Author Topic: dual CPU issues?  (Read 8930 times)

2021-07-31, 22:11:32

doillo

  • Active Users
  • **
  • Posts: 16
    • View Profile
Hello,

I answer a C4D post cos I have similar issue than an user posted, but I have not noticed the post was in c4d forum, and my issue is at 3dsmax.
Im going to open new threat here:

I have 2 computers one with Asus z10pe-d16 ws motherboard with 2x e5-2696v4 (88 threats). And my other computer is a e5-2678v3 (24 threats)

At several benchmarks, the dual system is more than 3 times powerfull than the single cpu computer (cinebenchR15, cinebenchR20, cinebenchR23, coronabenchmark 1.3...)

But when I'm rendering with corona DR (3dsmax 2021.3 and corona 6.2), dual system makes only 1.1~1.2 times more passes than the single cpu system... How it can be possible?
At first time, I though I had LAN issues, but doesn't matter wich computer is slaving or is the main. The result is the same. Also, if I put the same scene at the same time in each computer to render. Dual system is only 1.1~1.2 faster than the single one... So, what is going on?!

When dual system is rendering, 88 threats are full load (100%) all the time without any drop... So I guess corona is not handling properly dual CPUs computers. I guess, corona is rendering with both CPUs in "PARALLEL" instead of in "SERIAL" (adding passes). Cos something is missing or failing at corona programing code... And that makes sense why if I deactivate one cpu in the dual system, the performance rise a bit... Cos again, is just a guess, I think both cpus are fighting to render the same pass at same time, reading the same info at RAM at same time. But at the end, only 1 CPU passes are counting...

I made some test with vray also, and results are more or less expected (dual system renders 3 times more buckets than single cpu system)

So, anyone can send some light to this problem?
« Last Edit: 2021-08-02, 00:12:44 by doillo »

2021-08-02, 13:33:19
Reply #1

TomG

  • Administrator
  • Active Users
  • *****
  • Posts: 5434
    • View Profile
Ooops you were saying it is NOT C4D
Tom Grimes | chaos-corona.com
Product Manager | contact us

2021-08-02, 14:45:51
Reply #2

maru

  • Corona Team
  • Active Users
  • ****
  • Posts: 12708
  • Marcin
    • View Profile
What scores are your computers getting in the Corona Benchmark? https://corona-renderer.com/benchmark
Marcin Miodek | chaos-corona.com
3D Support Team Lead - Corona | contact us

2021-08-02, 19:32:13
Reply #3

doillo

  • Active Users
  • **
  • Posts: 16
    • View Profile
What scores are your computers getting in the Corona Benchmark? https://corona-renderer.com/benchmark

Hi, dual CPU system score are VERY far from the single CPU system:

2x E5-2696v4: 32 seconds

1x E5-5678v3: 1 minute and 51 seconds

That's why I am wondering something is wrong here....

« Last Edit: 2021-08-03, 01:31:44 by doillo »

2021-08-03, 02:38:33
Reply #4

doillo

  • Active Users
  • **
  • Posts: 16
    • View Profile
I am doing a fresh format and reinstalling everything on again... Let's see tomorrow doing test how it is going on.

Meanwhile, I have done Vray benchmark too, and dual system is clearly more than 3 times faster and powerfull than the single one... I don't know why this doesn't translate to 3 times faster renderings in corona and 3dsmax (or at least 2 times faster, but almost the same performance is strange).








2021-08-03, 13:59:14
Reply #5

doillo

  • Active Users
  • **
  • Posts: 16
    • View Profile
Hi again,

More or less same results with new fresh win10 installed... Seeking into the c4d forum, where someone said textures is making this issue I made this test:
Main computer (where render was launched) is the single CPU one, and the DR computer is the dual CPU one.



I made a simple scene: Teapot in a room with material/textures.

Render time: 16 minutes and 36 seconds
Single cpu computer made 95 passes
Dual cpu computer made 121 passes
Almost same performnace: dual cpu computer made only ~1.3 times more passes than single cpu computer



Same scene, but overwriting material. No maps, no textures, only simple white coronamaterial.

