This suggests that the AMD 7900 XTX has better utilization than the NVidia 4090 (which is extremely surprising to me!!). Is there a "smaller NVidia" that could be used to see if utilization scales with size? The 4090 is physically and architecturally larger than the 7900 XTX after all.
There's also a question about smaller AMD cards, as the 7900 XTX has the largest "infinity cache". Do 16MB sized infinity caches keep utilization high? Or is this something that's only going to work on the 7900 XTX?
Just some idle thoughts. I'm personally interested in how the architectures utilization scales going up or down, and why that is. That is, VALU utilization for AMD, specifically (and SM Issue Active for NVidia or maybe SM ALU).