If you're wondering, that's super computationally intensive and takes about 30 minutes per GPU on a Core i9-13900K. This, needless to say, led to much higher VMAF scores than intended! We have since recalculated all of the VMAF scores.
#Failed to open libx265 xvid4psp update#
Update : The initial VMAF scores were calculated backward, meaning we used the "distorted" video as the "reference" video, and vice versa. We also wanted to check how the GPU encoders fare against CPU-based software encoding, and for this we used the Core i9-12900K and Core i9-13900K. AMD ends up with the widest spread, at least in terms of speeds, so we ended up testing the RX 7900 XTX, RX 6900 XT, RX 5700 XT, RX Vega 56, and RX 590. For Intel, we're looking at desktop GPUs, with the Arc A770 as well as the integrated UHD 770. We'll test Nvidia's RTX 4090, RTX 3090, and GTX 1650 from team green, which covers the Ada Lovelace, Turing/Ampere (functionally identical), and Pascal-era video encoders. Thankfully, that leaves us with fewer GPUs to look at than would otherwise be required. We've checked the RTX 3090 Ti and RTX 3050 as an example - the fastest and slowest GPUs from Nvidia's Ampere RTX 30-series generation - and found effectively no difference. Generally speaking, the video encoding/decoding blocks for each generation of GPU will all perform the same, with minor variance depending on clock speeds for the video block. Here we're going to focus specifically on the video encoding performance and quality that you can expect from various generations of GPU.
Yes, games still get the most attention, but we like to look at the other aspects as well. Artificial intelligence training and inference, professional applications, video encoding and decoding can all benefit from having a better GPU. The best graphics cards aren't just for playing games.
0 kommentar(er)