Render time: 3 minutes and 11 seconds
Single cpu computer made 49passes
Dual cpu computer made 159 passes
Expected difference in performance: dual cpu computer made ~3 times more passes than single cpu computer


I will make more test, but something is wrong at corona handling textures with dual CPU system...


« Last Edit: 2021-08-03, 14:06:10 by doillo »

2021-08-03, 15:02:12
Reply #6

maru

  • Corona Team
  • Active Users
  • ****
  • Posts: 12708
  • Marcin
    • View Profile
Can you share your example scene?

It looks like your Corona Benchmark results are correct.

So the issue might be in:
- newer Corona versions
- 3ds Max
- specific scene
- plugins
-...

Marcin Miodek | chaos-corona.com
3D Support Team Lead - Corona | contact us

2021-08-03, 16:10:22
Reply #7

doillo

  • Active Users
  • **
  • Posts: 16
    • View Profile
Can you share your example scene?
Sure, tell me an e-mail and I will send to you

It looks like your Corona Benchmark results are correct.
Do you mean the coronabenchmark 1.3 or the test I did?

- newer Corona versions
Yes, I am going to try with the new corona 7 version, lets see...

- 3ds Max
I can try others max version... Its a pain, but I can try.

- specific scene
Yes, I will make more test with others scenes.

- plugins
I have only max and corona installed... So I don't think is any plugin doing something.


For me would be interesting if others dual CPU system users can make some test... but I know it going to be difficult.




2021-08-03, 16:15:13
Reply #8

maru

  • Corona Team
  • Active Users
  • ****
  • Posts: 12708
  • Marcin
    • View Profile
Can you share your example scene?
Sure, tell me an e-mail and I will send to you
The uploader is in my signature, here is the link: https://corona-renderer.com/upload
After the upload, please let me know what was your file name and which upload method you used.

Quote
It looks like your Corona Benchmark results are correct.
Do you mean the coronabenchmark 1.3 or the test I did?
I meant the Corona 1.3 benchmark (this: https://corona-renderer.com/benchmark ).

Quote
For me would be interesting if others dual CPU system users can make some test... but I know it going to be difficult.
We have some dual CPU systems in the office, we can do some testing too.
Marcin Miodek | chaos-corona.com
3D Support Team Lead - Corona | contact us

2021-08-03, 22:15:25
Reply #9

doillo

  • Active Users
  • **
  • Posts: 16
    • View Profile
Quote
The uploader is in my signature, here is the link: https://corona-renderer.com/upload
After the upload, please let me know what was your file name and which upload method you used.
Hi again, I sent the archive scene (corona_test_01.zip)

Just to be clear, is not about DR. I just mention DR cos is an easy way to compare 2 computers.

I made another test with the same scene (resolution: 1280x720p ; noise threshold 1.5%). Rendering only with each computer in local and I have this results:


Single cpu computer (e5-2678v3)
   OverWrited with white material(without textures):
  • render time: 11 min and 57 secs
  • rays/s total: 8.33M
  Original with textures at materials
  • render time: 35 min and 01 secs
  • rays/s total: 6.60M
We drop to 0.79 times in rays/s and 2.93 times slower with textures.

Dual cpu computer (2x e5-2696v4)
   OverWrited with white material(without textures):
  • render time: 3 min and 30 secs
  • rays/s total: 27.40M
  Original with textures at materials
  • render time: 28 min and 52 secs
  • rays/s total: 8.20M
We drop to 0.3 times in rays/s and 8.25 times slower with textures.



The drop in dual system is SO heavy, and I am almost certain, that drop is because an entire CPU of two is doing redundant work.
If you make simple calculations, the dual system should drop from 27.4M Rays/s to ~20.0M Rays/s and should take ~10 minutes with textures...

2021-08-04, 10:44:57
Reply #10

maru

  • Corona Team
  • Active Users
  • ****
  • Posts: 12708
  • Marcin
    • View Profile
I can't find the file corona_test_01.zip. Which upload method did you use? Dropbox or classic?
Can you try uploading again or use some other uploader like Wetransfer and send me the link via pm?

