Dual cpu speed problems

[Evolist_ua]

Hi guys, the 8th daily version of the Corona for the cinema has already been released, and the problem that I wrote about back in 2019 is still present (https://forum.corona-renderer.com/index.php?topic=24238.0)
The confirmation:
One cpu with no corona bitmap at materials, 9,7m rays actual

TWO CPU (not 2 core) with no corona bitmap at materials, 5,5m rays actual. One cpu is faster than two,LOL.

Two cpu with corona bitmap at materials, 16+m rays actual, this is NORMAL speed.


Are you planning to fix the render speed issue on dual-processor systems at all? If not, then at least add to the converter the ability to immediately convert all textures to the corona bitmap,because doing it manually takes a lot of time, thanks.

[maru]

We have a similar issue reported for Corona for Max, but we were not able to reproduce it on our end (it only happens to the user).

Are you using the Corona UVW Randomizer map here?
Are you using the Corona Triplanar Map here?

Can you please share the scene with us? (if you haven't yet)

Max report for our team - (Report ID=CRMAX-866)

[Evolist_ua]

We have a similar issue reported for Corona for Max, but we were not able to reproduce it on our end (it only happens to the user).

I tested in 3d max, even in 2019 everything was ok there with speed.
No  UVW Randomizer or Triplanar there, only maps at diffure,reflect\gloss, bump.

There is no need in dropping this test scene for you, created in a couple of minutes, it's just a set of simple geometric shapes, with sky and two materials, starting from 2019 with almost every assembly I create a similar test scene, and the result is always the same.

but we were not able to reproduce it on our end

Have you tested this on a DUAL PROCESSOR PC???

[TomG]

Just as a note, a scene is always handy, just in case some default render settings have been changed by a user etc. It's always useful to know we are testing EXACTLY the same thing, and not something "close to" or "similar to" or "kind of like" the scene where a user has a problem. So sharing the scene would still be useful, thanks :)

[maru]

but we were not able to reproduce it on our end

Have you tested this on a DUAL PROCESSOR PC???

We tested this ON BOTH SINGLE PROCESSOR PCS AND DUAL PROCESSOR PCS!!!

And to test it thoroughly we would need exactly the same scene that you are using so that we could compare with your results too.

Thanks for understanding.

[Evolist_ua]

We tested this ON BOTH SINGLE PROCESSOR PCS

On systems with one cpu everything is ok, it can be seen even by the fact that when I render with one processor (in the cinema settings I limit the number of threads so that the 2nd processor does not use) then everything is fine.

Created two test scenes, with and without bitmaps,  uploaded as 1629289008_scenes-for-dual-cpu.zip. Hope this helps, thanks.

[mmarcotic]

Hello,

thanks for reporting, we have logged it internally at (Internal ID=743369807)

Jan

[prince_jr]

tested a simple scene...

1. rendered with texture directly in texture slot, inkl. uvw-randomizer
2. rendered with corona bitmap in texture slot, inkl. uvw-randomizer > faster, more rays, less noise, but uvw randomizer seems not to work

> mac osx mojave, c4d r23, corona v6 (hotfix 2)
   no idea how it's behaving in v7 or v8 daily builds

[mmarcotic]

Hello,

can you also include this scene with textures included? I would like to rule out all input before creating a bug-report.

Additionally, this thread concerns a very different issue.

Thanks,
Jan

[Evolist_ua]

Hello,
For the guys of the corona team, this will probably be funny... But I just tested the 10th corona for c4d, and the problem with systems on 2 CPU remained...
Judging by the fact that I reported this problem back in 2019 (https://forum.corona-renderer.com/index.php?topic=24238.0), this bug is already more than 4 years old ... Do you have plans to fix this?

[Stefan-L]

i have also a few dual xeons in use, they not show any slow down for me, can you explain what you experience?
v10 runs fin here as expected, so it might ot be some general with 2 cpus.l

[Evolist_ua]

i have also a few dual xeons in use, they not show any slow down for me, can you explain what you experience?
v10 runs fin here as expected, so it might ot be some general with 2 cpus.l

Try to make a test scene (like at the beginning of the topic) and check the rendering speed when the textures are loaded (diffuse, reflect, gloss, bump) with the corona bitmap and without the corona bitmap.

