GPU simulations?

[Paul Oblomov]

Fire/smoke sims using a pure grid solver need to calculate the velocity in each voxel, no matter if its occupied by a visible channel or not

h18 introduces sparse pyro, not vdb, but now it doesn't calculate whole grid all the time, pretty fast tho.

[Svetlin.Nikolov]

This is a compromise with realism though. Even though you don't see it, it's important to calculate the air currents surrounding the visible fire/smoke.

[cb LLC]

This is a compromise with realism though. Even though you don't see it, it's important to calculate the air currents surrounding the visible fire/smoke.

Hi,
It would be nice to have it as an option if you wanna do fast calculations

[cb LLC]

Hey guys, since we don't have these CPUs here it would be very useful if you can send over your Phoenix logs from these simulations, so we can see how they compare with our setups.

HI
I want to add that after xref merge pf LIQUID grid and doing some sim test speed was around 3 mil per sec. After rebooting max it became 8. Then setting SPF to 6 from 1 speed is around 30-35 mil per sec. Not sure why

I remember restarting max and faster sim speed bug used to be long time ago

Thanks

[Svetlin.Nikolov]

Yup, changing SPF will change the voxel/sec. This is why it's unreliable. As I already suggested, the only case you should trust it would be when you haven't changed any dynamics settings.

This is why the merging issue sounds like something that needs follow up - so you get lower vox/sec, save the scene, restart max and without changing any settings it's quick again? Does it reproduce again if you try the same steps? It would be great if we manage to get this to happen here as well so we can understand why and figure out what needs to change.

Cheers!

[Paul Oblomov]

Well, I had this effect (restart and it is fast again) in earlier version. I think, I submitted information, but feels like it wasn't top priority. It can be reproduced IMHO with ease. I was using my little script, for batch sims with different settings and simtime was higher and higher each round. Tho different settings, so it can't be treated accurately, but my suggestion to devs - just batch sim same scene again and again, 10 times and you'll see that phx gets slower, or it is max get slower. I remember even viewport perfomance degradation, tho not sure is it was fixed.

[Svetlin.Nikolov]

Paul Oblomov , the issue you have with scene interaction was fixed, maybe you missed it. Voxelization of geometries took longer and longer. Could be a regression, or something different altogether...

[Paul Oblomov]

Yeah, just guessing, you know...