conda-forge
libgomp 15.2.0 he0feb66_19 conda-forge
libiconv 1.18 h3b78370_2 conda-forge
libjpeg-turbo 3.1.4.1 hb03c661_0 conda-forge
liblapack 3.11.0 8_h47877c9_openblas conda-forge
liblzma 5.8.3 hb03c661_0 conda-forge
libmamba 2.5.0 hd28c85e_0 conda-forge
libmamba-spdlog 2.5.0 h12fcf84_0 conda-forge
libmambapy 2.5.0 py312h1ca65c7_0 conda-forge
libnghttp2 1.68.1 h877daf1_0 conda-forge
libnsl 2.0.1 hb9d3cd8_1 conda-forge
libopenblas 0.3.33 pthreads_h94d23a6_0 conda-forge
libpng 1.6.58 h421ea60_0 conda-forge
libsolv 0.7.39 h9463b59_0 conda-forge
libsqlite 3.53.1 h0c1763c_0 conda-forge
libssh2 1.11.1 hcf80075_0 conda-forge
libstdcxx 15.2.0 h934c35e_19 conda-forge
libtiff 4.7.1 h9d88235_1 conda-forge
libuuid 2.42.1 h5347b49_0 conda-forge
libwebp-base 1.6.0 hd42ef1d_0 conda-forge
libxcb 1.17.0 h8a09558_0 conda-forge
libxcrypt 4.4.36 hd590300_1 conda-forge
libxml2 2.15.3 h49c6c72_0 conda-forge
libxml2-16 2.15.3 hca6bf5a_0 conda-forge
libxslt 1.1.43 h711ed8c_1 conda-forge
libzlib 1.3.2 h25fd6f3_2 conda-forge
lxml 6.1.1 py312h63ddcf0_0 conda-forge
lz4-c 1.10.0 h5888daf_1 conda-forge
lzo 2.10 h280c20c_1002 conda-forge
mamba 2.5.0 h9835478_0 conda-forge
markupsafe 3.0.3 py312h8a5da7c_1 conda-forge
menuinst 2.5.1 py312h20c3967_0 conda-forge
msgpack-python 1.2.1 py312h0a2e395_1 conda-forge
ncurses 6.6 hdb14827_0 conda-forge
nlohmann_json-abi 3.12.0 h0f90c79_1 conda-forge
numpy 2.5.0 py312h33ff503_0 conda-forge
nvidia-nvimgcodec-cu12 0.2.0.7 pypi_0 pypi
nvidia-nvjpeg-cu12 12.3.3.54 pypi_0 pypi
openjpeg 2.5.4 h55fea9a_0 conda-forge
openssl 3.6.3 h35e630c_0 conda-forge
packaging 26.2 pyhc364b38_0 conda-forge
pillow 12.3.0 py312h50c33e8_0 conda-forge
pip 26.1.2 pyh8b19718_0 conda-forge
platformdirs 4.10.0 pyhcf101f3_0 conda-forge
pluggy 1.6.0 pyhf9edf01_1 conda-forge
pthread-stubs 0.4 hb9d3cd8_1002 conda-forge
py-rattler 0.25.0 py310h70157a2_1 conda-forge
pybind11-abi 11 hc364b38_1 conda-forge
pycosat 0.6.6 py312h4c3975b_3 conda-forge
pycparser 3.0 pyhcf101f3_0 conda-forge
pydantic 2.13.4 pyhcf101f3_0 conda-forge
pydantic-core 2.46.4 py312h868fb18_0 conda-forge
pyproject_hooks 1.2.0 pyhd8ed1ab_1 conda-forge
pysocks 1.7.1 pyha55dd90_7 conda-forge
python 3.12.13 hd63d673_0_cpython conda-forge
python-build 1.5.0 pyhc364b38_0 conda-forge
python-gil 3.12.13 hd8ed1ab_0 conda-forge
python-installer 1.0.1 pyh332efcf_0 conda-forge
python_abi 3.12 8_cp312 conda-forge
readline 8.3 h853b02a_0 conda-forge
reproc 14.2.7.post0 hb03c661_1 conda-forge
reproc-cpp 14.2.7.post0 hecca717_1 conda-forge
requests 2.34.2 pyhcf101f3_0 conda-forge
ruamel.yaml 0.18.17 py312h5253ce2_2 conda-forge
ruamel.yaml.clib 0.2.15 py312h5253ce2_1 conda-forge
setuptools 82.0.1 pyh332efcf_0 conda-forge
simdjson 4.2.4 hb700be7_0 conda-forge
sniffio 1.3.1 pyhd8ed1ab_2 conda-forge
spdlog 1.17.0 hab81395_1 conda-forge
tk 8.6.13 noxft_h366c992_103 conda-forge
tomli 2.4.1 pyhcf101f3_0 conda-forge
tomli-w 1.2.0 pyhd8ed1ab_0 conda-forge
tqdm 4.67.3 pyh8f84b5b_0 conda-forge
truststore 0.10.4 pyhcf101f3_0 conda-forge
typing-extensions 4.16.0 h69aa097_0 conda-forge
typing-inspection 0.4.2 pyhcf101f3_2 conda-forge
typing_extensions 4.16.0 pyhcf101f3_0 conda-forge
tzdata 2025c hc9c84f9_1 conda-forge
unearth 0.18.2 pyhd8ed1ab_0 conda-forge
urllib3 2.7.0 pyhd8ed1ab_0 conda-forge
wheel 0.47.0 pyhd8ed1ab_0 conda-forge
xorg-libxau 1.0.12 hb03c661_1 conda-forge
xorg-libxdmcp 1.1.5 hb03c661_1 conda-forge
yaml-cpp 0.8.0 h3f2d84a_0 conda-forge
zipp 4.1.0 pyhcf101f3_0 conda-forge
zlib-ng 2.3.3 hceb46e0_1 conda-forge
zstandard 0.25.0 py312h5253ce2_1 conda-forge
zstd 1.5.7 hb78ec9c_6 conda-forge
</pre></details>Relevant log output
Other/Misc.
NOTE: The provided environment details pertain to the 0.2.0 release. Please rerun the minimum reproducible example after installing the 0.3.0 release.
pip install nvidia-nvjpeg2k-cu12==0.8.0.38
pip install nvidia-nvimgcodec-cu12==0.3.0.5
NOTE: All of the provided code was run inside the docker image condaforge/miniforge3:26.3.2-3.
Check for duplicates
Version
0.3.0 - 0.9.0
Describe the bug.
Updating from 0.2.0 to 0.3.0 results in a significant performance regression for large, tiled, JP2 images. Removing the cudaStreamWaitEvents around the tile decode step and restoring the looped stream wait event processing appears to resolve this performance hit. It appears that every release after 0.3.0 suffers from the same performance regression.
In order to observe a significant latency hit the Jp2 image must be relatively large and contain a large number of tiles. I have provided a code snippet demonstrating how to convert the Hurricane Alex Geotiff to a tiled JP2 file that meet both of these criteria. This new JP2 file is utilized in the minimum reproducible example.
Running a decode operation on this sample image results in a 7x slowdown between the 0.2.0 and the 0.3.0 release. My team and I would be interested in a solution involving an asynchronous option for decoding tiles on the GPU to restore the performance of the 0.2.0 release.
Thanks!
Minimum reproducible example
Environment details
Relevant log output
Other/Misc.
NOTE: The provided environment details pertain to the 0.2.0 release. Please rerun the minimum reproducible example after installing the 0.3.0 release.
NOTE: All of the provided code was run inside the docker image
condaforge/miniforge3:26.3.2-3.Check for duplicates