Rendering times much slower on Xeon?

[rich0550]

Hi
the Xeon gets 41,000 score on vray benchmark, and the other machine a 9700k gets 9,000 score. The 9700k is getting sometimes renders in half the time of the Xeon, that rendering VDB files. The 9700 has faster RAM and cores but the Xeon system is only a few years old. Is there a reason why rendering VDB files might go so much faster on the newer pc?

[Svetlin.Nikolov]

Hey,

There are many models of Xeons, can you share which one is yours?

Cheers!

[Paul Oblomov]

Vray when rendering phx uses some sort of instructions, that lead to cpu speed drop. I asked some time ago, whether this was famous avx - chaos said they not using avx. I swtched to 3960x and while rendering phx smoke-temp cpu temp can easily go way ahead of comfortable 75c to smth like 85c. I can't say what instructions used, but I still suspect avx, coz zen2 avx blocks are full speed, they don't have avx offset, like intel and avx are very freaking cpu melting instructions. I even have to lock my cpu to 4ghz, so it wont boil I don't like my cpu to get to its top temp.

[Svetlin.Nikolov]

I'm afraid if it was that simple, we would have changed it long ago and sped it up... rich0550 said it was during rendering VDB files, so I suppose it's not about the particle shader but just grid volumes. So we gotta know the grid sizes, number of empty voxels, what channels are in there, Phoenix logs from C:\PhoenixFD or V-Ray render logs will be very helpful, the scene would be even more helpful, and of course the full hardware specs. Otherwise, we could just speculate.

[rich0550]

Sorry not to be very descriptive my Xeon gets a Vray 41,674 score versus the 9700k which gets Vray 9635 score. I will send the VDB file to you guys to check it out. Its really weird when running backburner render the better machine is rendering at about 1.5 minutes a frame versus the other at around 2.5 minutes a frame, suddenly when the VDB file starts to load and is rendering, the Xeon goes to 17 minutes a frame, the 9700k goes to 5 minutes a frame.

Phoenix is not installed on the 9700k, only on the other Xeon. The VDB file was not created with Phoenix it was purchase from turbosquid he made it with Houdini. The settings I put pretty low fast render. The scene is too large to practically send, also its under NDA. I think i tested this with a backup hard drive that only had a fresh install never used, and had no Phoenix on it and saw similar numbers. Also I made a test not using network rendering and rendered only locally on both machines to compare too.

[Svetlin.Nikolov]

Thank you! The workflow for determining what's the culprit would be start turning off options that do extra work and compare after each change - this is what we would do here as well. First off, if the cache is an explosion, would be to set Fire's Based On to Disabled. Then turning on or off the Volume Light Cache. Turning off all V-Ray render elements as well. Then also unchecking Optimize Big Volumetric Grids. If neither of these turns around the time balance between the two machines, next steps would be to switch from Volumetric to Volumetric Geometry mode or vice versa. Next would also be turning on or off the Probabilistic Volumetrics option in the Phoenix Atmosphere. At this point something should change the balance between the two machines and would narrow down on the cause for the difference..

[Paul Oblomov]

Btw, yesterday my CPU temp gone 95c. Fire lights set to 80% killed my CPU So my idea is to look at the temps on both and CPU ghz at rendertime.

[Svetlin.Nikolov]

Not sure how we could help?

[Svetlin.Nikolov]

For us to be able to identify and issue and figure out a solution, we gotta have some minimum amount of information to work with. CPUs under load heat up, this is natural. When you use software that should squeeze the maximum out of the hardware, this is what is expected to happen. Is there something wrong with the cooling of your hardware?

[Paul Oblomov]

To figure out, which instructions lead to such behavior - you just have to use same hardware as me using (zen2) on a daily basis. When I asked about why my Xeon ES was lowering its CPU speed to a much lower level, than production unit - we came to nothing, since AVX wan't used (as we found out, if I remember) and to profile things up, there was a need to have such CPU and to have spare time to spend on this

So, I personally, don't need help, as I understand, that Chaos won't mod their code, to lower zen2 temps. But, maybe my case will help topic starter to figure out what is happening with his Xeon, since AVX offset (turned on by default) can lower its performance quite a bit in theory.

P.S. Btw, I knew, that my Xeon ES lowering CPU speed much aggressively, than prod. unit. But, was curious, if Vray was using AVX...not using as I was told. I'm not a software engineer, so I just can only guess. Also, my CPU cooling is smth, that most of the zen2 users are using. Just a small note to Chaosgroup, that Vray can sometimes push zen2 to its limits. linpack is not that agressive

[rich0550]

Thank you! The workflow for determining what's the culprit would be start turning off options that do extra work and compare after each change - this is what we would do here as well. First off, if the cache is an explosion, would be to set Fire's Based On to Disabled. Then turning on or off the Volume Light Cache. Turning off all V-Ray render elements as well. Then also unchecking Optimize Big Volumetric Grids. If neither of these turns around the time balance between the two machines, next steps would be to switch from Volumetric to Volumetric Geometry mode or vice versa. Next would also be turning on or off the Probabilistic Volumetrics option in the Phoenix Atmosphere. At this point something should change the balance between the two machines and would narrow down on the cause for the difference..

I'll try these things and report back.