DR render problem.

[d30evoco]

Hey guys hows it going?
We are in the middle of a new project and recently just for that project alone we upgraded our pcs. So the specs are 1950x threadrippers with msi399x motherboards, each one equipped with 64 gb ram (win 10,max 2018, corona 1.7.4). Everything looked dandy until we got to the dr part, heres a problem: so on average one render (4k) takes about 1 hour on 1 computer to reach 5% noise level. Now when we connected 2 additional pcs we were hoping to get the same noise level faster at least 2 times faster, but it didnt happen. It took exactly the same amount of time. I wonder if anyone else encountered this problem and if so how do you guys dealt with it? As for the things we tried: lan is fine 100mb, we tried different corona builds even corona 2.0 but it didnt really help, we didnt set the limit by number of passes or by time, only by noise level. So where did we go wrong? Your help appreciated, cheers.

 

[TomG]

Just some quick initial questions - what did DR report in the VFB? Did it show the machines contributing passes? Do you have the DR logs to show?

[d30evoco]

Hey thanks for the reply, yeah dr slaves indeed contributed passes (they were connected as far as i remember there were no mistakes) albeit real slow. Regarding logs unfortunately we didnt save them, but that would be easy to replicate and save them.

[TomG]

TY for the info! I'll have to let the developers say what would be the most useful info for diagnosing from this point on, could be DR logs, maybe even the scene etc. not sure what would be best for them.

[urbanite]

Hi

I have just install Corona 2 and I am having same/similar problem. On my workstation on test scene I am getting 15 passes in 6 minute. Adding 2 nodes (i7 and AMD TR) changed nothing in therms of speed although I see that they are 100% load in task manager. I cannot see the image preview corona dr slaves. Running max 2014.

[d30evoco]

Any luck with that problem?

[Frood]

Can someone please at least post a screen from "Stats" and "DR" tab of Corona VFB while rendering? Without any further information this is hard to judge. And just to be sure: you are not using some additional time limit?


Good Luck

[urbanite]

No additional time limits. Further tests shows that that only old scenes (corona 1.6 in my case) were having trouble with DR. Merging to the new one solved it.

[Mario Rothenbühler]

Same Problem here
render with 3 or 4 machines is not much more faster then 1 machine
alle machines contribute passes and CPU load is @ 100%

[maru]

If anyone is having similar problems, please report them to support@corona-renderer.com or through https://coronarenderer.freshdesk.com/support/tickets/new
That's the best way to have your issue looked into and resolved.

[Mario Rothenbühler]

Dear Team

I do some real tests here with 4 machines... i facing the weird rendertimes also on network rendering
first i think we have to solve this weird issue... and after then we can solve DR rendertimes

specs of the machines are (Corona benchmark 1.3)

T7500-1 (local, gbit):
Corona 1.3 Benchmark Finished
BTR Scene 16 passes
Intel(R) Xeon(R) CPU X5670 @ 2.93GHz
 Real CPU Frequency [GHz]: 3
Render Time: 0:02:28, Rays/sec: 3 271 340

T7500-2 (local, gbit):
Corona 1.3 Benchmark Finished
BTR Scene 16 passes
Intel(R) Xeon(R) CPU X5680 @ 3.33GHz
 Real CPU Frequency [GHz]: Undetected
Render Time: 0:02:16, Rays/sec: 3'551'220

T7500-3 (local, gbit):
Corona 1.3 Benchmark Finished
BTR Scene 16 passes
Intel(R) Xeon(R) CPU X5680 @ 3.33GHz
 Real CPU Frequency [GHz]: Undetected
Render Time: 0:02:09, Rays/sec: 3'746'800

Render-02 (offsite, VPN, 100Mbit):
Corona 1.3 Benchmark Finished
BTR Scene 16 passes
Intel(R) Xeon(R) Platinum 8168 CPU @ 2.70GHz (x2)
 Real CPU Frequency [GHz]: 2.7
Render Time: 0:00:32, Rays/sec: 14,957,200

