Cache Rendering Issue

[A_A_M_I_R]

Running Win 7 Professional x64. Houdini latest version 12.5.533.

I've been rendering this simulation for some time now. Anyway, what I noticed is that Windows has allocated around 14GB (stays around there) to the Cached Memory. Now from what I've been told, this is standby memory and should be allocated to a program if it needs it. In my render settings, I set the max cache ratio to 0.9 so shouldn't it all be allocated to the houdini render? My CPU is maxing out the whole time but the RAM keeps floating around 14GB.

My actual simulation (rop_geo) was around 6GB so even if I take that into account I should still have 8 GB left. I even closed almost everything down and still didn't see much change? I tried stopping my render, going to the Houdini Cache Manager and clearing it up but still had at least 13GB or so of cached physical mem. I'm not sure if it's Maya but even it's cache wouldn't even come close to 2GB when it's not open right? I cleared the After Effects cache files as well. Other than that, I only really run Nuke and I know it doesn't take up that much.

http://i296.photobucket.com/albums/mm182/A_A_M_I_R/houdini_cache.jpg

Any ideas?

[symek]

1. Render cache for Mantra related to settings you mentioned it's just a part of memory allocating by mantra. Specifically, your heavy geometry read from disk, won't be included in that number. Only procedurals (geometry created by mantra at render time) and textures are maintained with that cache. You may turn on Mantra statistics to 3 or so, to see cache info at the end of render.

2. Mantra and Houdini allocates separate pools of memory, so if you don't render from ifd file, you have probably doubled memory usage for sim geometry.

3. You can't really compare applications so different as Nuke and Houdini in memory consumption. Apples to apples would be Realflow hybrido to Houdini Flip, or similar Houdini Cops' network to Nuke network.

4. My brand new Windows 7 x64 chews up to 2GB memory on start jumping to 3GB with stupid Firefox opened compared to 200MB of my Centos 6.2, so it's really doesn't make sense to look at total system ram usage for debugging anything. Windows is famous for notorious problems with managing large memory pools.

5. It's hard to tell without details, but numbers seems correct, with 6GB of geometry render can take another couple of gigs depending on a scene.

[mangi]

Like Szymon mentioned about windows , "Measuring memory usage in Windows 7"

http://brandonlive.com/2010/02/21/measuring-memory-usage-in-windows-7/

Cached – Here things get a little more confusing.

[edward]

In addition to the notes that others have mentioned, it is also worth pointing out some things:

- Mantra maintains alternate representations of geometry after loading that can be more compact than the Houdini geometry representation

- Mantra will only use as much memory as needed by rendering. It might not even be maxing out your cache at all. Also double-check that it thinks your total memory is correct by looking at the statistics, and/or running the "memory" hscript command in Houdini.

- It is a misconception to think that you always need to use more memory in order to be faster.

[A_A_M_I_R]

1. Render cache for Mantra related to settings you mentioned it's just a part of memory allocating by mantra. Specifically, your heavy geometry read from disk, won't be included in that number. Only procedurals (geometry created by mantra at render time) and textures are maintained with that cache. You may turn on Mantra statistics to 3 or so, to see cache info at the end of render.

2. Mantra and Houdini allocates separate pools of memory, so if you don't render from ifd file, you have probably doubled memory usage for sim geometry.

3. You can't really compare applications so different as Nuke and Houdini in memory consumption. Apples to apples would be Realflow hybrido to Houdini Flip, or similar Houdini Cops' network to Nuke network.

4. My brand new Windows 7 x64 chews up to 2GB memory on start jumping to 3GB with stupid Firefox opened compared to 200MB of my Centos 6.2, so it's really doesn't make sense to look at total system ram usage for debugging anything. Windows is famous for notorious problems with managing large memory pools.

5. It's hard to tell without details, but numbers seems correct, with 6GB of geometry render can take another couple of gigs depending on a scene.

Thanks, yeah I assumed it should've been only mantra but I got worried after seeing the numbers. Also, I didn't render from idf as I really don't know how to use it yet! It seems the problem has resolved itself though after the render was complete, image below.

Like Szymon mentioned about windows , "Measuring memory usage in Windows 7"

http://brandonlive.c...e-in-windows-7/

Cached – Here things get a little more confusing.

Very interesting article, will be going over this soon. thanks!

In addition to the notes that others have mentioned, it is also worth pointing out some things:

- Mantra maintains alternate representations of geometry after loading that can be more compact than the Houdini geometry representation

- Mantra will only use as much memory as needed by rendering. It might not even be maxing out your cache at all. Also double-check that it thinks your total memory is correct by looking at the statistics, and/or running the "memory" hscript command in Houdini.

- It is a misconception to think that you always need to use more memory in order to be faster.

Thanks, will also check out the memory hscript command you mentioned! Is the Alfred progress supposed to show the memory available to mantra as well?

====================================================================

So after the render finished, and this is the day after it seems everything has gone back to normal! (image attached)

Edited October 20, 2013 by A_A_M_I_R