請問使用lada 5060TI16G 升級到5070ti會有感嗎

1970134465

用LADA和JASNA都能正常用GPU来破解

hodalahiruma

1970134465 发表于 2026-6-9 13:26
用LADA和JASNA都能正常用GPU来破解

你的问题与 Jasna 作者遇到的问题一样
也是 RTX 5090 无法使用 GPU 全速
https://codeberg.org/comman/lada-ex/issues/24

这问题我也不清楚
在我的PC环境，4060 与 5060 Ti 都能有 90% 以上的负载
我会帮你关注
等待LadaEx 制作群与 Jasna 作者的反馈

1970134465

感谢大佬的指点

1970134465

hodalahiruma 发表于 2026-6-9 13:54
你的问题与 Jasna 作者遇到的问题一样
也是 RTX 5090 无法使用 GPU 全速
https://codeberg.org/comman/lad ...

大佬，我换了4080S 来跑，还是出错，求指点
"python\python.exe" -m lada.cli.main --input "D:\Favorites\*\*.mp4" --output "D:\116" --temporary-directory "D:\116" --output-file-pattern "{orig_file_name}-U.mp4" --mosaic-detection-model "v4-fast" --encoding-preset "hevc-nvidia-gpu-uhq" --max-clip-length 180 --no-detect-face-mosaics --fp16 --tensorrt --torchcodec-encoder --fast-pipeline --device cuda:0

[06/10/2026-18:52:58] [TRT] [E] IRuntime::deserializeCudaEngine: Error Code 6: API Usage Error (The engine plan file is generated on an incompatible device, expecting compute 8.9 got compute 12.0, please rebuild. In nvinfer1::rt::`anonymous-namespace'::checkEngineDeviceCompatible at C:\_src\runtime\api\engine.cpp:918)
Traceback (most recent call last):
  File "<frozen runpy>", line 198, in _run_module_as_main
  File "<frozen runpy>", line 88, in _run_code
  File "D:\lada-ex\python\Lib\site-packages\lada\cli\main.py", line 516, in <module>
    main()
    ~~~~^^
  File "D:\lada-ex\python\Lib\site-packages\lada\cli\main.py", line 449, in main
    mosaic_detection_model, mosaic_restoration_model, preferred_pad_mode = load_models(
                                                                           ~~~~~~~~~~~^
        device, mosaic_restoration_model_name, mosaic_restoration_model_path, args.mosaic_restoration_config_path,
        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
    ...<4 lines>...
        fast_pipeline=args.fast_pipeline
        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
    )
    ^
  File "D:\lada-ex\python\Lib\site-packages\lada\restorationpipeline\__init__.py", line 116, in load_models
    mosaic_detection_model = Yolo11SegmentationModel(
        mosaic_detection_model_path, device, classes=classes, conf=0.15,
        fp16=fp16, vr_mode=vr_mode, imgsz=imgsz, use_tensorrt=use_tensorrt, batch_size=4
    )
  File "D:\lada-ex\python\Lib\site-packages\lada\models\yolo\yolo11_segmentation_model.py", line 61, in __init__
    self._trt_predictor = Yolo11TensorRTPredictor(engine_path, self.device, self.imgsz, self._conf, self.fp16, self._tracker_yaml)
                          ~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\lada-ex\python\Lib\site-packages\lada\models\yolo\yolo_tensorrt_predictor.py", line 86, in __init__
    in_name = engine.get_tensor_name(0)
              ^^^^^^^^^^^^^^^^^^^^^^
AttributeError: 'NoneType' object has no attribute 'get_tensor_name'
[06/10/2026-18:53:03] [TRT] [W] Functionality provided through tensorrt.plugin module is experimental.
W0610 18:53:04.172000 7552 Lib\site-packages\torch\distributed\elastic\multiprocessing\redirects.py:29] NOTE: Redirects are currently not supported in Windows or MacOs.

hodalahiruma

1970134465 发表于 2026-6-10 18:51
大佬，我换了4080S 来跑，还是出错，求指点
"python\python.exe" -m lada.cli.main --input "D:\Favorites ...

