Apple does not use HBM, just LPDDR5 with wider interfaces. Xbox X/One, PS4/5, AMD server/workstation chips (threadripper (and pro), siena, and epyc), and apples above the base model m1/m2/m3 all use wider memory interfaces.
AMD agpu (and Intel with iGPUs) do use unified memory. As well as the high end AMD (MI300A) and Nvidia (Grace+hopper) server+iGPU chips.
This REALLY helps GPUs performance which is very bandwidth intensive, but even normal workloads have cache misses. Part of the benefit isn't just more bandwidth, but that you can have 4,8, or 16 cache misses running in parallel which can help keep more of your cores doing useful work instead of waiting on memory. This doesn't decrease memory latency, but can help improve throughput when using multiple cores.
Sad that even a M3 pro (the midrange, 3rd fastest in the apple M series lineup) available in Apple laptops or mac minis has more bandwidth than every AMD Ryzen and every Intel i3/i4/i7/i9. Even Intel's fastest i9-14900KF using a maximum of 250 watts has 1/8th the memory bandwidth of the M2 Ultra, 1/4th the bandwidth of the M3 max, and 1/2th the bandwidth of the M2 Pro. Same (or less actually) then the MBA or base Mac mini.
AMD agpu (and Intel with iGPUs) do use unified memory. As well as the high end AMD (MI300A) and Nvidia (Grace+hopper) server+iGPU chips.
This REALLY helps GPUs performance which is very bandwidth intensive, but even normal workloads have cache misses. Part of the benefit isn't just more bandwidth, but that you can have 4,8, or 16 cache misses running in parallel which can help keep more of your cores doing useful work instead of waiting on memory. This doesn't decrease memory latency, but can help improve throughput when using multiple cores.
Sad that even a M3 pro (the midrange, 3rd fastest in the apple M series lineup) available in Apple laptops or mac minis has more bandwidth than every AMD Ryzen and every Intel i3/i4/i7/i9. Even Intel's fastest i9-14900KF using a maximum of 250 watts has 1/8th the memory bandwidth of the M2 Ultra, 1/4th the bandwidth of the M3 max, and 1/2th the bandwidth of the M2 Pro. Same (or less actually) then the MBA or base Mac mini.