> I have 80GB/s memory bandwidth [1] on my 1950X. The MBP has 400GB/s memory bandwidth. 5x(!)

Your 1950x also doesn't have a GPU. For comparison the rx6700xt, and Vega 56 have ~400GB/s. The 6 year old R9 nano has 500GB/s.

The M1 Max *CPU* cores can access RAM at 400GB/s. My 1950x CPU cannot use the 448GB/s memory bandwidth on my RTX 2080 (which only has 8GB of GDDR6 RAM). This unified memory is wild.

1. M1 Max has 50+ billion transistors (only a part of that count is CPU, so adjust the number downwards), 1950X (first gen 14-nm Ryzen) has 10 billion.

2. Multicore Cinebench R23 for M1 Max: 14970; Ryzen 1950X: 18780

Geekbench tests in short bursts, not taxing the CPU thermals. Cinebench is the opposite. In short, CPUs do well on some benchmarks and not so well on others.

The Cinebench R23 results for 1950x[1] and M1Max[2], including single and multi-core:

           single     multi
  M1Max:    1562      14970
  1950X:    1035      18780

Remember that the M1 Max has 8 high performance and 2 efficiency cores vs. the 16 cores of the 1950x. Again, this is a pretty convincing victory - 1.5x single core perf in a laptop - this is not a HEDT part.

[1] https://www.cpu-monkey.com/en/cpu-amd_ryzen_threadripper_195...

[2] https://www.cpu-monkey.com/en/cpu-apple_m1_max

So you are comparing a laptop that is just coming out to something 4 years old.

a 4 year old workstation/HEDT that was widely viewed as being an out-of-band excellent performer for its time, and the laptop is outperforming it in multithreaded performance by almost a factor of 2, with half the (performance) cores, with no SMT.

yeah, that actually is fairly wild. Granted, there are good technical reasons there (lower memory bandwidth, the 1950X having a NUMA/NUCA arrangement that often hampers the ability to bring all the cores to bear on a single task, Zen1 having extremely high memory latency and cross-CCX latency in general, etc) but it still actually is a very impressive feat.