Slaves with 8 CPU cores rendering slower than slaves with 4 CPU cores

[tricky]

We have a mixture of render slaves, 4 of them have 4 CPU cores and 4 of them have 8 CPU cores. They all have 4GB ram.

Ordinarily, the 8 core machines are significantly faster than the 4 core machines, close to twice as fast. We are running through an animation at the moment, and for some reason, the 8 core machines are running about 25% slower than the 4 cores. I have obviously changed something that has affected things somewhat.

I have changed the face/level coefficient to 2.0 and changed the dynamic ram to 2000MB. The scenes are reaonably complex with quite a few proxies, textures and wind-animated trees (which are not part of the pre-calced GI solution). Memory useage whilst rendering is around 3GB.

Would either of these two settings slow down the renders from the faster machines?

[leehumphries]

my comment isnt an answer to your problem but something that was brought to my attention recently from Boxx Tech....I have Xeon 2.26 16 core workstation and recently i was asked in the bios to change this

Please ask Lee to check this bios setting due to some newly discovered changes.
Advanced – Processor & Clock Options – Simultaneous Multi-Threading – Disable

For some reason im told and maybe corrected that some people disable this and means rather than me getting 16 buckets for the 16 cores while rendering i get 8,which i think im told is more powerful,how much i dont know.

Maybe someone could expand on this if suitable to the topic

Hope u get it sorted

[vlado]

Would either of these two settings slow down the renders from the faster machines?

Nope. Did you try disabling the bitmap pager?

Best regards,
Vlado

[trick]

Is memory usage the same (3GB) on both 4 and 8 core machines?
My experience is that machines with more cores reach the dynamic memory limit far sooner then machines with less cores: try setting the limit on the 8-core at 3GB.

[tricky]

Nope. Did you try disabling the bitmap pager?

Best regards,
Vlado

No. Funnily enough, the bitmap pager is the other item that I changed over the face/level coef and the dynamic RAM. I've never really used the bitmap pager, but due to the amount of memory my scene has been using whilst rendering (up to 7GB), I enabled the Bitmap Pager. I also enabled it on all the render slaves. How would this affect things, and what should I do to improve it? I worry that disabling the bitmap pager may mean that my render slaves fail to render because they don't have enough memory.

[tricky]

Is memory usage the same (3GB) on both 4 and 8 core machines?
My experience is that machines with more cores reach the dynamic memory limit far sooner then machines with less cores: try setting the limit on the 8-core at 3GB.

I believe it is the same. How would I enable a different limit on some of the machines and not others? I though this variable was set by the scene that is sent to backburner.

[trick]

I believe it is the same. How would I enable a different limit on some of the machines and not others? I though this variable was set by the scene that is sent to backburner.

You'll need to split your scene and sent jobs with different limits to the different nodes (4GB and 8GB mem)

[Morne]

You'll need to split your scene and sent jobs with different limits to the different nodes (4GB and 8GB mem)

Although that may be a temp solution, it all sounds a bit backwards to me, and there must be another solution

[tricky]

This is the 3rd time I have written this - I keep pressing 'reply' instead of 'post quick reply'! It drives me mad.

Anyway, in order to get my slaves to have bitmap pager enabled, I copied the max.ini file from my local quad-core workstation to each of the slaves, both 4-core and 8-core. Is there something in the ini file which tells max the machine only has so many cores???

[Donald2B]

At one point there was a Maxscript command to set the number of CPUs to render with. This would get baked into the file. I don't remember why people were doing this, but the following command would basically reset that to use all available CPU/cores.

Just type the following into the maxscript window and hit enter. It's a long shot, but who knows!

renderers.current.system_numThreads=0