onnx-preview2 #1121
Replies: 8 comments 29 replies
-
|
之前的体验记录一下。禁用了重复帧检测。 用的是 |
Beta Was this translation helpful? Give feedback.
-
|
Magpie.App.zip 我这里的测试结果 720p->1440p |
Beta Was this translation helpful? Give feedback.
-
|
这个速率对我来说常规算法基本可以退休了 |
Beta Was this translation helpful? Give feedback.
-
|
Is this meant to be here? I'm guessing the answer is yes, but the file is so large that I just want to make sure. |
Beta Was this translation helpful? Give feedback.
-
|
This file is essential for building the TensorRT engine. Its size is the reason it was not bundled with the program. |
Beta Was this translation helpful? Give feedback.
-
|
Thank you very much for taking the time to provide such a fantastic update to the application! I am using an RTX5070Ti. However, unfortunately, it doesn't seem to work correctly on my PC environment, which might be the cause. During the initial startup of TensorRT, the GPU usage rose close to 100% for about a minute, creating an engine. The engine files are located in the cache folder. After that, even though I turned on the upscaling, the onnx filter effects weren't noticeable. Additionally, there seemed to be no significant increase in GPU usage. Switching to DirectML also yielded no noticeable effect from the onnx filter. DirectML was working during the onnx-preview1 version. Strangely enough, after installing onnx-preview2, the DirectML functionality of onnx-preview1 stopped working as well. Currently, only the Lanczos upscaling specified in the scaling mode is effective for the visuals. I have attached a log file in case it might be helpful. |
Beta Was this translation helpful? Give feedback.
-
|
According to the logs, the ONNX model has been applied successfully. However, the overlay hasn't been adapted yet, and the processing time of the ONNX model is currently included in the first effect timing. Have you tried other models, such as rewaifu_anime_rtmosr_gan_x2_fp16_op18? |
Beta Was this translation helpful? Give feedback.
-
|
After restarting the PC, it started working normally! I deeply apologize for causing trouble without taking basic measures. This app is truly wonderful! I am eagerly looking forward to a version that allows filter swapping and input resolution changes. Once again, thank you very much! |
Beta Was this translation helpful? Give feedback.
-
|
Really good i tested on my gtx 1050 ti and the quality was good the fps was not it's not for me my gpu its so bad but the idea its awesome good work |
Beta Was this translation helpful? Give feedback.
-
|
trt还有个 --int8 参数，昨天用别的工具测了下（实际使用了--int8 --fp16混合量化，因为根据别的评价只启用int8效果可能会非常不理想），4080也只有大概5-10fps的变化（相比fp16），蚊子腿再小也算肉，至于兼容性和副作用还没观察 |
Beta Was this translation helpful? Give feedback.
-
|
文档说 tensorrt 的隐式量化已经弃用，推荐用校准数据来降低精度损失。这需要用户自己优化模型，但比隐式量化效果更好。 tensorrt 的工具：https://github.com/NVIDIA/TensorRT-Model-Optimizer/blob/main/examples/onnx_ptq/README.md |
Beta Was this translation helpful? Give feedback.
-
|
嗯最佳方案还是得原开发者使用匹配的校准数据进行正确优化 |
Beta Was this translation helpful? Give feedback.
-
|
不需要开发者自己适配，有典型数据就可以量化了，自己截图应该可行 |
Beta Was this translation helpful? Give feedback.
-
|
门槛还是有点高，一是没习惯存大量这类截图，而是好像因为版权问题没有这类的靠谱开源数据集 |
Beta Was this translation helpful? Give feedback.
-
|
要看值不值得，如果 int8 性能有革命性提升，我可以让 Magpie 在缩放时自动截图，每个游戏都可以有自己的校准数据 |
Beta Was this translation helpful? Give feedback.
-
|
Intel Arc A750 1080P→4K 註 若使用rewaifu_mahou_rtmosr_x2_fp16_op17.onnx約為15FPS |
Beta Was this translation helpful? Give feedback.
-
|
改使用Zarxrax_DigitalFlim_SuperUltraCompact_x2_fp16_opset14.onnx |
Beta Was this translation helpful? Give feedback.
-
|
AMD Radeon RX9070XT 以下都為1080P→4K windows11 24H2 驅動版本25.4.1 使用rewaifu_anime_rtmosr_gan_x2_fp16_op18.onnx 註1 使用rewaifu_mahou_rtmosr_x2_fp16_op17.onnx約為100-110FPS |
Beta Was this translation helpful? Give feedback.
-
|
似乎支持 Windows，但文档都是 Linux+Python |
Beta Was this translation helpful? Give feedback.
-
|
https://github.com/ROCm/AMDMIGraphX/blob/develop/CHANGELOG.md#migraphx-211-for-rocm-630 |
Beta Was this translation helpful? Give feedback.
-
|
发现 vs-mlrt 的作者搞了一个编译 CI，也提供了二进制文件，技术上没有阻碍了 |
Beta Was this translation helpful? Give feedback.
-
|
我测试migx的dll一点反应都没有，看了下编译过程还是用的rocm6.2的包，溯源去amd官网看了下好像windows还没放出6.4的版本 |
Beta Was this translation helpful? Give feedback.
-
|
虽然对标 tensorrt 但还差的远，我没兴趣折腾 |
Beta Was this translation helpful? Give feedback.
-
|
To the amazing developers behind Magpie, I wanted to express my deepest gratitude for creating such an incredible and game-changing tool. I’ve just witnessed performance that borders on magic — for the first time, I was able to run a modern, graphics-intensive game on my MX110 GPU — a graphics card considered nearly ancient by today’s standards. Not only did Magpie breathe life into my aging hardware, it delivered visual fidelity that I never imagined possible on this setup. The optimization, the rendering techniques, the ability to push frames with advanced scaling — it outperformed tools from major industry players. You’ve done what entire companies couldn’t. I will be proudly promoting Magpie through a video showcasing my system specs (which are closer to a museum piece than a gaming rig) and the in-game results I achieved using your software. I’m still in shock — the game looked too good to be true. Thank you for empowering users with low-end PCs. This is more than software — this is a miracle for gamers like me. With appreciation and respect, |
Beta Was this translation helpful? Give feedback.
Uh oh!
There was an error while loading. Please reload this page.
-
onnx-preview1...onnx-preview2
This discussion was created from the release onnx-preview2.
Beta Was this translation helpful? Give feedback.
All reactions