Thanks
Marcin Miodek | chaos-corona.com
3D Support Team Lead - Corona | contact us

2021-08-04, 16:15:25
Reply #11

maru

  • Corona Team
  • Active Users
  • ****
  • Posts: 12708
  • Marcin
    • View Profile
Got the scene. Logged for further testing.

(Report ID=CRMAX-866)
Marcin Miodek | chaos-corona.com
3D Support Team Lead - Corona | contact us

2021-08-05, 13:47:18
Reply #12

rowmanns

  • Corona Team
  • Active Users
  • ****
  • Posts: 1892
  • Corona for 3ds Max QA Team
    • View Profile
Hi,

I would suggest using a time or pass limit for testing performance, unfortunately the noise threshold is not a good benchmark for testing the performance.

We will also investigate here :)

Thanks,

Rowan
Please read this before reporting bugs: How to report issues to us!
Send me your scene!

2021-08-05, 16:19:39
Reply #13

doillo

  • Active Users
  • **
  • Posts: 16
    • View Profile
Hi,

I would suggest using a time or pass limit for testing performance, unfortunately the noise threshold is not a good benchmark for testing the performance.

Yes, I made extra tests with time timit instead of noise limit. And I sent it to maru.

Also, I made a video showing the performance drop suffered by my dual system when different maps are applied into a material... It is like only coronabitmap is allowed to works fine in a dual system. When you apply a simple output or colorcorrection between the coronabitmap and the material, performance drops so heavy. It drops from ~30M rays/s to 8~10M rays/s... (when at single CPU system, the drop is from 8M rays/s to 6.5M rays/s)


« Last Edit: 2021-08-05, 20:32:16 by doillo »

2021-08-12, 09:46:29
Reply #14

maru

  • Corona Team
  • Active Users
  • ****
  • Posts: 12708
  • Marcin
    • View Profile
@Doillo - unfortunately so far we were no able to reproduce this issue, even on a dual CPU machine. We are getting the expected performance.

Can you please:
- Make sure that you have all Windows Updates installed (you may need to restart your computer multiple times and check for new updates again, then repeat)
- Make sure that you have all the drivers up to date - especially the "system" drivers such as the chipset driver
- Update your BIOS to the newest available version
Marcin Miodek | chaos-corona.com
3D Support Team Lead - Corona | contact us

2021-08-13, 22:11:14
Reply #15

doillo

  • Active Users
  • **
  • Posts: 16
    • View Profile
@Doillo - unfortunately so far we were no able to reproduce this issue, even on a dual CPU machine. We are getting the expected performance.

Can you please:
- Make sure that you have all Windows Updates installed (you may need to restart your computer multiple times and check for new updates again, then repeat)
- Make sure that you have all the drivers up to date - especially the "system" drivers such as the chipset driver
- Update your BIOS to the newest available version

Wow, that sounds disapointed for me... I have everything updated. (drivers, windows updates (the optional ones), the bios, everything). My first thougts were to think that was my system fault, but as you could see, my coronabenchmarks were fine (cinebench and vray also was good). Moreover, other render engines like vray works fine as expected.
Moreover, as you saw, the problem were at textures at corona material. Without textures, everything was runing fine. I am building another dual system, and in a few weeks I will be able to do more test with new dual system.

I was thinking also about the ECC RAM memory (required for my mother board). Could be the problem there? I have 4x16 Gb sticks ECC Ram.

2021-09-02, 10:36:43
Reply #16

maru

  • Corona Team
  • Active Users
  • ****
  • Posts: 12708
  • Marcin
    • View Profile
We did some more tests on a dual CPU machine (picked a pretty old CPU on purpose - 2x E5-2690 "v0") and we are still getting the expected performance in both your scene and the Corona Benchmark.
I am really not sure what else we could try. I am afraid only reproducing this issue on our end would give us some data.
Marcin Miodek | chaos-corona.com
3D Support Team Lead - Corona | contact us

2021-09-09, 23:13:00
Reply #17

doillo

  • Active Users
  • **
  • Posts: 16
    • View Profile
We did some more tests on a dual CPU machine (picked a pretty old CPU on purpose - 2x E5-2690 "v0") and we are still getting the expected performance in both your scene and the Corona Benchmark.
I am really not sure what else we could try. I am afraid only reproducing this issue on our end would give us some data.

