If we look at the Xbox One and PS4 as examples of shared memory, they have 8GB shared but obviously they don't come close to an 8GB video card, they're more similar to a 2GB or 3GB video card with 6GB of system memory. It's not all for the GPU, the part of the game that runs on the CPU needs that memory too!
As for the speed of the shared memory, remember that the memory bandwidth has to be shared between the GPU and CPU. For reference the PS4 had 176GB/sec of memory bandwidth. That sounds great, right? Well it still can't do 16x anisotropic filtering (a feature that relies on extra memory bandwidth) in most games, something that is easy for almost any dedicated GPU. The GTX 950 is way more powerful than the PS4, yet it only has 106GB/sec of memory bandwidth. Even the GTX 960 has less memory bandwidth at 112GB/sec. This is partly due to more efficient architecture with caching and delta compression and tiled rendering, but it's also due to the PS4 having to share that RAM. The CPU needs to use the memory too! The XXX has 326GB/sec while the RX 580 has 256GB/sec. Again, the GTX 1070 actually has the same amount of memory bandwidth as the RX 580 at 256GB/sec, but we know it's much faster.
Microsoft has been talking about 6 TFLOPS a lot, and people have really latched onto it. That's more TFLOPS than a GTX 980 Ti and near the GTX 1070! Oh wait, an AMD RX 580 gets over 6 TFLOPS with boost clocks or overclocking, and that isn't close to a 980 Ti or 1070. You just can't compare TFLOPS on different architectures when talking about game performance. the 6GB GTX 1060 which compares very closely to the RX 580 gets only 3.9 TFLOPS (4.3 with boost clocks). For another comparison, the PS4 gets 1.84 TFLOPS while the GTX 1050 only gets 1.7-1.8, and the GTX 1050 is definitely more powerful.
I've seen lots of people claiming that because the XXX has an 8 core CPU that means you would need an 8 core Ryzen or Skylake-X to compare. This is completely ridiculous because it's the same old Jaguar architecture as the 8 core CPUs in the Xbox One and PS4, which we know sucks. It's like people forgot that the original releases also had 8 core CPUs. It's an architecture designed for tablets. The CPU in the Xbox One and PS4 gets handily beaten by the dual core Pentium G4560, in most games the Pentium gets more than double the performance. Obviously the number of cores is not everything. The XXX increases the clockspeed by about 30%, not enough to close that gap.
The new integrated GPU on the XXX is a modified version of the RX 580, it has more cores but a lower clockspeed. In the end they get pretty similar TFLOPS as the XXX increases the number of cores by about 11% but decreases the clockspeed by about 7% to 13% for boost clock.
Sauce, with sauces
