FDS7: Expected computational cost vs. FDS6, and work in the area of natural smoke ventilation.

[Lukako26]

Hi,

I’m currently working on my master’s thesis focusing on natural smoke ventilation and CFD modeling in FDS. The following questions are interesting topics for me to include in my discussion.

I have two related questions regarding current and future development of FDS:

1) Computational cost in FDS7
From the transition between FDS5 and FDS6, it seems that improvements in physical modeling (e.g. combustion, radiation, numerical treatment) have generally increased the computational cost per simulation.

Should we expect a similar trend in FDS7, where increased physical fidelity leads to longer runtimes for comparable cases?
Or are there ongoing efforts to offset this through numerical improvements or performance optimizations?

2) Natural smoke ventilation modeling
Is there any ongoing work or discussion within the FDS community regarding natural smoke ventilation, particularly in relation to under-resolved openings?

I'll appreciate any and all answers. It is of course also understood if there's information that can't be disclosed.

Thank you and best regards
Lukas Larsen

[mcgratta]

There were certain changes in the algorithm going from 5 to 6 that were costly. We don't expect the same for 7. As for ventilation, you can see from the Discussion forum that questions arise on a regular basis, but I haven't seen a grid resolution study on it. That might be an interesting topic.

[rmcdermo]

From the current user guide...

[rmcdermo]

Actually, the comment about ULMAT needs to be revised, since UGLMAT HYPRE is now the default for GEOM and this does not slow with scale since it is an iterative method.

[marcosvanella]

Wait ULMAT is the default for GEOM. You have to choose UGLMAT to run with the global solver.

[rmcdermo]

OK, I thought we had changed that. Is there any reason not to use UGLMAT as default now? I'd say let's flip that switch.

[marcosvanella]

One reason is when multiple pressure zones are present in a multiple mesh case. Ulmat solves sequentially per zones in the mesh (likely a subset of all zones) whereas UGLMAT runs a sequential loop per zone and does a solve on each (all processes). We should be able to deal with this by restructuring the code such that matrices (potentially condensed if indefinite) are loaded as diagonal submatrices of a large matrix containing all zones unknowns, and sending that with a global rhs across all zones to the solver. Other than that which solver is faster than the other on a single pressure zone-multiple mesh case depends how many pressure subiterations Ulmat needs to do. If it has to run up to 5-10 subiterations UGLMAT is generally faster.

…

________________________________ From: Randy McDermott ***@***.***> Sent: Thursday, March 26, 2026 10:08 AM To: firemodels/fds ***@***.***> Cc: marcosvanella ***@***.***>; Comment ***@***.***> Subject: Re: [firemodels/fds] FDS7: Expected computational cost vs. FDS6, and work in the area of natural smoke ventilation. (Discussion #16093) OK, I thought we had changed that. Is there any reason not to use UGLMAT as default now? I'd say let's flip that switch. — Reply to this email directly, view it on GitHub<#16093?email_source=notifications&email_token=ABY23VMLRCRWIRITKLD63UL4SU2VFA5CNFSNUABIM5UWIORPF5TWS5BNNB2WEL2ENFZWG5LTONUW63SDN5WW2ZLOOQXTCNRTGI3TONBZUZZGKYLTN5XKOY3PNVWWK3TUUVSXMZLOOSWGM33PORSXEX3DNRUWG2Y#discussioncomment-16327749>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/ABY23VM7YX2GKPQM4WL2RAD4SU2VFAVCNFSM6AAAAACXADP7VWVHI2DSMVQWIX3LMV43URDJONRXK43TNFXW4Q3PNVWWK3TUHMYTMMZSG43TIOI>. You are receiving this because you commented.Message ID: ***@***.***>

[rmcdermo]

OK, should we use ULMAT if we have multiple zones and UGLMAT otherwise?

[marcosvanella]

Yes, let's see where in the source we should move the test, if needed. Currently the Poisson solver test is done at the end of READ_GEOM IF (PRES_FLAG/=UGLMAT_FLAG) THEN PRES_METHOD = 'ULMAT' PRES_FLAG = ULMAT_FLAG ENDIF We have to see if N_ZONE has already been computed by then.

…

________________________________ From: Randy McDermott ***@***.***> Sent: Thursday, March 26, 2026 11:22 AM To: firemodels/fds ***@***.***> Cc: marcosvanella ***@***.***>; Comment ***@***.***> Subject: Re: [firemodels/fds] FDS7: Expected computational cost vs. FDS6, and work in the area of natural smoke ventilation. (Discussion #16093) OK, should we use ULMAT if we have multiple zones and UGLMAT otherwise? — Reply to this email directly, view it on GitHub<#16093?email_source=notifications&email_token=ABY23VOGGJXY7ME5RH646A34SVDL7A5CNFSNUABIM5UWIORPF5TWS5BNNB2WEL2ENFZWG5LTONUW63SDN5WW2ZLOOQXTCNRTGI4DQNBWUZZGKYLTN5XKOY3PNVWWK3TUUVSXMZLOOSWGM33PORSXEX3DNRUWG2Y#discussioncomment-16328846>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/ABY23VKI6DG62FQJ4UQY4BD4SVDL7AVCNFSM6AAAAACXADP7VWVHI2DSMVQWIX3LMV43URDJONRXK43TNFXW4Q3PNVWWK3TUHMYTMMZSHA4DINQ>. You are receiving this because you commented.Message ID: ***@***.***>