Hello again, well I think I know where is the issue from. I search a lot and I find more ppl having this issue. Not only with corona, with vray and other softwares also. And not only with dual intel xeons, also with EPYC system, and TR 3990x.

Windows 10 "can not handle more than 64 threats". Well, it can, but when you have more than 64 threats, w10 splitt those into 2 "numa cores". And I think 3dsmax can not handdle those "numa cores"... Thats why, when I use 3dsmax maps, performance drops HUGE and an entire CPU "dissapears". It seems corona renderer is well implemented, and coronabitmap works fine (but not all of ur maps, cos triplanar also drops heavy my system) with those "2 numa cores" but 3dsmax not.

As I said, all of this is my summ and my thoughts, reading posts, making my owns test, etc.

It would be nice if you could try the test I show you in the video with a system with more than 64 threats.

2021-09-16, 13:59:00
Reply #18

maru

  • Corona Team
  • Active Users
  • ****
  • Posts: 12708
  • Marcin
    • View Profile
@doillo - just one more request:

Could you please:
- download the attached archive
- open the two included scenes and render both of them on the problematic dual-cpu computer (do not change any settings)
- save the render output (there is a 1 minute time limit set)
- share your results with us

Thanks in advance!
Marcin Miodek | chaos-corona.com
3D Support Team Lead - Corona | contact us

2021-09-16, 15:47:21
Reply #19

alexyork

  • Active Users
  • **
  • Posts: 699
  • Partner at Recent Spaces
    • View Profile
    • RECENT SPACES
We are also using some 128-thread machines here in the studio. Not particularly noticing the same issue as OP but thought I'd contribute my results too from my own workstation in case helpful.

Cheers.
Alex York
Partner
RECENT SPACES
recentspaces.com

2021-09-22, 15:55:38
Reply #20

maru

  • Corona Team
  • Active Users
  • ****
  • Posts: 12708
  • Marcin
    • View Profile
@doillo ^could you please try that as well?
Thanks in advance!
Marcin Miodek | chaos-corona.com
3D Support Team Lead - Corona | contact us

2021-09-24, 10:12:49
Reply #21

doillo

  • Active Users
  • **
  • Posts: 16
    • View Profile
Hi,

Sorry if I am late answering but I am a bit bussy nowdays.

I will send you the test on monday

2021-09-24, 10:45:06
Reply #22

hrvojezg00

  • Active Users
  • **
  • Posts: 270
    • View Profile
    • www.as-soba.com
I`m imterested in this as well as we are having alot of issues with it lately.

H

2021-10-12, 13:05:32
Reply #23

maru

  • Corona Team
  • Active Users
  • ****
  • Posts: 12708
  • Marcin
    • View Profile
Hi,

Sorry if I am late answering but I am a bit bussy nowdays.

I will send you the test on monday

Hi Doillo, I shared a simple test scene here https://forum.corona-renderer.com/index.php?action=dlattach;topic=33941.0;attach=154014
Could you please try follow the instructions from that post and share your results with me?
Thank you!
Marcin Miodek | chaos-corona.com
3D Support Team Lead - Corona | contact us

2021-10-12, 13:23:32
Reply #24

maru

  • Corona Team
  • Active Users
  • ****
  • Posts: 12708
  • Marcin
    • View Profile
I`m imterested in this as well as we are having alot of issues with it lately.

H

I have just created a support ticket in our system on your behalf. Please check your email and it would be great if we could discuss this further.
Thank you in advance!
Marcin Miodek | chaos-corona.com
3D Support Team Lead - Corona | contact us

2022-03-01, 22:35:10
Reply #25

doillo

  • Active Users
  • **
  • Posts: 16
    • View Profile
Hello again,

Srr if I am reviving a "dead topic", but after 4 months I have more data to share.

