I am seeing significant streaming performance differences between two ESP32-CAM generic AI-Thinker modules (off of AliExpress). The first one gives me ~12fps at UXGA resolution, the second one ~3fps at UXGA, using the library "ESP32/examples/Camera/CameraWebServer.ino" code unmodified. Distance from the WiFi router is the same (~3m), and switching OV2640 camera modules between the two makes no difference.
Both are new modules from different suppliers. Upon physical inspection, I found that the first module (faster) has a IPS6404L‐SQ PSRAM chip, and the second (slower) has a LY68L6400. Looking at the datasheets, the LY68L6400 appears to be a faster chip, but both chips run at 84MHz when doing cross page boundary operations, which I assume (without digging into the code) both are doing due to the size of the mpeg stream.
Any ideas why the modules would have such a big speed difference (one 4x faster than the other?)
Ok, I'm late to the party but this may be of interest anyway.
Since they are the same type, my immediate guess would be different quality antennas, especially since they are from two different suppliers. There are sketches out there which will list the various WiFi APs within range and also displays their signal strengths, try one of those and compare your two units
I have done some experimenting myself lately, using AI Thinker, with and without an external antenna and the difference in fps is striking. I was using one (w/o ext. antenna) to keep an eye on my 3D printer, and like you had just a few meters between the ESP32CAM and the router, I got fairly decent speed from it. But when I had it in another room to do some work on it, the rate dropped to next to unusable, presumably because it now had to communicate through two brick walls. I popped an external antenna on it and the frame rate was suddenly better than ever! I've tried this with other AI Thinkers too, with similar results.