[xuegy]

This time it's configurable up to 128GB RAM so I'd say it's finally usable. For $4800($4319 for edu) you get 16 big cores, 128GB RAM, 800GB/s memory bandwidth.
I know it's still very expensive, but for how much $ can you build a comparable x86 workstation?

[wkernkamp]

Yes, should do the benchmark in 15 seconds, I think.

[flotus1]

Originally Posted by xuegy

I know it's still very expensive, but for how much $ can you build a comparable x86 workstation?

From a raw specs perspective: you can't.
If you have everything you need within the Apple ecosystem, and your CFD applications can utilize this theoretical performance somewhat effectively, there is just no competition for this in x86 land.
This new Mac might be within the realm of actually not having enough cores to saturate the memory bandwidth. So that's where a regular workstation could still compete with 48-64 cores. albeit in a different form factor, at a higher price point, and with higher power consumption. Apple's M1 chips have this extreme memory bandwidth because it also serves as the GPU memory.

[masofshad]

Some discussion here regarding memory bandwidth on M1 Max:

"Unfortunately, the news here isn’t the best case-scenario that we hoped for, as the M1 Max isn’t able to fully saturate the SoC bandwidth from just the CPU side"
. . . .
"While 243GB/s is massive, and overshadows any other design in the industry, it’s still quite far from the 409GB/s the chip is capable of. More importantly for the M1 Max, it’s only slightly higher than the 204GB/s limit of the M1 Pro, so from a CPU-only workload perspective, it doesn’t appear to make sense to get the Max if one is focused just on CPU bandwidth."

Only way to know for sure is to try it.

[Simbelmynë]

Originally Posted by flotus1

From a raw specs perspective: you can't.

Can we guesstimate? If it finishes the benchmark, single-core in 400-500 seconds then perhaps 25-31 seconds with the 16 performance cores is a reasonable guess if bandwidth is not an issue. That would be really good, but not the best workstation you can get. A version with 64 cores though, and full compatibility with Linux. They could probably sell that north of 100k Euros

[flotus1]

Of course we can guesstimate, extrapolation from incomplete data is one of my favorite pastimes
Let's take an optimistic 400s for a single-core result. That's where the M1 chips in the OF benchmark thread landed. Assuming ideal scaling on 16 cores gives us 25s.

We have a result from 2x Epyc 7313: 29s on 32 cores: OpenFOAM benchmarks on various hardware
And 2x Epyc 7532: 16s on 64 cores: OpenFOAM benchmarks on various hardware

The first one is in the same ballpark, the second one clearly faster than even the most optimistic projections for an M1 Ultra. Looking slightly ahead into the future, the introduction of DDR5, even more memory channels, and huge last-level caches. x86 will make a big leap in the next 1-2 years.

Both of these machines are a much larger form factor. I personally don't care too much about that.
It is safe to assume both machines use much more energy. And when I think about the price per kWh in my part of the world, I am starting to care.
And lastly, neither of these machines will be cheaper than 4800$, even if we only count the parts cost. Get it ready to run from some system integrator, and you can multiply the parts cost by a factor of 1.5x to 2x.

All of this is of course comparing literal apples to oranges. I would not buy an Apple computer for too many reasons to count. But if it works for you, why not?
I am genuinely surprised by Apples price policy for this machine. Only charging 800$ extra for the upgrade to 128GB of RAM is fairly reasonable. Let's ignore how much you have to pay if you want a height-adjustable monitor stand to go with it
Btw: Apple hinted that the Mac Pro will get a similar treatment next. Maybe 4 of these chips in a single package?

Some discussion here regarding memory bandwidth on M1 Max:

An interesting find, thanks.
That's the kind of stuff I had in mind when I hedged my bets in the first post