4 months ago, I thought the problem was with my "exotic CPUs", because they are the OEM version (e5-2696v4). I thought that was the problem... So I decided to buy 2 original e5-2699v4 and change them... And surprise... I have the same problem :/

I found a post on the official Autodesk forum (they're old, but I'm pretty sure they haven't fixed it yet) that said 3dsmax can NOT properly handle +64 threats.
In addition, Windows 10 also has some problems with +64 threats... for example, I cannot "set affinity" in task manager and manage the number of threats for a certain application (Windows denies this action if it has more than 64 threats) And TR 3990x is known to have issues with its performance: https://www.anandtech.com/show/15483/amd-threadripper-3990x-review/3


And on top of this, corona is NOT working properly too. There are maps that SINK render performance when I use them (like triplanarTex) or native 3dsmax output or bitmap (and I think there will be more maps for sure).
I'm using corona 6/7 and I don't know if in corona 8 any of this is fixed or at least in a "fix schedule". I know you can not fix 3dsmax maps, but at least, try to fix your own maps like tryplanarTex, and check the others.

I'm still shocked that it took me like a whole year to know what was going on here. Finding almost 0 information on this. And I can not believe nobody is complaining of this.


There is any alternative? Can I render in other OS? do you have Linux render node soft?



2023-02-07, 08:23:56
Reply #26

simtub

  • Active Users
  • **
  • Posts: 13
    • View Profile
I've been having some issues getting Corona to run at a decent speed with our new DUAL EPYC Racks systems. Came across this thread and it has given me a bit more knowledge over the issues pertaining.

We thought it was a Windows 11 issue too and just upgraded to Windows 11 Pro Workstation Edition which handles up to 4 CPU's and 6TB Ram (We have Dual EPYC's at 64C 128T with Total Core count of 256 and 256GB Ram) But it seems Corona is still rendering slower than many of the benchmarks on out there on the Corona Benchmark Page and also CPU Utilization jumps up and down during rendering time.
 

Hello again,

Srr if I am reviving a "dead topic", but after 4 months I have more data to share.

4 months ago, I thought the problem was with my "exotic CPUs", because they are the OEM version (e5-2696v4). I thought that was the problem... So I decided to buy 2 original e5-2699v4 and change them... And surprise... I have the same problem :/

I found a post on the official Autodesk forum (they're old, but I'm pretty sure they haven't fixed it yet) that said 3dsmax can NOT properly handle +64 threats.
In addition, Windows 10 also has some problems with +64 threats... for example, I cannot "set affinity" in task manager and manage the number of threats for a certain application (Windows denies this action if it has more than 64 threats) And TR 3990x is known to have issues with its performance: https://www.anandtech.com/show/15483/amd-threadripper-3990x-review/3


And on top of this, corona is NOT working properly too. There are maps that SINK render performance when I use them (like triplanarTex) or native 3dsmax output or bitmap (and I think there will be more maps for sure).
I'm using corona 6/7 and I don't know if in corona 8 any of this is fixed or at least in a "fix schedule". I know you can not fix 3dsmax maps, but at least, try to fix your own maps like tryplanarTex, and check the others.

I'm still shocked that it took me like a whole year to know what was going on here. Finding almost 0 information on this. And I can not believe nobody is complaining of this.


There is any alternative? Can I render in other OS? do you have Linux render node soft?

2023-02-11, 05:47:24
Reply #27

simtub

  • Active Users
  • **
  • Posts: 13
    • View Profile
Some updates on our experience with a DUAL CPU system

On our Dual EPYC node, we realised that 4 x 64Gb Dimm configuration was the issue. Supposedly on a Single CPU, filling 4 Dimm slots would enable the system to run in Quad Channel but since we have 2 CPU's then 4 Dimms is just a Dual Channel Memory set up and we would need 8 Dimms for Quad Channel or 16 Dimms for 8-Channel Memory set ups. (Our school boy error when setting out the original specifications).

