OctaneRender™ Standalone 4.00

Forums: OctaneRender™ Standalone 4.00
VIP Information, news and announcements regarding new Octane Render commercial products and releases.

Re: OctaneRender™ Standalone 4.00 [current 4.xx]

Postby Notiusweb » Sat Nov 24, 2018 1:55 am

Notiusweb Sat Nov 24, 2018 1:55 am
Notiusweb wrote:Abstrax,
I found why I am getting the crashes-
I checked the DPC Latency Checker program and it spikes every time the "AI Scene Caching mesh" notice appears.

Is that caching happening in the CPU or GPU....or when the one is feeding the other info
Because the more GPU I have, the higher the spikes are when it happens.
In my case, if I have 5 or less GPU, there are almost no spikes, but as I add in my 6, 7, and 8, the spikes go much much higher.
To the point, where it is showing that the GPUs are stalling the CPU. This is why my GPU is freezing and shutting down.
I know this now because I can see when I change the frame on the timeline in a scene with an ABC animation, when too high of a spike is generated, it correlates with a freeze suddenly.

What is interesting is that
(1) restarting the render with the backwards arrow does not trigger a high spike, or raise the DPC Latency really at all.
(2) moving the scene around with the cam does not trigger a change
(3) changing a texture does not trigger a change

It is when I select a frame, or when each frame is being processed in the Batch Renderer, and the "AI Scene Caching Mesh" notice appears
Also, it occurs whether the denoiser is on or off, or the render-resolution is low or high.

Also, I tried to change the max thread cores for AI in app settings, and if I made 6 or less, I could predictably immediately crash the system into a BSOD when I then used >5 GPU.
Don't know why that is, but it was interesting to see maybe it is related to the rest of it.


My question-
Would there be a way to make a selection box, like how you made for the Tonemapping per-card selection back in V3, for the AI Scene Caching Mesh?

-THX so much.


Abstrax, just read the V4 Cuda Error 2 post, and that there is a memory component to it...

When I'm getting BSODs, the Blue screen error message will often say:
"Page Fault In Nonpaged Area".
Since this message is memory-related in nature, thought I would share it here.
Win 10 Pro 64, Xeon E5-2687W v2 (8x 3.40GHz), G.Skill 64 GB DDR3-2400, ASRock X79 Extreme 11
Mobo: 1 Titan RTX, 1 Titan Xp
External: 6 Titan X Pascal, 2 GTX Titan X
Plugs: Enterprise
User avatar
Notiusweb
Licensed Customer
Licensed Customer
 
Posts: 1285
Joined: Mon Nov 10, 2014 4:51 am

Re: OctaneRender™ Standalone 4.00 [current 4.xx]

Postby Notiusweb » Sat Nov 24, 2018 11:56 am

Notiusweb Sat Nov 24, 2018 11:56 am
I will note that using the diagnostic DPC Latency Checker I can see off same scene with same exact settings, that there are much more higher spikes into the red danger zone with the stable V4 than with RC7 when I use all 8 of my GPU. Something that was changed between V4_Stable compared to V4_RC7 is making a difference.
Maybe some enhancement is taxing CPU/GPU memory heavier now.
Win 10 Pro 64, Xeon E5-2687W v2 (8x 3.40GHz), G.Skill 64 GB DDR3-2400, ASRock X79 Extreme 11
Mobo: 1 Titan RTX, 1 Titan Xp
External: 6 Titan X Pascal, 2 GTX Titan X
Plugs: Enterprise
User avatar
Notiusweb
Licensed Customer
Licensed Customer
 
Posts: 1285
Joined: Mon Nov 10, 2014 4:51 am

Re: OctaneRender™ Standalone 4.00 [current 4.xx]

Postby Notiusweb » Sat Nov 24, 2018 1:49 pm

Notiusweb Sat Nov 24, 2018 1:49 pm
Wow, I now ran the same scene through RC6, and DPC Latency is way down, even compared to RC7.

So, I can load up the same scene in RC6, RC7, and V4_Stable, within a 10 minute span of eachother, say, and let the scene render on several separate frames of the animation, while tracking through an ongoing DPC Latency Check instance, and I can see red line 'danger zone' instances increasing as the version date increases.
So, whatever phenomenon increased from RC7 to V4_Stable, also had previously increased from RC6 to RC7.

FYI - Here is what is the same-
1) Render speed, does not change between RC6 and V4_Stable
2) Latency surge - occurs always at the moment "AI Scene Caching Mesh" appears


Noting what was listed between RC6 and RC7 as memory related:
-Fixed excessive usage of system memory during the compilation of heavily overlapping geometry.



Here is related between RC7 and V4_Stable as memory related:
-Refactored the way how we manage image textures on the GPU and in out-of-core memory. This should reduce the stress on the CUDA driver and solve update issues when out-of-core memory is being used for textures.
-Trying to make out-of-core memory management more robust in conjunction with the denoiser: Octane makes now sure that enough memory is reserved for the denoiser if out-of-core memory is enabled. We also re-allocate out-of-core memory if an update fails due to a lack of GPU memory.
-Simplified CUDA texture handling to reduce stress on CUDA driver.
-Fixed out-of-core usage in the simple mesh case, i.e. if the scene contains only 1 mesh without a transformation.