因你先前是使用 5090 且使用了 --tensorrt 参数
这会预先编译好 BasicVSR++ 引擎使编码更快
但现在使用了 4080，架构上不同
因此先前的引擎不适用
移除:
model_weights\lada_mosaic_detection_model_v4_fast.onnx
model_weights\lada_mosaic_detection_model_v4_fast.bs4.imgsz640.fp16.v11_total_manual.engine.lock
model_weights\lada_mosaic_detection_model_v4_fast.bs4.imgsz640.fp16.v11_total_manual.engine
注: 文件名或许与我的不同，若不清楚就移除 *v4_fast.onnx、*v4_fast.engine.lock、*v4_fast.engine
注: 若 v2、v4_accurate 也曾编译过，也要移除
移除后再次使用命令
会再重新产生适用于 40x0 架构的引擎

另外请问你用的 CPU 型号是?

1970134465

12400+ b660g

1970134465

hodalahiruma 发表于 2026-6-10 20:16
因你先前是使用 5090 且使用了 --tensorrt 参数
这会预先编译好 BasicVSR++ 引擎使编码更快
但现在使用了  ...

删了，还是不行。
"python\python.exe" -m lada.cli.main --input "D:\Favorites\*\*.mp4" --output "D:\116" --temporary-directory "D:\116" --output-file-pattern "{orig_file_name}-U.mp4" --mosaic-detection-model "v4-fast" --encoding-preset "hevc-nvidia-gpu-uhq" --max-clip-length 180 --no-detect-face-mosaics --fp16 --tensorrt --torchcodec-encoder --fast-pipeline --device cuda:0

Compiling YOLO TensorRT engine for D:\lada-ex\model_weights\lada_mosaic_detection_model_v4_fast.pt. Output: D:\lada-ex\model_weights\lada_mosaic_detection_model_v4_fast.bs4.imgsz640.fp16.v11_total_manual.engine
  Building engine (this may take a few mins, workspace=11812MB)...
[06/10/2026-21:38:29] [TRT] [W] Functionality provided through tensorrt.plugin module is experimental.
W0610 21:38:30.457000 7744 Lib\site-packages\torch\distributed\elastic\multiprocessing\redirects.py:29] NOTE: Redirects are currently not supported in Windows or MacOs.

hodalahiruma

1970134465 发表于 2026-6-10 20:19
12400+ b660g

lada-ex 支持 Intel Quick Sync Video
这应该可以解决你的显卡性能因等待 cpu 解码时一直等候而无法将 gpu 负载拉高
1.首先先下载附件文件，文件中有下载位置，可更新最新 lada-ex 及 torchcodec
移除或备份以下文件夹:
lada-ex\python\Lib\site-packages\lada
lada-ex\python\Lib\site-packages\lada_ex-0.12.1.dev0.dist-info
lada-ex\python\Lib\site-packages\torchcodec
lada-ex\python\Lib\site-packages\torchcodec-0.11.1.dist-info
2.接着解压附件 lada-ex_update.rar，附件里有:
lada-ex\python\Lib\site-packages\lada
lada-ex\python\Lib\site-packages\lada_ex-0.12.1.dev0.dist-info
lada-ex\python\Lib\site-packages\torchcodec
lada-ex\python\Lib\site-packages\torchcodec-0.14.0+cu126.dist-info
放至:
lada-ex\python\Lib\site-packages
3.利用新版 lada-ex 与 torchcodec，开启 GPU 解码
打开命令提示字符，输入:
lada-ex.bat --input "D:\Favorites\*\*.mp4" --output "D:\116" --temporary-directory "D:\116" --output-file-pattern "{orig_file_name}-U.mp4" --mosaic-detection-model "v4-fast" --encoding-preset "hevc-nvidia-gpu-uhq" --max-clip-length 180 --no-detect-face-mosaics --hw-decode nvidia --hw-decode-detector intel --fp16 --tensorrt --torchcodec-encoder --fast-pipeline --device cuda:0
多了两个参数: --hw-decode nvidia --hw-decode-detector intel
解码交给 GPU、马赛克侦测器交给 Intel QSV
如此你的GPU性能只有10几的负载数字应可更高
这是参照 lada-ex 开发者所述
https://codeberg.org/comman/lada-ex/issues/24
你尝试下