So it seems Corona is much more bandwidth dependent on the number of memory channels much more so than Vray and Cinebench. When we ran benchmarks for Vray and Cinebench in Dual Channel, the scores were very good and matched to other systems of similar spec. It was only Corona that had issues, Dual Channel Ram literally halved the performance on the Corona Benchmark giving us around 20seconds where as Quad Channel gave a respectable 11 seconds. I suspect 8-Channel may push the Corona benchmark to 8-9 seconds but we don't have enough Ram to test this.

Other things we considered to resolve this speed issue before switch the RAM were:

- Upgrading to Windows Pro 11 Workstation Edition (This version can run more CPU's and RAM than Windows 11 Pro but it didn't affect our case even with 128 Cores and 256 Threads)
- Turning off Virtualization (Didnt really do much)
- Turning off VBS, Core Isolation and Memory Integrity on Windows (Didn't really do much)
- Power Plan on High Performance (Both in Windows and BIOS - Not sure how this affects the system performance as yet)
- NUMA settings on BIOS (Not yet looked into it) 
- Windows Updates and latest BIOS

Hope this can help others out there configuring Dual Socket Systems.  I would be interested to know from the Corona Development team why there is such a dramatic performance difference between DUAL and QUAD/8 Channel RAM configs as Vray doesn't have this issue at all.

Cheers.

2023-04-24, 23:32:24
Reply #28

doillo

  • Active Users
  • **
  • Posts: 16
    • View Profile
Oh my god!

I just assume my RAM was working in quad channel (octo in 2 slots), using 4 sticks per cpu (8 in total). In fact, typical monitor software tells me my RAM is working in quad-channel, but I didn't try to populate all dimms (cos My boards have 16 slots!).

I will try this solution but I have to buy RAM to check it cos I don't have 16 RAM sticks...

2024-01-05, 10:16:03
Reply #29

andrew1988

  • Active Users
  • **
  • Posts: 10
    • View Profile
have you found the sollution in the end ?
I have some issue with you with my new build dual XEON 8480+

2024-01-05, 10:17:16
Reply #30

andrew1988

  • Active Users
  • **
  • Posts: 10
    • View Profile
have you found the sollution in the end ?
I have some issue with you with my new build dual XEON 8480+


Oh my god!

I just assume my RAM was working in quad channel (octo in 2 slots), using 4 sticks per cpu (8 in total). In fact, typical monitor software tells me my RAM is working in quad-channel, but I didn't try to populate all dimms (cos My boards have 16 slots!).

I will try this solution but I have to buy RAM to check it cos I don't have 16 RAM sticks...

2024-01-08, 11:46:33
Reply #31

Nejc Kilar

  • Corona Team
  • Active Users
  • ****
  • Posts: 1245
    • View Profile
    • My personal website
If it helps knowing I'm running a dual Epyc build that has 8 memory slots per CPU (16 total). Just for the fun of it I tried running the benchmarks (and one rather simple interior) in a 8 sticks vs 16 sticks scenario.

The results were for all intents and purposes identical between the two configs.

I do think that going with less than 8 sticks though would result in slower rendering - I haven't tested it out though.

Do note that each platform can be a bit different so the above might work for my 7B13 Epycs but might not for your dual Xeons.
« Last Edit: 2024-01-09, 10:35:00 by Nejc Kilar »
Nejc Kilar | chaos-corona.com
Educational Content Creator | contact us

2024-01-10, 11:59:32
Reply #32

andrew1988

  • Active Users
  • **
  • Posts: 10
    • View Profile

thanks for your reply !
yes , it have 4 sticks per socket installed , 8 in total, memory runs on quad channel

every benchmark tool runs perfect but the corona 10 benchmark
its weird



If it helps knowing I'm running a dual Epyc build that has 8 memory slots per CPU (16 total). Just for the fun of it I tried running the benchmarks (and one rather simple interior) in a 8 sticks vs 16 sticks scenario.

The results were for all intents and purposes identical between the two configs.

I do think that going with less than 8 sticks though would result in slower rendering - I haven't tested it out though.

Do note that each platform can be a bit different so the above might work for my 7B13 Epycs but might not for your dual Xeons.