How easy/difficult is it for developers to update their software for M1 Ultra?

[leman]

This link has been posted in some other forum thread IIRC, showing the M1 Ultra scaling pretty nicely.

That’s the advantage of high bandwidth memory.

 

[theorist9]

This link has been posted in some other forum thread IIRC, showing the M1 Ultra scaling pretty nicely.

That's interesting; thanks for sharing that. It sounds like those results are for an older program he's been using for a while, which enables him to see generational evolution in processor speeed:

"To really convey my experience, I want to set the stage with previous results from the computational fluid dynamics (CFD) benchmark I’ve been running for the last 10 years."

I don't know if that program is still used in practice today; if not, it would be nice to see some examples of scaling at high core counts on the kind of programs researchers use currently.

 

[Bodhitree]

It really depends how much mileage you get out of multithreading a system. if you look at games, there was a time when games were all single-threaded, then they moved to hardcoding specific tasks to certain threads. The first step you traditionally do when moving a codebase onto multiple threads is to move the rendering engine onto a seperate thread. This kind of approach gives you some rough parallelism, but it’s hard to scale when moving to 4 threads or 8 threads.

These days, most large games have switched architecture to a job-based approach, where you split the work the engine is doing into many small jobs, which are then farmed out to a pool of threads. This is a lot more scaleable, and it really depends on how many jobs you’ve got whether you can keep your thread pool at 100% utilization.

With more typical Mac software, utilities and creative programs and so on, there are generally fewer tasks than in game software. It will take some time for developers to get used to the new way of working.

 

[Yebubbleman]

So in simple terms, if an app is already compiled for AS and runs well on the multi-core of M1, Pro and Max, why does it require more work for Ultra?

In simple terms, it's the fact that M1 Ultra is two M1 Max SoCs. Yes, "Ultra-Fusion" is a thing and, yes, it's better than you'd ordinarily get with two distinct SoCs (and, no, it's not the same as just adding on more cores). But that link is not insubstantial and why apps not optimized for M1 Ultra perform similarly to a standard M1 Max.