PS3 actually had a famously weak GPU.
It was quite good for the time. It was somewhere between the midrange 7600 and high-end 7800 GTX. The problem was that the Xbox 360 had a fully custom solution that was simply better.
The idea was that the GPU would only be used for the 'boring' 3D stuff and most of the cutting-edge high-perf shit would be run on the Cell's wacky architecture.
No, the idea was the Cell would be a hybrid CPU-GPU, with the massive bandwidth between the PPC core and the SPEs overcoming existing limits on traditional accelerated setups, somewhat like an AMD APU. Problem was, the yields were nowhere near what they needed to be for a chip as massive as they intended, they were way behind where they needed to be on developer tools, among other problems. Also, CPU-GPU bandwidth wasn't really an issue as more and more computation stayed GPU-native, unlike on the PS2. So they cut down the design a bunch, added in a commodity GPU, and told developers "like probably just use the SPEs for graphics preprocessing idk lol." Developers did indeed find useful things to do with the SPEs, but the original idea was that would be the GPU, full stop.
The final product that hit the markets was basically a bunch of compromises, kludges, and cuts to get
something out the door because Xbox was eating their lunch, and none of their original ideas were working.
Incidentally, SPEs look a lot more like modern CUDA cores than they do the vertex & pixel units of the time.