The M* ultra has more cores/execution units on the gpu and cpu side of things, but the best m4 max is comparable. Quick rundown: 1. Memory bandwidth is abou

The M* ultra has more cores/execution units on the gpu and cpu side of things, but the best m4 max is comparable. Quick rundown: 1. Memory bandwidth is about the same (same width bus but 533 MT/s difference favoring Apple) 2. Total core count is the same: 16 3. SH is has only 1 types of CPU core (no zen5/zen5c mix like strix point). The cores are also "full fat" (no double pumping the avx-512 instruction) Zen 5 cores 4. M4 Max has 12P/4E cores split, but Apples M4 E-cores are famously very competent. Definitely faster than intel's e-core, probably not as fast as zen 5c (but I'm just guessing a bit as no one really does e-core vs e-core benchmarks) 5. Performance really depends on the power envelope, the cpu performance split likely swing M4 Max on top until at least 100w. Beyond that, it could go either way. Regardless it is not night and day anymore. Basically everyone (Apple/Qualcomm/AMD/Intel) makes a fast p-core at this point (+/-10%). The real difference is in the power/performance curves and Apple still does so much better in the lower power portions of their curve (less so in the upper) 6. Both GPUs have 40 cu, but that isn't really apples to apples (I don't know how many shaders/fp/int/rt/etc is in a single m4 CU) 7. In general, AMD RDNA 3.5 GPU cores are just faster and have better support in both gaming and ML (ROCm... is fine. It's no CUDA, but lots of libraries support it now) 8. Apple will probably support it's ML stack longer. ROCm may be better now, but AMD is famous for dropping support for 'old' devices really quickly 9. Apples GPU/CPU shared memory arch is just better. The GPU and NPU get full access to all the RAM without any futzing. On AMD you can set the GPU VRAM to 75% of total memory, anything else is GTT (at least on linux). Lots of ML libraries ignore GTT 10. SH has more NPU TOPS if that matters. 38 vs 50, I think Basically, for a small desk unit, it's a wash. However, considering that the M4 Max has been out for a few months now and no SH device has a release date... Still, credit to AMD for finally showing up, no one else has been.