This comment is being furiously deleted by the site moderator for some unknown reason:
---
iGPU suffers much more from the RAM bandwidth, because it does not have a dedicated GDDR6/HBM3 like dGPU.
In addition, historically, it turned out that AMD controllers from review to review have proven worse efficiency relative to the nominal frequency of memory modules compared to Intel. Although the palm of primacy is undoubtedly Apple - it is impossible to imagine a worse memory controller. 400 GB / s is declared, but in reality everything rests on 120 GB / s. This is a shame.
The regular Zen5 Strix has an outdated 128-bit memory controller, but its efficiency was brought to 80%, which is extremely good. For the first time on AMD, I saw numbers in the region of 100 GB / s for laptops with AMD and lddr5 7500, which in theory should have had an extremely positive effect on the performance of igpu, so the results presented "slightly" surprise me. Something is wrong here.
In any case, today Apple has the absolute lead in RAM bandwidth - 120 GB/s on laptops.
We can only wait sadly for the release of Zen5 Strix Halo with a 256-bit memory controller interface, where, with the same lpddr5 7500, the RAM bandwidth should theoretically exceed 200 GB/s. But as we have already seen from the shameful experience of the Apple development team, theory and reality are very far from each other...
I do not expect any serious progress in the iGPU until they are provided with dedicated HBM3 memory on a crystal with a real bandwidth of 300 GB/s.
When could this happen theoretically? Not earlier than 2028...