Regarding the Corona Benchmark the Azure RS-2 have to be the x4 speed of the other machines.
But in real case it is only x2. this makes no sense... compared to vray the scaling of CPU renderpower is linear... more cores more clock = shorter rendertimes.... linear maybe 2-5% loss in network communication. but not more

attached screenshot of rendering all 4 machines
and screen of the backburner rendertimes

also attached the scene file... very basic.. not much in the file.

thank you for your help in advance

/.mario'

[maru]

Dear Team

I do some real tests here with 4 machines... i facing the weird rendertimes also on network rendering
first i think we have to solve this weird issue... and after then we can solve DR rendertimes

specs of the machines are (Corona benchmark 1.3)

T7500-1 (local, gbit):
Corona 1.3 Benchmark Finished
BTR Scene 16 passes
Intel(R) Xeon(R) CPU X5670 @ 2.93GHz
 Real CPU Frequency [GHz]: 3
Render Time: 0:02:28, Rays/sec: 3 271 340

T7500-2 (local, gbit):
Corona 1.3 Benchmark Finished
BTR Scene 16 passes
Intel(R) Xeon(R) CPU X5680 @ 3.33GHz
 Real CPU Frequency [GHz]: Undetected
Render Time: 0:02:16, Rays/sec: 3'551'220

T7500-3 (local, gbit):
Corona 1.3 Benchmark Finished
BTR Scene 16 passes
Intel(R) Xeon(R) CPU X5680 @ 3.33GHz
 Real CPU Frequency [GHz]: Undetected
Render Time: 0:02:09, Rays/sec: 3'746'800

Render-02 (offsite, VPN, 100Mbit):
Corona 1.3 Benchmark Finished
BTR Scene 16 passes
Intel(R) Xeon(R) Platinum 8168 CPU @ 2.70GHz (x2)
 Real CPU Frequency [GHz]: 2.7
Render Time: 0:00:32, Rays/sec: 14,957,200

Regarding the Corona Benchmark the Azure RS-2 have to be the x4 speed of the other machines.
But in real case it is only x2. this makes no sense... compared to vray the scaling of CPU renderpower is linear... more cores more clock = shorter rendertimes.... linear maybe 2-5% loss in network communication. but not more

attached screenshot of rendering all 4 machines
and screen of the backburner rendertimes

also attached the scene file... very basic.. not much in the file.

thank you for your help in advance

/.mario'

I am sorry, but I do not fully understand your report. For example, I do not see the Corona 1.3 Benchmark results of the "AZURE-RS-2" computer. What exactly is that computer, and what is its hardware?
I think it would be best if you could contact us using this form https://coronarenderer.freshdesk.com/support/tickets/new and then we would ask you some further questions to find out how exactly we could help. One issue per one support ticket please, so if you have two separate issues (one with render performance, and another one with DR), then please send them separately.

Thank you in advance,
Marcin

[Mario Rothenbühler]

Hi Marcin, Thank you for your Feedback

The Specs and Benchmark Results are the one from "Render-02"

Thank you

/.mario'

[nicolasZ]

Hi,

I think I have the same problem as you.

When I render with the node with noise level limit enabled, it takes the same amount of time it would with only one machine.

I have tried to render only with 1 thread on my master ( set 1 on #of theards in system settings) and with one slave machine :  the slave only adds data on a little stripe and that doesn't affect the noise level limit.

I don't know why, but I'm pretty sure it has been doing this since the last Windows update.

Before this Windows update everthing worked fine with Corona DR (with 1.7.4 and V2). FYI : I'm working with Corona 2 and I have tried with 3dsmax 2016 and 2018


@ TomG and maru : I have already reported this problem (ticket 12673).

I hope you guys will find a solution to that problem, because we have big projects that will begin soon.

Many thanks

[dj_buckley]

I'm seeing similar results.  If I set a render going on the host, leave it for 20 or so passes, I get a decent time estimate.  If I then fire up the Node, the time estimation doesn't change at all once the node has kicked in and rendered a few passes.  I'd expect it to nearly half (if not more as the node is far quicker than the host).