First of all, I would suggest trying running this software in the virtual machine on your Mac.
I'm not really familiar with virtualization (or probably emulation in case of Apple Silicon) on MacOS host, but something like Oracle VirtualBox, VMware Fusion or Apple dedicated UTM and Parallels should work.
There are might be some compatibility issues, but I think it should be a good idea to try virtualized/emulated Win11 on your Mac first instead of spending additional money on hardware.
I'm not really familiar with virtualization (or probably emulation in case of Apple Silicon) on MacOS host, but something like Oracle VirtualBox, VMware Fusion or Apple dedicated UTM and Parallels should work.
There are might be some compatibility issues, but I think it should be a good idea to try virtualized/emulated Win11 on your Mac first instead of spending additional money on hardware.
I’d need to run an ARM virtual machine on my M4 Mac—I’m not optimistic that would yield a workable result. But if someone can show I’m wrong, I’d be interested to hear about it.
You inspired me to check this. I skimmed thru some YouTube videos and it seems like people are able to play modern games (even those without native ARM support on Windows) on Macs using Parallels or Crossover.
If really demanding games run without any problems, I see no reason why you shouldn't try much less resource-intensive software. Especially when AKABAK developer states it works using WINE on Linux.
If really demanding games run without any problems, I see no reason why you shouldn't try much less resource-intensive software. Especially when AKABAK developer states it works using WINE on Linux.
@Few Which PC is suitable depends on how physically large the model is, the model symmetry, the number of freqs simulated, the max freq required, and of course your patience level 😉. You could get started, to see if you like this type of simulation, with a used PC of 8-12 threads, 16-32GB RAM, and an SSD (M2 preferred).
If you want to compare CPUs, you can use this benchmark CPU Benchmark i9-7920x (the CPU I currently use). Cooling is important otherwise the CPU will be thermally limited and throttle back to save itself.
Akabak BEM needs 1 thread per freq simulated. If there are 48 freq samples on a 24 thread machine, the sim will get divided into 2 sets (ie. 2x24). The RAM required depends on #threads and #mesh_elements. A large 6K mesh element model (1/4 symmetry) requires ~1GB per thread, and there are 24 threads, then approx. 24GB of RAM will be needed to avoid swapping to disk (M2 or SSD preferred). The #mesh_elements depends on the mesh size (related to max freq simulated) and model's physical size. The CPU I currently use, with an 6K mesh element model, with 64 sample freq, requires 32GB RAM (Win11 needs some too) and it solves in approx. 1.5Hrs. If that model was 3K mesh elements it would have solved in <30 min, and if it were 2K elements it would solve in <10min.
If you want to compare CPUs, you can use this benchmark CPU Benchmark i9-7920x (the CPU I currently use). Cooling is important otherwise the CPU will be thermally limited and throttle back to save itself.
Akabak BEM needs 1 thread per freq simulated. If there are 48 freq samples on a 24 thread machine, the sim will get divided into 2 sets (ie. 2x24). The RAM required depends on #threads and #mesh_elements. A large 6K mesh element model (1/4 symmetry) requires ~1GB per thread, and there are 24 threads, then approx. 24GB of RAM will be needed to avoid swapping to disk (M2 or SSD preferred). The #mesh_elements depends on the mesh size (related to max freq simulated) and model's physical size. The CPU I currently use, with an 6K mesh element model, with 64 sample freq, requires 32GB RAM (Win11 needs some too) and it solves in approx. 1.5Hrs. If that model was 3K mesh elements it would have solved in <30 min, and if it were 2K elements it would solve in <10min.
Mac can be dual booted, different parttions for different OS.
Can be more space saving / portable.
Can be more space saving / portable.
Not much to add to @DonVK’s answer than mentioning some examples… I have 3 PC’s running ABEC and Akabak fine on Win7, 10 and 11. The machines are respectively Dell T3500 (W3520), Dell M4600 (i7-2720QM) and Lenovo P15v (i7-12800H). On ABEC the P15v (20 threads) is about 4.5x faster than the two Dells (based on solve time/number of frequencies) when the number of frequencies is specified as a multiple of number of threads (optimal use). As you see, you get significant but not orders of magnitude real-world gain from a recent PC. It really depends what you are trying to simulate, but remember that solving high frequency is expensive since the number of elements required for a given accuracy goes up with frequency.
