Corona not using all cores ?

[visualart]

Hi, just acquired myself a new workstation with a dual 18-core, so 36 physical cores, 72 logical processors.
When rendering, I see only 36 processes are used.  With iray I also only see 36cores listed as maximum, anyone knows if  this is an autodesk limitation ?

Would be a shame not to have all potential unleashed here... :)

[Juraj]

What infographic are you looking it ? Task manager/Logical processors ? And it shows only half of them are working and CPU usage is 50perc. ?

Depending on what you're judging the use upon, you either already are using 72 threads, or have Hyper-threading off, but I don't think that would be the case.

[visualart]

I'm looking at task manager & resource monitor.

In task manager it shows 3dsmax application CPU 55.4%, in resource monitor it shows CPU average 49.44
When looking at the cpu curves, I can see CPU 0 to 36 (node0) not working, CPU 0 to 36 (node1) working 100%

So it kind of looks like 1 cpu is neglected.

In the bios hyperthreading is enabled.

[Juraj]

Hmm, well, that's the new 2699v3 ?

Do you have Windows8, perhaps latest update, chipset drivers installed ? Everything in UEFI looks right ?

[visualart]

Looks like the system wasnt installed using UEFI.
I hope i don't have to reinstall everything here... So far it looks like only 1 of the 2 cpus isnt used for rendering.

I'll check all latest drivers. It's the new E5-2699v3 indeed :)

Just turned off hyperthreading and now it's using all cores. (except the logical HT's of course)

[Juraj]

Do you have the "Professional" version of your Windows ? Although that would probably not even read those additional cores, not only ignore their use.

By being installed without EUFI you mean the boot doesn't give you option to even choose UEFI shell (All current boards support simultaneous use, I suppose you have the Asus board) ? It just loads into regular bios ?

[visualart]

I have a supermicro X10 DRI-T4+ motherboard. Windows version is 8.1 Pro

It loads just regular bios. I'll have a look into using the uefi shell.

[Juraj]

You might already have it or not. By default it seems to have its own regular bios, but the web also states support for generic 2.7.x uefi support. You bought expensive server-grade board with features towards data centers and not workstation. You paid for Quad-Lan, 24dimm slots and sas3 ports, none of which you you will ever use, and instead miss better thermal design, comfortable UEFI, and option to squeze another 10perc. performace. Strange choice ?

The issue will be somewhere else. Time to perhaps dig into manual :- ) But I don't think that has much to do with Corona at all.

[visualart]

I got this adviced as I'm using 256GB of ram, a shitload of raiddisks and using heavy loads of network traffic.

What would be your choice as motherboard ?

[Juraj]

I got this adviced as I'm using 256GB of ram, a shitload of raiddisks and using heavy loads of network traffic.

What would be your choice as motherboard ?

Workstation oriented Asus like Z10PE-D16 WS. I am using Z10PE-D8 WS for current build in process as I don't need more than 64gb (but using 16gb sticks you still go up to 192), but D16 has enough to go up to 1024. Looks like it's not easily available outside of OEM builders yet.
Pretty much the same features and usability in workstation environment plus far more comfortable use and better build for performance, minus the server stuff (sas3 ports,...)

But ok, given that you're on rendering forum I had no idea you could use 256gb or prefer to build disk raids inside workstation, so I presume this is more video-editing machine ? No honestly, now I am quite curious :- )

I am not really fond of aggregated links (using multiple NICs on boar), does that really help you with traffic ? I've setup "cheap" 10gbit network (dual-NIC intel cards inside every workstation, plus netgear prosafe unmanaged switch), maybe this is something you can look into given it's probably within your budget ? It's best thing I did since I use SSDs inside file-server, but don't setup any storage inside workstations since I need all three of us in office to access them.

Sorry for side-tracking, I'll still look into what could cause your issue.

[visualart]

Thing is, I do a variety of stuff ranging from video-editing/compositing, 3D, simulations, audio. I take my workstation a lot on the road so I want everything maxed out in as less machines as possible. I'm working a lot for entertainment and resolutions/file sizes get huge rather quickly (imagine making 1hour of videocontent on 16k resolutions).
Then when I'm working from home and can comfy use my rendernodes & server I don't want to have the feeling I'm on a network.

I'm still looking into optimizing my network doh, I have the advantage of usually being with maximum 3 persons on the network, however the filesizes and amount of data transfer can bring it down fast. (last time happened with a 20TB krakatoa project). So any tips are defenatly welcome !

In meantime I contacted autodesk, adobe & intel, none of them have a clue so far and are looking further into it.

Thanks for the help !

[Stan_But]

Is there some task that use all of 72  logical processors on your system?

[arqrenderz]

If you use some of the Stress test software out there like Everest ?

[daniel.reutersward]

I´m using a Supermicro motherboard, the X10DAi, with dual Intel Xeon E5-2697v3 (didn´t know Asus released a new dual cpu board until now...would have gotten it if I knew!) and Windows 8 and it´s using all 28 physical and all 56 logical cores.

Could there be some limitation in Windows? I read somewhere that there was a limitation but can´t remember how many physical/logical cores.

/Daniel

[Juraj]

I´m using a Supermicro motherboard, the X10DAi, with dual Intel Xeon E5-2697v3 (didn´t know Asus released a new dual cpu board until now...would have gotten it if I knew!) and Windows 8 and it´s using all 28 physical and all 56 logical cores.

Could there be some limitation in Windows? I read somewhere that there was a limitation but can´t remember how many physical/logical cores.

/Daniel

There is but not in Pro version. x64 Windows8.1 pro can go into 256 I think, but using 2 CPUs only.

[daniel.reutersward]

There is but not in Pro version. x64 Windows8.1 pro can go into 256 I think, but using 2 CPUs only.

I see! I thought less than 256, anyway I´ve been trying to think of something that could cause the problems but right now I can´t think of anything.

[visualart]

Just did the AIDA benchmark and there it's using full 72 cores at 100%.

Doing some further tests now and looks like everything within max seem to not go beyond 50-60%

- fumefx : 55%
- mentalray : 60% (only 36 buckets shown)
- vray : 59% (only 36 buckets shown)
- iray : 57% (only 36 processors visible in hardware settings)
- corona : 55%

For the sake of it: realflow : 56%

[Juraj]

Which 3dsMax version and service pack is it ? And which exact OS do you have ?

[Juraj]

There is but not in Pro version. x64 Windows8.1 pro can go into 256 I think, but using 2 CPUs only.

I see! I thought less than 256, anyway I´ve been trying to think of something that could cause the problems but right now I can´t think of anything.

It's 32 for regular and all x32 versions, and 256 for x64 versions. 2 CPUs only, anymore and you need server Windows (2008/2012/etc..). Just checked. So I guess this is out of way, since he gets the performance in synthetic benchmarks.

[daniel.reutersward]

It's 32 for regular and all x32 versions, and 256 for x64 versions. 2 CPUs only, anymore and you need server Windows (2008/2012/etc..). Just checked. So I guess this is out of way, since he gets the performance in synthetic benchmarks.

I see, thanks for the info! I´ve googled my @## off trying to find an answer, I think it´s weird if it´s 3ds max that´s causing the problem?

[daniel.reutersward]

I found a thread where someone said:

"Guys, guys, I HAVE A SOLUTION FOR YOU: Open 3dsmax. Hit F11 to bring up the Maxscript window. Then, copy paste this line, and hit ENTER on the Numpad:
renderers.current.system_numThreads=64"

It´s a really old thread but perhaps worth a try but change "64" to the correct number?

Here´s a link to the thread: http://forums.autodesk.com/t5/3ds-max-3ds-max-design-general/64-cores-support/td-p/4239199

[visualart]

Hey Daniel, thanks for the link, tried it, akward doh, I only get 58 buckets and resource monitor still shows around 60% which makes no sense at all...

In meantime as far as your other questions: i'm using max 2014 SP5 on Windows 8.1 Pro

[visualart]

I just had an interesting chat on the intel support. What version of intel Embree is corona using ?
The newer version should be more optimised towards xeon Phi and v3 generation processors. (planning on putting in a xeon Phi in the future :))

[Juraj]

But you have the same issue in Vray(3?), which by default doesn't have Embree enabled (which is good, because it's horrible).

[arqrenderz]

"But you have the same issue in Vray(3?), which by default doesn't have Embree enabled (which is good, because it's horrible)."

But corona uses embree by deafult on intel procesors no??  why is it horrible?

[visualart]

Vray 3 is using embree as well (as an option)

According to the chat I had it's more a software issue about how  the new processor"s power is adressed. Physically there's nothing wrong with the system hardware. It's more about the software not being optimised in how to use it.

[Juraj]

"But you have the same issue in Vray(3?), which by default doesn't have Embree enabled (which is good, because it's horrible)."

But corona uses embree by deafult on intel procesors no??  why is it horrible?

Vray 3 is using embree as well (as an option)

But you have the same issue in Vray(3?), which by default doesn't have Embree enabled (which is good, because it's horrible).

It was pretty clear what I said...

But corona uses embree by deafult on intel procesors no??  why is it horrible?

Vray's implementation is horrible which is what I refered to.

Up until 3.007 it had hundreds of bugs which would arise anytime with scene complexity. After 3.008 this is ok, but it's still unusable because it basically switches off ALL instances (with no threshold to size), making it impossible to memory manage even medium complex scenes.
Corona is far more flexible in this regard and is build primarily around it, not only as after thought.

[Stan_But]

Is there some task that use all of 72  logical processors on your system?

and yet
Try to find at least one task that will use 100% CPU

[johan belmans]

@Visualart:

Hi Tom

how are you doing?
Looks like a massive workstation, lets hope in the near future that Xeon Phi can be used for corona.

@Keymaster:
Last year you mentioned that at that moment the memory space of the Phi was not compatible with Corona.
You gave ohter reasons aswell why it could not be supported.