[Stefan-L]

i have threadrippers with 1 CPU (32/64 cores) and dual xeons (56 cores),
they all behave as expected from speed as said.(also i test my machines always with Cinebanch and Coronabench.

thats why i asked

speed greatly depends on how file search path are set, disc speed/network speed of the machine and so on, also number of cpu cores cant be compared 1:1.

[Evolist_ua]

i have threadrippers with 1 CPU (32/64 cores) and dual xeons (56 cores),
they all behave as expected from speed as said.(also i test my machines always with Cinebanch and Coronabench.

thats why i asked

speed greatly depends on how file search path are set, disc speed/network speed of the machine and so on, also number of cpu cores cant be compared 1:1.

Please re-read this topic more carefully, also the topic to which I gave a link.

If you do not use a corona bitmap, then one processor renders faster than two (In the settings of the cinema 4d, there is a limit on the number of threads. I have a 2x2698v4 system. I set the limit to 40 threads). If the textures are loaded with the corona bitmap, then the rendering speed on two processors is ok.

But manually adding corona bitmap for each texture takes time... Why don't they make a script for that? Or that the textures would immediately work as if through a corona bitmap? There are also problems with the corona bitmap itself, about which many topics have already been created.

"also number of cpu cores can't be compared 1:1."
That difference is almost 2 times, when using two processors, in comparison with one. If checked in render without textures, or when all textures are through corona bitmap.

It is incorrect to compare AMD and Intel in this case.

[Stefan-L]

well i think i did read the thread.

i do not use corona bits , or only rarely. speed is ok with our without here.
i have quite some Corona render experience, and it works very well for most things with the classic c4d bitmap.

modern systems (also single cpu threadrippers) can be 2x or even 3x and more faster than old dual xeon as you or i have,
best compare the cinebench values for example of both machines (to have a neutral comparance), so there a 2x speed difference is expected.

EDIT: on cinebench 15, the xeon you have a score of 2940 points, where the single threadripper has, dependent on model scores between 6000 and 10400 points, and intel i9 13900 has still more than 6000 points, more than double the speed as the dual xeons ( and i think it is valid to compare amd and xeons for render tasks, it is all about speed, amd cpus are just most times faster today for rendering)

this is due the progress modern cpus have made last years.

[Evolist_ua]

well i think i did read the thread.

i do not use corona bits , or only rarely. speed is ok with our without here.
i have quite some Corona render experience, and it works very well for most things with the classic c4d bitmap.

modern systems (also single cpu threadrippers) can be 2x or even 3x and more faster than old dual xeon as you or i have,
best compare the cinebench values for example of both machines (to have a neutral comparance), so there a 2x speed difference is expected.

EDIT: on cinebench 15, the xeon you have a score of 2940 points, where the single threadripper has, dependent on model scores between 6000 and 10400 points, and intel i9 13900 has still more than 6000 points, more than double the speed as the dual xeons ( and i think it is valid to compare amd and xeons for render tasks, it is all about speed, amd cpus are just most times faster today for rendering)

this is due the progress modern cpus have made last years.

Why are you writing here about the difference between old xeons and new other processors? Topic about it?

I have a 5950x system, it is in terms of rendering speed like my system with two xenons. I use the xeon system as a node, but I can only use 50% of its capabilities.
Dual processor systems can give a lot of performance, and I have a desire to build a few more nodes, but this bug severely limits it.


"i have quite some Corona render experience, and it works very well for most things with the classic c4d bitmap."

I didn’t just write that the bug was described back in 2019 ... But it’s still not working.

[Evolist_ua]

"well i think i did read the thread."

If you do not understand the problem, then you did not look carefully ((

[maru]

Please report this as a support ticket here: https://support.chaos.com/hc/en-us/requests/new
Please attach any scene where you are able to reproduce the problem and add information about the version of Corona and Cinema 4D you are using, and the hardware you are using.

[Evolist_ua]

Please report this as a support ticket here: https://support.chaos.com/hc/en-us/requests/new
Please attach any scene where you are able to reproduce the problem and add information about the version of Corona and Cinema 4D you are using, and the hardware you are using.