Though my advice earlier is broadly correct it assumed an acoustic BEM code that was parallelised for shared or distributed memory. I hadn't realized that akabak's BEM assembly and solve stages were single threaded. This would be unusual but seems to be confirmed in the hardware requirement part of the akabak website? (I am also not a Windows user). Akabak can run multiple frequencies at the same time in different threads which is good but it means an appropriate computer will need large amounts of memory to have a copy of the problem in memory for each core. The data in memory caches will only be relevant to one thread which means larger memory caches will be beneficial.
In your position and wishing to run akabak rather than an alternative I would be looking at running it on your current hardware rather than buying more that you don't seem to have much use for? The akabak website hardware page states it runs under wine with caveats.
In your position and wishing to run akabak rather than an alternative I would be looking at running it on your current hardware rather than buying more that you don't seem to have much use for? The akabak website hardware page states it runs under wine with caveats.
Out of curiosity I installed the current wine and the current demo akabak solver, plotting program and examples on a nearly 10 year old dual processor 16 core Xeon system running linux. I am not a wine or windows user and required a fair amount of time to work out where the installed files were and what wine wanted me to do to run windows programs.
I have no idea how to run the akabak point-and-click interface and the demo version doesn't store files and so everything disappeared after the run (I think one can store some values but I don't know what or how). I opened the first speaker example, a tower with 4 midwoofers and a waveguide tweeter, and selected run everything. It seemed to work in the sense plausible(ish) looking plots turned up at the end. The memory usage was reasonable. Most of the time it used all the cores with the rest of the time mostly a single core. It took about 5 to 10 minutes but I wasn't really paying attention. Unfortunately I don't know what calculations were performed on what grid resolutions and so can't really read across to other BEM solvers. Nothing appeared amiss though in terms of memory usage or expected calculation time.
I didn't use the arm version of wine so that remains untested but doing something similar on your machine would seem worth testing.
I have no idea how to run the akabak point-and-click interface and the demo version doesn't store files and so everything disappeared after the run (I think one can store some values but I don't know what or how). I opened the first speaker example, a tower with 4 midwoofers and a waveguide tweeter, and selected run everything. It seemed to work in the sense plausible(ish) looking plots turned up at the end. The memory usage was reasonable. Most of the time it used all the cores with the rest of the time mostly a single core. It took about 5 to 10 minutes but I wasn't really paying attention. Unfortunately I don't know what calculations were performed on what grid resolutions and so can't really read across to other BEM solvers. Nothing appeared amiss though in terms of memory usage or expected calculation time.
I didn't use the arm version of wine so that remains untested but doing something similar on your machine would seem worth testing.
Thanks, all, for the detailed advice and even test-driving! I’m definitely on firmer footing now than when I started. In the background, I have an inquiry into the tech folks at the college where I teach, in case they have a server they’d be willing to load Akabak on. If I could tap in remotely I could use the machine I have while taking advantage of some horsepower. Unless they no longer are maintaining a suitable server…. I hope to hear soon! In the meantime, I’ll have to look more seriously at the options for running Akabak on the Mac. I didn’t have much trouble getting hornresp running under Wine several years ago, but that was very different hardware, and age hadn’t burned through so many neurons at that point. My bio-based processor supports one thread at a time at a kHz clock speed.
Thanks again. If nothing else, maybe this informative thread will be helpful to others with similar questions.
Few
Thanks again. If nothing else, maybe this informative thread will be helpful to others with similar questions.
Few
I know you were talking about the installation effort. Nevertheless, it doesn't hurt to point out again that Akabak clearly benefits from many CPU cores and the highest possible memory bandwidth and -capacity, something that hardly plays a role in programmes like Hornresp.
It just depends on how complex the model to be simulated is and how long you may be prepared to wait.... 🙂
Yes, I realize HornResp is a completely different animal. That's why I started this thread. I knew I couldn't base my assessment on HornResp running under Wine. I just brought it up to point out I have a bit of experience with running Wine, albeit quite a few years ago. Including that unnecessary information probably just caused confusion.
Last edited: