this is mostly wrong. The real issue has always been memory bandwidth. the highest end consumer x86 CPU has about the same memory bandwidth as a dGPU from 10 years ago. The M1 is extremely competitive with modern dGPUs, only a bit behind a 6900 XT.
If Intel/AMD are seriosus about iGPU, they implement solution for memory bandwidth (they did, Intel Iris Pro eDRAM, AMD gaming consoles using GDDR, upcoming AMD stacking SRAM). So I believe the core problem is that market didn't seriously want great iGPU but just fine with poorer iGPU or rich dGPU by Nvidia.
Apple compares the M1 Max as having similar performance to a nVidia's 3080 Laptop GPU, which scores around 16,500 on passmark. For comparison, the AMD 6900 XT desktop CPU scores 27,000, while the nVidia 3080 desktop GPU scores 24,500.
So the M1 Max is not as fast as a high end desktop GPU. Still, it is incredible that you are getting a GPU that performs slightly less than a last generation 2080 desktop GPU at just 50-60 watts.
Yes, apple's marketing materials claim 400 GB/s, while the 6900 XT is 512 GB/s. This is very easily Googled. While memory bandwidth isn't everything, it is the major bottleneck in most graphics pipelines. An x86 cpu with 3200 MHz memory has about 40 GB/s of bandwidth, which more or less makes high end integrated graphics impossible.
Ah, I misunderstood your comment. When you said it was competitive with the 6900 XT I thought you were talking about GPU performance in general, not just in terms of memory bandwidth.
According to the numbers Apple is touting, the M1 Max is competitive with modern GPUs in general, being on par with—roughly—a 3070 (laptop version) or a 2080 (desktop version). They've still got a ways to go but this is shockingly close, particularly given their power envelope.
> this is mostly wrong. The real issue has always been memory bandwidth.
Not really wrong. Memory bandwidth is only a limitation for a very narrow subset of problems.
I've gone back and forth between server-grade AMD hardware with 4-channel and 8-channel DDR4 and consumer-grade hardware with 2-channel DDR4. For most of my work (compiling, mostly) the extra memory bandwidth didn't make any difference. The consumer parts are actually faster for compilation because they have a higher turbo speed, despite having only a fraction of the memory bandwidth.
Memory bandwidth does limit certain classes of problems, but we mostly run those on GPUs anyway. Remember, the M1 Max memory bandwidth isn't just for the CPU. It's combined bandwidth for the GPU and CPU.
It will be interesting to see how much of that memory can be allocated to a M1 Max. It might be the most accessible way to get a lot of high-bandwidth RAM attached to a GPU for a while.
GP is talking specifically about GPUs. iGPUs are 100% bottlenecked by memory bandwidth; specifically it is the biggeset bottleneck for every single purchasable iGPU on the market (excluding M1 Pro/Max).
Your compute anecdotes have no bearing on (i)GPU bottlenecks.