Hi,
I sent a report and a scene back in 2021... nothing has changed during this time.
Someone from the corona team wrote in this thread

"thanks for reporting, we have logged it internally at (Internal ID=743369807)"

[maru]

Thanks, we do have it reported indeed.
Which versions of Corona and C4D are you using?
Also, which OS and version? 
And lastly, what are your exact hardware specs? What CPU model, how much RAM, what RAM configuration? (2 RAM sticks or more) Are you 100% sure that your RAM is installed correctly according to the motherboard manual?

I am sorry if you have already answered those questions in the past, but if you could do it once again, that would be great.

[Evolist_ua]

Thanks, we do have it reported indeed.
Which versions of Corona and C4D are you using?
Also, which OS and version? 
And lastly, what are your exact hardware specs? What CPU model, how much RAM, what RAM configuration? (2 RAM sticks or more) Are you 100% sure that your RAM is installed correctly according to the motherboard manual?

I am sorry if you have already answered those questions in the past, but if you could do it once again, that would be great.

Corona 3-10 (yes, all versions from 3 to last 10)
C4D R16-R23
Windows 7-10-11 different versions

cpu`s: 2x intel xeon e5-2698v4
ram tested with different configurations, 1-8 pieces


" 100% sure "
Yes, with pc all ok. In 3dsmax+corona ALL OK. In C4D standart render ALL OK. In C4D+corona, but without textures (grey geometry render) ALL OK too.
If use team render, same problem.

The problem is textures (without corona bitmap) and dual-processor configuration.

[Stefan-L]

"The problem is textures (without corona bitmap) and dual-processor configuration. "

what i meant in short, that i not can reproduce that on my dual machines. so you might have some specific problem.
i hope the corona team finds what is  causing this for you.

[Evolist_ua]

"The problem is textures (without corona bitmap) and dual-processor configuration. "

what i meant in short, that i not can reproduce that on my dual machines. so you might have some specific problem.
i hope the corona team finds what is  causing this for you.

People in two of my topics complained about the same problems.

What xeons do you have? Maybe in newer generations that do not have this problem on socket 3647 or newer?

I will put together two test scenes in the coming days (with and without corona bitmaps). Can you test them and share the results?

[maru]

Thanks for the extra details!

[Stefan-L]

"What xeons do you have? Maybe in newer generations that do not have this problem on socket 3647 or newer?

I will put together two test scenes in the coming days (with and without corona bitmaps). Can you test them and share the results?"

yes i can test them.

the machine i use in office still as main working machine (as it is super smooth in all work 2d and 3d, is a dual Intel(R) Xeon(R) CPU E5-2697 v3) so slightly older than yours.
for render clients, i have a different machine single and dual as a v4 xeons and thread rippers (1950x, to 3790x)

i see no difference in corona bitmap or c4d bitmap use.

what does make a difference in both is the handling of file paths, corona (including corona bitmap)  works yet slightly differently here in my experience, so as i recommended one of my guesses is  that your issue it is connected to c4d searching more for files when using corona bitmaps as it might not use c4d bitmap search features identical. (you see differences also in c4d asset browser from both). The c4d bitmap might find a texture where the corona bitmap doesn't. If something is unusual slow in c4d it is very often about asset files not found or searching for them ineffectively..
i also had the case where a slight different file search setup did make slowness on one pc or new c4d versions

To make both file handlings really same would be actually really some great addition if the Corona team could finally make Corona work really the same as the c4d bitmap in c4d internal in regard to file paths and search paths. this includes asking if to copy texture assets when copying materials or objects with materials from one scene to another.

[Evolist_ua]

yes i can test them.

Hi, this is a very simple scene
https://drive.google.com/file/d/1bBznCFUIdCMLZeKV94bprGh-G9VdNyCQ/view?usp=sharing

Here are the test results on my system.
1) Without the corona bitmap, two processors work completely, the performance is about 5m rays.



2) Without corona bitmap, one processor works, because of the Windows 11 scheduler it does not divide perfectly, as it was in Windows 10, but the performance is higher, about 6m+ rays.


3) With corona bitmap, one processor works, because of the Windows 11 scheduler it does not divide perfectly, as it was in Windows 10, but the performance is already much higher, about 9m+ rays. (on win10 with one processor there is no difference with or without bitmap, on win11 there is still a problem due to the fact that the system still divides the load between two processors)


4) With corona bitmap, two processors work, everything is fine, correct performance, about 13m rays.