Is this still the case with new Phi and the latest version of embree?

cheers

johan

[Ondra]

adding support for phi would still cause a lot of problems

[visualart]

Got a mail back from Chaosgroup :

"We do not have any thread count limitation, but there is some "issues" with NUMA
technology.
Please tell us the exact version of V-Ray you are using and the version of 3DS Max.
We will attach document how to optimize CPU utilization on your machine, starting
multiple DR spawners."

[visualart]

Onto further research, I tried the vray trick but it's not working. When I check the affinity the processes are sent to it seems that 3dsmax.exe is only sent to 1 processing group thus using only 1 cpu. When I try to assign the spawned version to the other group this one fails to work.

So kind of narrowed down to the problem being processors are split due NUMA and using the 2 groups on the same process is not possible.

[Juraj]

The NUMA issue was dominant in Opteron architecture, but not needed in 2p Xeons. "Distributed rendering" in single machine is also very stupid workaround, absolutely defeating the point of machine.
There must be something else. If 40 and 48 threads machines are still used fine, it's strange it's 72 that suddenly causes issue to only use half. Maybe just don't use HT until then, it's better then to use just half.

[racoonart]

well, I'm not helpful but.. WTF I didn't even know there were processors with more than 6 cores :D This is madness... (and I'm seriously bad informed about CPU development as it seems)

[daniel.reutersward]

As far as I know the one with most cores is the one he has, the 18 core one! Just wish it would be possible to have them at 4Ghz! :)

[visualart]

Ok, gave up, going without HT untill some solution drops out. In meantime some testing, made a little benchmark scene,
in case you want to try as well, the scene is attached, just hit render.

1 is result from the new workstation


other is distributed with old dual x5680 and 4 dual E5-2670

[arqrenderz]

Here is my PCs
1 Is i7 4930 K occ 4.2ghz                                     Time 12:04    rays 7.772.300       
2 is i7 4930 K occ 4.2ghz + 4770k Occ 4.0ghz    Time  8:15     Rays 11.421.914

[fire3d]

here's mine:

I7 3770k oc 4.7ghz - no DR

I7 3770k oc 4,7ghz + I5 3340 oc 3.5Ghz DR

I think i need new hardware :P

[zacuttialberto]

Good day to everyone,

I'm very confused with the update of my workstation (i posted i new subject just to collect suggestions) but one  question if you are using only one processor with HT on the software is using all CORE and Thread or not?

this because one my option was to set up a small workstation based on As Rock MB and install a 12 or 18 core XEON.

Ciao

Alberto

[johan belmans]

Hereby my results

first:  workstation i7-3970X @3,5GHz without DR

Second: workstation i7-3970X @3,5GHz with DR (2x dual E5-2630 v2 @2,6 GHz + 1x dual X5650 @2,66GHz)

[visualart]

Would it be possible to run the same machine as distributed node as well with corona ?

I wonder as atm i'm rendering a huge project in the background while still working on another max instance and i'm still getting like 20M rays/sec while the background continues it's task at full speed.

[johan belmans]

I don't know, haven't tried it.

[Dimer]

Hi!
My test on workstation dual intel xeon E5-2697 v2 (Engineering sample)

[sbrusse]

i7-4930k 3.9Ghtz

13m22s

[borisquezadaa]

Hey!... look a this explanation why...

http://area.autodesk.com/blogs/maxstation/n268-how-many-cores-does-3ds-max-support

It seems not worth the effort buying those expensive rig for use in 3ds max unless...

https://msdn.microsoft.com/en-us/library/windows/hardware/dn653313%28v=vs.85%29.aspx

[arqrenderz]

So can Visualart try Vlado solution??
"Posted 4 days ago
For rendering specifically, you can also run a DR server on your local machine and do a local DR session, with 3ds Max running on one half of the cores, and the DR server - on the other half.
Best regards,
Vlado"

[borisquezadaa]

Seems a solution worth trying. Sadly i dont have such rig to test it.

[Ondra]

I can confirm that this will work in Corona.

[LuckyFox]

Sorry to up such an old topic.
We experienced something like only 25% usage while rendering of our 2x Xeon workstations.
Found a solution by reseting the Corona (assign a different renderer and back to Corona).
Also found it is happening for us when working on the same project with 2 different versions of corona (1.3 - 1.4)
Hope this post helps somebody who is experiencing the same problems as we did.

[maru]

Sorry to up such an old topic.
We experienced something like only 25% usage while rendering of our 2x Xeon workstations.
Found a solution by reseting the Corona (assign a different renderer and back to Corona).
Also found it is happening for us when working on the same project with 2 different versions of corona (1.3 - 1.4)
Hope this post helps somebody who is experiencing the same problems as we did.

My first guess would be making sure all Windows Updates and MSVC++ 2015 redistributable are installed. See: https://coronarenderer.freshdesk.com/support/solutions/articles/12000009767-i-am-getting-errors-when-starting-3ds-max-with-corona-1-4-
Although this is just a shot in the dark.