Is ~19 hours for a 7-page complex PDF reasonable on RTX 3060 Ti + i9?

[doneself]

Hi,

I ran Chandra on Windows 10 (via PowerShell) to process a 7-page PDF containing tables, graphics, and Chinese text. My system specs are:

• GPU: NVIDIA GeForce RTX 3060 Ti (8 GB VRAM)
• CPU: Intel i9-10900KF (10 cores, 20 threads)
• RAM: 128 GB
• PyTorch: Confirmed torch.cuda.is_available() returns True and detects the RTX 3060 Ti

I ran the following command on Windows 10 via PowerShell:

chandra input.pdf ./output --method hf

The entire job took about 19 hours to complete.

To help illustrate the layout and content complexity, I’ve attached page 1 of the PDF as a reference.

Given this hardware and document type (including Chinese), is this processing time expected? Or could it indicate a configuration or performance issue?

Thanks for your help!

[loowace]

显存至少24g，最好有32g

[doneself]

显存至少24g，最好有32g

那就是这么复杂的PDF，用8G的显卡，跑19个小时，是合的吧。

[loowace]

显存至少24g，最好有32g

那就是这么复杂的PDF，用8G的显卡，跑19个小时，是合的吧。

www.datalab.to 直接用官网的ocr，很快的，跟pdf复杂不复杂关系不大，就是显存小了

[doneself]

我用简单一点的带表格的PDF，一页要20分钟。如果像上面那么复杂的页面，要两个小时。

[loowace]

我用简单一点的带表格的PDF，一页要20分钟。如果像上面那么复杂的页面，要两个小时。

www.datalab.to 直接用官网的ocr，很快的，一两分钟一页