Performance Issue on new workstation Xeon Gold

[ossiantove]

Hello, I am experiencing a really strange performance issue on my new workstation. It starts at the renderer performance level I expect in comparison to my old workstation, around 275% faster. But after a few minutes, it drops to a fraction of where it started, about 27% of the render performance of the old workstation.

• In activity monitor, I see no signs of thermal throttling. The CPUs are at 2.9 GHz all the time. (Or maybe thermal throttling doesn't appear in activity monitor?)
• The CPUs are all on 100% workload
• The CPUs all have sufficient power supply.
• The CPU temperatures start at 50° celcius.
• The temperatures increase to 69°, but performance start dropping after only a few minutes and CPUs are not even above 55° when performance level starts to drop.
• When looking at rays/s it starts at 3.5 million, drops to 380 000 after about 4 minutes.
• Old workstation starts at 1.38 million and stays there throughout the whole render time.
• After 1 hour and 24 minutes, the new workstations made 6 passes, the old one 16!
• It show similar results on ART render. Hardware related?


Specs on new workstation:
• Asus Sage C621 Motherboard. 2x LGA 3647 Socket.
• Dual Xeon Gold 6142, 16 cores 2.6 GHz.
• 128GB 2667 MHz ECC RAM
• GTX 1080ti
• 2TB Samsung Evo Pro
• Windows 10
• 3ds max 2019
• Corona Renderer 1.7 hotfix 4

Specs on old workstation:
• Mac Pro late 2013 (Darth Vader's Coffee Machine)
• Single Xeon E5-2697v2, 12 cores, 2.66 GHz.
• 64GB ECC RAM
• AMD FirePro D700
• 1TB PCIe Flash
• Windows 10
• 3ds max 2019
• Corona Renderer 1.7 hotfix 4


Any suggestions?

[Juraj]

Not even throttling would cause such slowdown.

What about other engines outsides of Corona/Art ? Does Vray do it ?
And some looping CPU benchmark outside of 3dsMax ? You can even try exporting scene to Corona Standalone to isolate 3dsMax influence.

Very curious about this.

[ossiantove]

Thank you for your reply. I will try with an exported scene to Corona Standalone and then I will also try a scene with blender as soon as I can, since I don't have Vray. Will be back with results tomorrow.

[Bzuco]

Can you confirm that when it slows down, the 3dsmax.exe process is still showing 99 or 100% cpu usage in process tab?

[maru]

Do you have Windows, 3ds Max, drivers and BIOS updated to the latest versions? These things can sometimes do crazy things. Another thing to check is the currently selected power plan in Windows and power management in your motherboard drivers. Just make sure you have some extreme performance mode enabled and try again.

[alexyork]

Please keep us updated on this - we're looking into those chips, so this is interesting to hear about. I hope you find a solution!

As Maru said power management would be the most likely culprit I guess.

Would be good to test using Cinebench or any other longer-term tests like Intel Burn Test etc. It would be most interesting if it's related to a particular instruction set that's being used...

[ossiantove]

Webhallen (they built this machine) are looking in to this issue at the moment and they are reporting back to me. The blender test did show similar results. I was also thinking about relation to a certain instrucion set, but it does not seem to be the case. The last test they did recently was to test the same blender scene but with the fans set to run at 100% and it actually helped. Now they are testing with Corona again and I am waiting to get results from them. They are also in contact with intel, since this seem to be related to temperature in the end. I will be back with more information soon.

[maru]

It seems that some Xeon CPUs start throttling heavily if the temperature exceeds ~74 C. According to our conversation over the support it might be the case here. Since it happens in Blender and other apps too, it most likely does not have anything to do with Corona.

[Juraj]

Xeon Gold 6142 is 150W CPU multiplied by two it is some heat, but nothing too serious to cause throttle.

What build is this ? Which CPU coolers are on ?

Could be incorrectly behaving turbo.

[hrvojezg00]

There are no good coolers atm (noctua to release this month both square and narrow socket). We are having the same problems with temps too, 200W cpu`s are jumping t 82-83 C in no time. Currently cooling with Dynatron cooler with alot of Noctua fans all around to help it cool down. Corsair don`t plan on delivering 3647 socket all in ones too. 

[Vuk]

Anyone got some tips on whta cooler to use for this socket for 24c Dual Intel Xeon Platinum 8160? I found this one >SUPERMICRO 2U Active CPU Heat Sink Socket v2 LGA3647-0 but not sure if it can keep up with the CPU?

Here is a link of the cooler:
https://ipon.hu/shop/termek/supermicro-2u-active-cpu-heat-sink-socket-v2-lga36470/1665364

*EDIT

Actually found that Noctua released their new LGA3647 coolers so they are already available for purchase. Has anyone tried them yet?

[arqrenderz]

ossiantove , have you looked to energy plan settings on windows?
By deafult those are on energy saving mode, have you solved the issue??

[hrvojezg00]

Anyone got some tips on whta cooler to use for this socket for 24c Dual Intel Xeon Platinum 8160? I found this one >SUPERMICRO 2U Active CPU Heat Sink Socket v2 LGA3647-0 but not sure if it can keep up with the CPU?

Here is a link of the cooler:
https://ipon.hu/shop/termek/supermicro-2u-active-cpu-heat-sink-socket-v2-lga36470/1665364

*EDIT

Actually found that Noctua released their new LGA3647 coolers so they are already available for purchase. Has anyone tried them yet?

We have all Platinums on Noctua, all cpu`s lower then 65 C, highly reccomend!

H