Maybe excessive usage of system memory is better for stability in some ways?
Well, I'll keep looking to see if I can find anything correlating or noteworthy.
Win 10 Pro 64, Xeon E5-2687W v2 (8x 3.40GHz), G.Skill 64 GB DDR3-2400, ASRock X79 Extreme 11
Mobo: 1 Titan RTX, 1 Titan Xp
External: 6 Titan X Pascal, 2 GTX Titan X
Plugs: Enterprise
User avatar
Notiusweb
Licensed Customer
Licensed Customer
 
Posts: 1285
Joined: Mon Nov 10, 2014 4:51 am

Re: OctaneRender™ Standalone 4.00 [current 4.xx]

Postby manalokos » Sun Nov 25, 2018 9:54 am

manalokos Sun Nov 25, 2018 9:54 am
Dear Abstrax,

We use Octane standalone, it crashed several times with a windows error, not CUDA error, I think it happens
when we are changing some nodes while compiling. Using Windows 7 and 4x1080ti

abstrax wrote:
manalokos wrote:I'm getting pretty frequent crashes with this new version. It is not useable in our current production

Thank you for your report. We are really keen to fix all issues as quickly as possible, but a bit more information:

- Which plugin do you use or are you working with the Standalone?
- What do you mean with "crash"? CUDA failures or crashes where the application dies?
- If it's a CUDA failure, what is the very first error message in the log?
- How can we reproduce the crash?
manalokos
Licensed Customer
Licensed Customer
 
Posts: 438
Joined: Fri Nov 18, 2011 4:43 pm

Re: OctaneRender™ Standalone 4.00 [current 4.xx]

Postby haze » Sun Nov 25, 2018 7:33 pm

haze Sun Nov 25, 2018 7:33 pm
FoolooF wrote:I am having a weird transparency issue with Ver.4

I was working on a project in October using Ver. 3.08.5.
I had a DAZ character wearing a dress, with a ragged cuff to the sleeve.
I just used a .jpg transparency map for the sleeve.
Everything worked fine.

I tried using one of the RC versions for the same project, (I believe RC7 at the time).
The transparency map worked correctly, the sleeve edge was ragged, but the forearm of the character was invisible.

Opened it in Ver. 3.08.5 again and it looked fine
I thought, ver.4 is still experimental so it's going to have some bugs, and just used 3.08.5.


I was playing with VizPeople free fireplace in 3.08.5.
I had two versions of it loaded, one occupying the exact same space as the other.
I did this so I could set one fire as surface visible emission, and the other as an invisible (opacity 0) light source.
It worked fine and looked good.

Then Ver.4 was officially released.
I upgraded my license and installed ver.4.

I opened the same fire scene in ver.4 and the fire was gone, invisible.

All other parts of the mesh were visible as in ver.3.08.5, (logs and debris), but the flames were gone.
In this instance, it seemed to me any geometry occupying the same space with opacity at zero was rendering both invisible.


I decided to do some simple testing.


In ver.3.08.5:
I Loaded a sphere into a scene with it's own placement node connected to a geometry group node.
Then I copied the sphere and placement nodes, and connected those copies to the same geometry group node.
So both spheres are occupying the same space.
If I set the opacity of one sphere to zero, I can still see the other sphere,(as it should be).

In ver. 4.0:
I did the exact same procedure as above, but both spheres disappeared.
If I set one sphere slightly out of sync with the other, (0.000010), I can still see one sphere at some angles, and see through it at others.


Am I missing something simple, or doing something wrong?
Does ver.4.0 have some kind of different setting I'm not aware of?


My system:

Win 7 sp1 64bit
16 gigs system ram.
Geforce GTX 760, with 2 gigs ram.
Drivers : 416.94


I just tried this in 3.08.5 and in 4.00. I set up a scene with a green sphere and a red sphere, both occupying the same space. Ideally this shouldn't be done in Octane (since it can't be done in the real world anyway).

In the scene the red sphere is connected to the first input of the geometry group, and the green sphere is connected to the second input.

In 3.08.5 if I render this scene I see the green sphere. If I adjust the opacity of the red sphere, nothing changes. If I adjust the opacity of the green sphere, both spheres disappear.

In 4.00, if I adjust the red sphere, both disappear, but if I adjust the green sphere, nothing changes.

I'll find out if there is an ordering issue.
User avatar
haze
OctaneRender Team
OctaneRender Team
 
Posts: 961
Joined: Sun Feb 08, 2015 8:57 pm

Re: OctaneRender™ Standalone 4.00 [current 4.xx]

Postby abstrax » Sun Nov 25, 2018 7:41 pm

abstrax Sun Nov 25, 2018 7:41 pm
Notiusweb wrote:Wow, I now ran the same scene through RC6, and DPC Latency is way down, even compared to RC7.

