>>105952302
>>105952316
>>105952383
Another chart to compare but yeah, the numbers are only relative. One thing to note is that although Nvidia is using TensorRT and Intel is using OpenVINO, ONNX is what is being used on AMD so they are massively undermeasured right now on the benchmarks. Partially their fault. The benchmark should be updated once ROCm actually works on Windows. Still, as good as it gets for performance measurements. I really wish you could do actual comparisons roughly looking at the SD.NEXT benchmark data.