Maybe but Intel's and nVidia's products will be insaner if you are looking at raw power. Not every buyer is that bothered by energy constraints. Energy is money but so is time.As much as Apple wants to sell it there is a lot of scenarios when Ultra is not really 2x Max. Either way, its a great start. M2, M3..... will be insane
The issue is, that in the chart the line stops at about 300W despite the axis still going, suggesting that's is where the '3090' caps out. I agree that Apple is trying to show how the M1 Ultra's maximum W compares with a leading dGPU as the same W, but in doing so they made implications about performance limitations.
Maybe the 3090 should run Metal? Or maybe the benchmark is to run a video render or something using software optimized for each platform? It is likely that there is not a single purpose benchmark software that actually gets to the result optimized for the individual platform.
You beat me to it.
A interestedly take away is from the green standpoint, Apple is achieving a lot with far less electric usage. Yes the 3090 can scale a lot higher, but all you need is the equivalent of 2 M1 Ultras using 200 watts to somewhat compare to a single 3090 using 300 watts with that thought. If that was doable, then you have a 40 core CPU and GPU together this becomes a even more lop sided when thinking of enterprise expenses with use of power grids.
I'd like to see the math on that one
And is the card using that energy 100% of its uptime? (Or is that the minimum cost?)
It is not. Time consuming tasks like machine learning or rendering take long. Software specialists are somewhere between €100 and €200 an hour. AI specialists or people working at Apple or Google earn somewhat around $200K (can also be 100K or 300K).
I don't think the charts are going to make people who are already pre-disposed to Apple products think Apple has lost integrity.
You will need the GDP of a small nation just to afford this beasty chip and very deep pockets too boot.
They're not lying, Apple just cut off the top half of the chart where the 3090 keeps going. The M1 Ultra does beat it out at lower wattage. So misleading, yes, are they lying, no, technically it does beat it at performance per watt.
Historically linked GPUs a drop of 50% performance for each GPU added. Such that if GPU1 is 100 then GPU1 + GPU2 is 150, and GPU1 + GPU2 + GPU3 is 175. Apple complex bridging appears to allow them to get near 100% performance added, but linking two M1 Ultras would see 75% of the performance of a 3090. Three M1 Ultra, at 300W would see 82.5% of a 3090. Normalizing for performance the M1 Ultra tends towards infinite power.
But at 500W the 3090 performed twice as well as the M1 Ultra. This can't be extrapolated from the graph because the 3090's performance is showing decreased performance gains per watt at 300W. Are we to believe the 3090 performance per watt is a cubed function?