(my 5950x system gives about 16m rays)

I also tested physical materials, the situation is similar in everything.

Thank`s.

[Evolist_ua]

Hi, I just tested the 11th corona, there is still a problem with the speed on 2 socket systems ((

Is there any progress with this problem?

[andrew1988]

same here with dual intel xeon 8480+ build
disapppointed

Hi, I just tested the 11th corona, there is still a problem with the speed on 2 socket systems ((

Is there any progress with this problem?

[Evolist_ua]

disapppointed

Disappointed is this...

I switched to 3990x, and here I have the same problem as with the dual socket...(mayby sees this as gluing together two processors ((8+8)+(8+8)) + ((8+8)+(8+8)) ). I have to use bitmaps so that there is no loss of speed (many thanks to the guy who made the conversion script).

[andrew1988]

yes  i found some 3ds max map type slow the render speed dramatically ,
such as   Composite map ,   output , and  corona multi-texmap

and also  Coronabenchmark 10  with a uexpected performance

disapppointed

Disappointed is this...

I switched to 3990x, and here I have the same problem as with the dual socket...(mayby sees this as gluing together two processors ((8+8)+(8+8)) + ((8+8)+(8+8)) ). I have to use bitmaps so that there is no loss of speed (many thanks to the guy who made the conversion script).

[Evolist_ua]

Hello, I found out about one feature of Windows, maybe this is the problem?

"Windows supports more than 64 logical cores, and they are divided into groups to perform tasks more efficiently. A logical core can be either a separate processor in a socket, or a physical CPU core or SMT thread.

During boot, the operating system creates groups of logical processors, but their number cannot exceed 64 in one group. If there are more cores, then two groups are created. In the case of 32 cores/64 threads or a CPU with fewer cores, there are no problems - there is one group. But in the case of 65++ threads there are two groups.

On systems with 64 or fewer logical cores, applications run smoothly because Windows associates them with the same group. If there are two groups, then the tasks are alternately assigned to the first or second group. And an application can only access its group's threads - unless it recognizes multiple groups. Thus, the application must be able to work with several groups of logical cores. If there is no such support in the application, then the number of logical cores is limited to 64, and processor resources are not fully used."

[Nejc Kilar]

Howdy folks! You are not forgotten :) Or shall I say we are not forgotten because I also have 64 core Threadripper which has some of the issues you mentioned.

The issue is a bit hard for us to diagnose right now but we are putting effort into resolving it. We tried a couple of solutions internally with not very good results but then some solutions are already being rolled out as they do improve performance. These latest improvements mainly affect high thread count CPUs not necessarily 2S systems.

So please if you have issues similar to the ones outlined in this thread please let us know. Any new info about different system configurations struggling with these kinds of issues will make it easier for us to have a more complete improvement.

[Evolist_ua]

I saw in the latest daily build “Improved performance of Corona renderer with bucket rendering on high-core count machines”, I tested it, nothing has changed compared to the release version of the 11th corona. Still bad speed without corona bitmap, the difference in the number of rays is enormous

[TomG]

The fix is for high core count single CPU machines - we're still investigating potential issues on high core count multi-CPU systems. Thanks!

[Evolist_ua]

The fix is for high core count single CPU machines - we're still investigating potential issues on high core count multi-CPU systems. Thanks!

In this case, this is a single-processor system, I specifically attached the task manager, where you can see that it is 3990x (64 cores/128 threads). There is no way to test it on a dual-processor system right now.

[TomG]

Interesting, as this is the "Dual cpu speed problems" thread ;)

[Evolist_ua]

Interesting, as this is the "Dual cpu speed problems" thread ;)

Yes))

I have a system with a dual processor, but I thought that switching to a high-performance single-processor system would help me, but as it turned out, I have the same problem))

P.S. on my system with 5950x (16 cores/32 threads) everything is fine.

[TomG]

The report is welcomed of course! I was looking to see if we had a better thread to report it under, but for now there isn't one, we might set one up though if we hear more issues with single CPUs. On the 3990X we have for testing, this fix did indeed make a difference, so this is interesting.