Render Speed issue with Threadripper 16 core and 32 core.

[nickmilitello]

Hello all, So doing some test with v3 and Cinema 4d on my render farm.  I did a medium complex scene with 40 passes and denoising on, no camera move so every computer is rendering the exact same image.  Doing a 700 frame test.  I am getting some very odd render times.
My render farm is a few 22 core 2.1 Xeons, 2 x 32 (OC to 4.0) threadrippers, 2x16 core (OC 4.0) Threadripper, a Dual core 2.6 14 Core Xeon and some other random PCs.
First:
Cinebenchs:
32 core - about 5900
16 Core - 3300
2.1 22 Core Xeons - about 2600
Dual 2.6 14c Xeon - 4650

Now when I render a corona scene on the farm, these are the results (average time)
32c TR - 18 min
16c TR -  20 min
2.1 22 Xeon - 20min
Dual 14c Xeon - 21-22 min

The render pass times also slightly vary at around 20 sec a pass.
Also I checked and on all machines it seems that all cores are pegged at 100% during render

So I guess my big question is what is going on?  I know that I probably wouldn't get a full 200% better performance in the 32c vs the 16c, or the Dual Xeon vs single Xeon, but this definitely seems wrong.  Any thoughts on why the render times are about the same. 

[sprayer]

It seams Xeons also have the same issue. Can you run corona 1.3 benchmark too?

[nickmilitello]

Benchmark Scores. (all have 32gb ram DDR4 2133) also put Cinebench Score there as well for reference. 

2990wx (OC 4.0) X3999 STRIX E GAMING---- 1:15s---------------------CB 5800
1950x (OC 4.0)----------------------------------1:11s---------------------CB 3300
Dual 22c 2.1 Xeon-------------------------------0:44s---------------------CB 4600
22c 2.1 Xeon-------------------------------------1:24s---------------------CB 2600

It blows my mind that the Dual 14c is way faster than the 2990wx.  I know that Corona does rely on RAM and other resources, but can anyone explain how the scores are so far off. 

[sprayer]

It seams all results are fine except 2990wx
In which slots you insert Ram modules? Did you check recommendation image in manual?
https://imagizer.imageshack.com/img921/9003/tsXTIN.jpg

And yes your RAM working at slowest frequency, do not know what modules you are using but faster ram gives better performance, amd cpu more dependable of ram frequency than intel cpu

[nickmilitello]

The manual Saadi to use A1 and C1.  Ordered 3600 ddr4 ram today to test. Will see if that helps. Thanks for the info

[diffuus]

If you use A1 and C1, does it mean u are using only 2 sticks of RAM?
In that case, the threadripper performance is cut down a whole lot because it requires at least 4 sticks of RAM to work at full potential. (It's a quad channel design)
So i would advice to test with 4 or 8 sticks of the same fast RAM.

[nickmilitello]

I was just thinking that I should try 4 sticks instead of the 2.  Makes sense if it the same render speed as the 16core which is 2 channel.  Thanks for the info all.  Ram should be here today, will report back

[urbanite]

have a look.

feature=youtu.be

[Juraj]

Don't over-quote that regression video :- ). So far it helped with two particular benchmarks and did opposite in others.

But foremost, it's not necessary for Corona (While 3dsMax itself does benefit from local dynamic mode as it's single-core process).

With 2990WX, you absolutely do need quad-channel memory at 2933MHz running properly.

[nickmilitello]

It was 4 dimms...  put 4 3200 DDR4 instead of the 2 2133 and it went from a 1:20 benchmark to 0:37!  Im sure the ram speed had something to do with it as well.  Thanks all very much for the info!!

[urbanite]

Don't over-quote that regression video :- ). So far it helped with two particular benchmarks and did opposite in others.

But foremost, it's not necessary for Corona (While 3dsMax itself does benefit from local dynamic mode as it's single-core process).

With 2990WX, you absolutely do need quad-channel memory at 2933MHz running properly.

Good to hear that Corona is save Juraj:)