So, I can load up the same scene in RC6, RC7, and V4_Stable, within a 10 minute span of eachother, say, and let the scene render on several separate frames of the animation, while tracking through an ongoing DPC Latency Check instance, and I can see red line 'danger zone' instances increasing as the version date increases.
So, whatever phenomenon increased from RC7 to V4_Stable, also had previously increased from RC6 to RC7.

FYI - Here is what is the same-
1) Render speed, does not change between RC6 and V4_Stable
2) Latency surge - occurs always at the moment "AI Scene Caching Mesh" appears


Noting what was listed between RC6 and RC7 as memory related:
-Fixed excessive usage of system memory during the compilation of heavily overlapping geometry.



Here is related between RC7 and V4_Stable as memory related:
-Refactored the way how we manage image textures on the GPU and in out-of-core memory. This should reduce the stress on the CUDA driver and solve update issues when out-of-core memory is being used for textures.
-Trying to make out-of-core memory management more robust in conjunction with the denoiser: Octane makes now sure that enough memory is reserved for the denoiser if out-of-core memory is enabled. We also re-allocate out-of-core memory if an update fails due to a lack of GPU memory.
-Simplified CUDA texture handling to reduce stress on CUDA driver.
-Fixed out-of-core usage in the simple mesh case, i.e. if the scene contains only 1 mesh without a transformation.



Maybe excessive usage of system memory is better for stability in some ways?
Well, I'll keep looking to see if I can find anything correlating or noteworthy.

We made two changes to improve upload speeds / GPU updates: One is reordering operations to allow faster GPU uploads and the second one is using a different method of uploading.

If you want I can make you a test build of the Standalone with the fast upload being disabled to see if that is causing the issue. Please let me know if you are interested.
In theory there is no difference between theory and practice. In practice there is. - Yogi Berra
User avatar
abstrax
OctaneRender Team
OctaneRender Team
 
Posts: 5483
Joined: Tue May 18, 2010 11:01 am
Location: Auckland, New Zealand

Re: OctaneRender™ Standalone 4.00 [current 4.xx]

Postby Notiusweb » Sun Nov 25, 2018 8:05 pm

Notiusweb Sun Nov 25, 2018 8:05 pm
abstrax wrote:If you want I can make you a test build of the Standalone with the fast upload being disabled to see if that is causing the issue. Please let me know if you are interested.


Thx, PM sent.
Win 10 Pro 64, Xeon E5-2687W v2 (8x 3.40GHz), G.Skill 64 GB DDR3-2400, ASRock X79 Extreme 11
Mobo: 1 Titan RTX, 1 Titan Xp
External: 6 Titan X Pascal, 2 GTX Titan X
Plugs: Enterprise
User avatar
Notiusweb
Licensed Customer
Licensed Customer
 
Posts: 1285
Joined: Mon Nov 10, 2014 4:51 am

Re: OctaneRender™ Standalone 4.00 [current 4.xx]

Postby funk » Tue Nov 27, 2018 4:54 am

funk Tue Nov 27, 2018 4:54 am
I was testing light linking IDs by disabling them all in the kernel, but was still getting light in my interior.

I have portals in my windows, and it seems they still emit light even when you disable all light IDs.
Win10 Pro/ Ryzen 5950X / 128GB / RTX 4090 / MODO
"I am the resurrection, and the life: he that believeth in me, though he were dead, yet shall he live" - Jesus Christ
User avatar
funk
Licensed Customer
Licensed Customer
 
Posts: 1204
Joined: Mon Feb 07, 2011 1:24 pm
Location: Australia

Re: OctaneRender™ Standalone 4.00 [current 4.xx]

Postby Jolbertoquini » Tue Nov 27, 2018 10:52 am

Jolbertoquini Tue Nov 27, 2018 10:52 am
Hi Guys,

Thank you for the update. I have quick questions, is that no way to have cryptomatte working with V4 then only 2018.1 because is kind of features we are looking for a while and until we can really use on production will take a huge time.?

Could we have maybe a v4.01 with Cryptomatte Pleeease??

Cheers,
JO
Octane Render for Maya.
https://vimeo.com/jocg/videos
https://www.linkedin.com/in/jocgtd
http://www.hmxmedia.com/
--------------------
Join MAYA OCTANE USERS Skype discussion here :
https://join.skype.com/LXEQaqqfN15w
User avatar
Jolbertoquini
Licensed Customer
Licensed Customer
 
Posts: 1067
Joined: Sun Aug 31, 2014 7:08 am
Location: London

Re: OctaneRender™ Standalone 4.00 [current 4.xx]

Postby manalokos » Tue Nov 27, 2018 5:39 pm

manalokos Tue Nov 27, 2018 5:39 pm
Recently we've tried to do some rendering with Denoiser on, and we already got 2 sudden reboots.
Using windows 7 and Nvidia driver 416.34
manalokos
Licensed Customer
Licensed Customer
 
Posts: 438
Joined: Fri Nov 18, 2011 4:43 pm
PreviousNext

Return to Commercial Product News & Releases (Download here)


Who is online

Users browsing this forum: No registered users and 2 guests

Tue Mar 19, 2024 5:41 am [ UTC ]