hodalahiruma

1970134465 发表于 2026-6-10 21:37
删了，还是不行。
"python\python.exe" -m lada.cli.main --input "D:\Favorites\*\*.mp4" --output "D:\ ...

没看到报错
编译需要30分钟以上时间
与 jasna 相同
第一次编译都要等待

如果不想编译
将 TensortRT 参数关闭
--no-tensorrt

1970134465

"python\python.exe" -m lada.cli.main --input "D:\Favorites\*\*.mp4" --output "D:\116" --temporary-directory "D:\116" --output-file-pattern "{orig_file_name}-U.mp4" --mosaic-detection-model "v4-fast" --encoding-preset "hevc-nvidia-gpu-uhq" --max-clip-length 180 --no-detect-face-mosaics --fp16 --tensorrt --torchcodec-encoder --fast-pipeline --device cuda:0

Compiling YOLO TensorRT engine for D:\lada-ex\model_weights\lada_mosaic_detection_model_v4_fast.pt. Output: D:\lada-ex\model_weights\lada_mosaic_detection_model_v4_fast.bs4.imgsz640.fp16.v11_total_manual.engine
  Building engine (this may take a few mins, workspace=11812MB)...
[06/10/2026-21:48:06] [TRT] [W] Functionality provided through tensorrt.plugin module is experimental.
W0610 21:48:06.839000 1316 Lib\site-packages\torch\distributed\elastic\multiprocessing\redirects.py:29] NOTE: Redirects are currently not supported in Windows or MacOs.
WARNING: [Torch-TensorRT] - Using an engine plan file across different models of devices is not supported and is likely to affect performance or even cause errors or deadlock.
WARNING: [Torch-TensorRT] - Using an engine plan file across different models of devices is not supported and is likely to affect performance or even cause errors or deadlock.
WARNING: [Torch-TensorRT] - Using an engine plan file across different models of devices is not supported and is likely to affect performance or even cause errors or deadlock.
WARNING: [Torch-TensorRT] - Using an engine plan file across different models of devices is not supported and is likely to affect performance or even cause errors or deadlock.
WARNING: [Torch-TensorRT] - Using an engine plan file across different models of devices is not supported and is likely to affect performance or even cause errors or deadlock.
WARNING: [Torch-TensorRT] - Using an engine plan file across different models of devices is not supported and is likely to affect performance or even cause errors or deadlock.
WARNING: [Torch-TensorRT] - Using an engine plan file across different models of devices is not supported and is likely to affect performance or even cause errors or deadlock.
WARNING: [Torch-TensorRT] - Using an engine plan file across different models of devices is not supported and is likely to affect performance or even cause errors or deadlock.
WARNING: [Torch-TensorRT] - Using an engine plan file across different models of devices is not supported and is likely to affect performance or even cause errors or deadlock.
WARNING: [Torch-TensorRT] - Using an engine plan file across different models of devices is not supported and is likely to affect performance or even cause errors or deadlock.
Processing video:   1%|▋                                                                                                Processing video:   1%|▏                        |Processed: 00:19 (1302f) | Remaining: 56:4Processing video:   1%|█▏                                                                                                                  |Processed: 00:34 (2236f) | Remaining: 57:32 (225180f) | Speed: 65.2fps

依然是CPU100%，GPU没用