But alas for me x86_64 > arm64.

Unified memory does go fast

macbook m4 max has ~5x the memory bandwidth of the next best high end x86 laptops, and can be configured with 128gb memory. i prefer x86 too, but for laptops the performance is not in the same league

Just a question, Does anyone has experience compiling FOAM in the M4 mini? I'm seriously considering buying one.

Just run docker, arm64 is easier on linux than mac

I saw a thread on cfd-online.com with detailed instructions for various methods

not quite what i remembered. there's another guy with a github repo (other than Alexey Matveichev) with details on compiling on apple silicon

hopefully asahi linux gets update to m4

so i was bored and doing some blender benchmarks, damn even the m2 ultra gives a run for its money, its only beating out (in cpu) by desktop cpu, mobile not so much.

To bad most software i use (outside of openfoam, blender and paraview, i,e, Ansa, Fluent, Starccm+, NX, Catia) dont have arm ports and/or lose a bunch of performance during translation

i know fluent and starccm+ have some arm port beta, but damn

time flies https://cdn.discordapp.com/attachments/727096272346939472/1327327269018013796/image.png?ex=6782a959&is=678157d9&hm=bb582f33c008f8277b2c676ab40d2143d924dad3c64d5a72df1dd44f6deee867&

Looks like createPatch is somehow broken in OF2412, gonna create another bug report 😣 Maybe this has something to do with it https://www.openfoam.com/news/main-news/openfoam-v2412/pre-processing#create-patch

So, what are the implications of Keysight buying ESI to OpenFOAM.com? Any guesses?

No but I'm wondering. I just realized the former director of OpenCFD left few months ago. I have no idea if this is related to Keysight or not though.

But as a consequence, nobody has been announced yet to replace him as the chair of the steering committee.

Have you noticed anything on gitlab? I feel like activity slowed down over the past months but it's just a feeling

No, I haven't been active there before now

just a follow up, ryzen ai max 395+ looks like it will be a pretty decent mobile chip. 16 cores & 64mb of cache which is decent - but most importantly for those of us that sometimes want to run a sim locally, in order for them to push out 50 TOPS they've used a unified memory architecture that in theory has 256GB/s of bandwidth for reference, this is more bandwidth than than an 8-channel threadripper pro 5995WX, more than 4th gen epyc 8000 series chips, and more than any of the 1P epyc 3rd gen or older systems

I'm excited if that memory bandwidth can be fully used by the cpu. because Apple advertise memory bandwidth that can only be fully available if GPU is used.

yeah that's my concern as well. supposedly the m4 cpu can access a larger portion of the bandwidth than the earlier chips but i've not seen any actual data on it

I saw a interview where one of amd engineers said "cpu should be able to fully saturate it as we tested that case, but that's not our target workloads"

is there some module for solid fuel combustion with openfoam?

createBaffles

documentation says: Makes internal faces into boundary faces. Does not duplicate points. Is there a tool to duplicate (and merge) the mesh points for baffles only?. Update: Found it:

mergeOrSplitBaffles

I found an indirect definition of baffle in

etc/caseDicts/annotated/mergeOrSplitBafflesDict

in the split option:

// Detect baffles (on selected patches) and duplicate the points. This is
// used if e.g. the two sides need to move separately. Note that since the
// points are duplicated the two faces are no longer baffles.

So, the requirement for a baffle is that they share points 🤔 but clearly the faces are not shared. Learned something new today: baffle != two overlapping patches, although they look the same in the polyMesh/boundary file. Is there a way to tell if a mesh has baffles? Or overlapping patches?

Wow, the duplication of baffle points works. Note: You need to tell the dict location explicitly to mergeOrSplitBaffles like

mergeOrSplitBaffles -dict system/mergeOrSplitBafflesDict

. If the baffle is inside a region it only duplicates the internal baffle points, not edge points. 👍

that's why I never got any useful results from this utility

can I ask what is the difference/impact? i am dealing with an internal wall that i treating with baffle, and didnt thougth too much about it and simply put it there to fast the meshing process...

When points are shared (so it's a baffle), the topology of point-to-cells is different. You can navigate right through a baffle via point-cell connection. But not with face-to-cell connection. It's really irritating to handle this special case in mesh algorithms. Also, e.g. Blender does not allow two faces to share exactly same points --> can't represent baffles with faces in Blender without workarounds.