Anyone had this? - VRaySampleRate pass shows DR buckets with less samples than the WS

[TomasEsperanza]

 

All the other render elements look fine, that's the confusing thing.

[siddharthkolte]

I would like to know why/how would anyone use this pass like what good does it make knowing that the red areas are going to take more time to render...

 

I would say the pass should be only run on the WS as for me it is just a testing thing like how much time would it take to render a specific area. I wouldn't use it for distributed rendering... But if anyone is using it for Post then I would like to know that...

[Nic H]

hi siddath

the best way to use it in post

[siddharthkolte]

lol

[TomasEsperanza]

I would like to know why/how would anyone use this pass like what good does it make knowing that the red areas are going to take more time to render...

 

(Assuming that you're not being deliberately obtuse in an attempt to be funny...) The vraysamplerate element is an extremely useful measuring device for balancing render settings. Nevertheless your response is totally unrelated to my query. Thank you.

 

I would say the pass should be only run on the WS as for me it is just a testing thing like how much time would it take to render a specific area. I wouldn't use it for distributed rendering... But if anyone is using it for Post then I would like to know that...

 

Post?! Er... No.

[siddharthkolte]

Well no seriously I have never used this pass I always saw it in the Render Passes but never actually used it.

 

I am sorry if my answer was unrelated, are you wondering why there is difference in the calculation by the DR bucket and the WS bucket?

[ionutteudean]

From my point of view, nothing's going wrong there. There's probably reflected the sun/light in your scene and it adds more samples to to get it right. It can depend a lot on how you set up your subdivs for materials/lights. Maybe you have few material subdivs on that sphere and that might be why it uses more AA. Hope I'm not too far from the truth

[Jeff Mottle]

Let's try to keep it professional nicnic, please.

[TomasEsperanza]

From my point of view, nothing's going wrong there. There's probably reflected the sun/light in your scene and it adds more samples to to get it right. It can depend a lot on how you set up your subdivs for materials/lights. Maybe you have few material subdivs on that sphere and that might be why it uses more AA. Hope I'm not too far from the truth

 

Thanks for your replies guys. Just to clarify, I am generally pretty knowledgeable concerning V-Ray usage, so this isn't really a case of a "newbie" missing something really fundamental.

 

The thing that is (to me quite clearly) wrong with this image is that some buckets (only ones from the render nodes) are showing as if a different (smaller) amount of samples were being used by the render nodes. To me this makes no sense, as the scene samples; lights, materials, etc. plus the GI samples and Anti-Aliasing are all set up in the scene, which is then sent the the nodes to render buckets.

 

So... Why would DR effect the outcome of this specific VrayRenderElement? As a guess, it may be that the nodes can't see the GI maps on the network... but given that the other render elements (RawGI etc.) look clean, this is baffling. Is this a known issue in the V-Ray user community?

[TomasEsperanza]

Here is the same issue illustrated by other VRayRenderElements, from the same scene (different material used this time).

 

[braddewald]

I've never seen that before. It's strange that you can't tell any difference with the other passes as I would have assumed it was a missing map causing the problem (a single server not having a map used for reflection glossiness for example). Could you be running different versions of v-ray? I was also thinking it could be Embree being used on the Intel nodes and not AMD nodes, but I have a mixed farm myself and I've never had this happen...

[Corey Beaulieu]

Is there any chance you are using "Bucket Mode" for your Irradiance Method?

 

if not,

 

What happens when you Brute Force?

[TomasEsperanza]

I've never seen that before. It's strange that you can't tell any difference with the other passes as I would have assumed it was a missing map causing the problem (a single server not having a map used for reflection glossiness for example). Could you be running different versions of v-ray? I was also thinking it could be Embree being used on the Intel nodes and not AMD nodes, but I have a mixed farm myself and I've never had this happen...

 

Yeah, weird innit? The paths are all proper network paths (\\shared folders). The CPUs are all Intel, and the V-Ray version is 2.4 (so no Embree here), on the three machines; Laptop, Workstation, and Rendernode.

Edited April 21, 2015 by TomasEsperanza

[TomasEsperanza]

Is there any chance you are using "Bucket Mode" for your Irradiance Method?

 

if not,

 

What happens when you Brute Force?

 

Re: Bucket mode - I did double check that, but no it's just the standard single frame mode.

 

Brute force, makes no difference.

 

I have looked again at the VRayDRBucket element, and realised it's only the Laptop's buckets that are displaying as less samples. This is the case whether the job is sent from the Workstation or the Laptop. So that narrows it down to being related to the Laptop (or perhaps it's place on the network?). The CPU of the Laptop is much slower than the other machines (maybe a factor?). It's all Ethernet, but the laptop goes to the switch via the router. The other two machines are connected directly to the switch. (Perhaps worth investigating?)

 

Obviously, so long as the render looks fine, this is academic, but I think it's good to run these things down, to be clear on what is happening.

 

Thanks for your suggestions.

[Francisco Penaloza]

it is strange for sure, maybe you can post it in Chaosgroup forum, Vlado may jump on it for sure there.

Different colors in buckets during distribute rendering is nothing new in VRay, I feel with Max 2014 and VRay 2 there is lots of problems affecting this area.

 

Now since you mentioned the performance difference between machines, it may be related. if you think about it this pass or element try to show us how much samples are used over the scene, and it can be logical to think that one machine will process more samples that other slower machine. more blue color less samples used.

Now this contradict the the basic of adaptivity of VRay that depend of using color threshold to control how many passes you'll need and not being time based. I don't know I am trying to guess what it can be, but maybe it is just simple bug and I already had to much coffee for today

 

I did some test here and the only thing I noticed is that the area or edges where one bucket touch the other machine bucket it seems to be more blue than the rest, so my final pass look like a soft grid over my pass.

If you find a solution or explanation I would be very interested in knowing.

[TomasEsperanza]

I did some test here and the only thing I noticed is that the area or edges where one bucket touch the other machine bucket it seems to be more blue than the rest, so my final pass look like a soft grid over my pass.

 

Yep, I know the artifact you speak of, it appears to be typical.

 

I'm (reluctantly) going to muck around with the network cabling configuration soon, as apparently all machines should go directly into the switch. We'll see. I suspect the issue will disappear when I move over to Max 2016 and V-Ray 3 (or stop using the laptop on renders), but that's not very practical at the mo'.

[Corey Beaulieu]

Okay. Last thought and then I'm out of it. The issue is above me for sure, but things like this cause to me to re-read the Help menu and learn more about Vray itself.

 

This is from the notes section of the Image Sampler (Anti-Aliasing) Chapter:

 

A note on RAM usage: image samplers require substantial amount of RAM to store information about each bucket. Using large bucket sizes may take a lot of RAM. This is especially true for the Adaptive subdivision sampler, which stores all individual sub-samples taken within a bucket. The Adaptive DMC sampler and the Fixed rate sampler on the other hand only store the summed result of all sub-samples for a pixel and so usually require less RAM.

 

My thinking is that you are hitting your RAM limit on the Laptops and thus they are quitting out early. Since this is a bucket to bucket issue and assuming you are using the the Adaptive Subdiv Sampler (which is maybe not that typical).

 

Bucket issues are obviously more commonly due to Network problems, especially when DR-ing, but that is something I know nothing about.

 

I hope you figure it out and please share when you do.

[elipan]

From my understanding of DR, the render node will crash (or will show NODE-NAME is not responding in the log) when hitting RAM limit.

 

This looks like a LAN issue to me - are you sure your node have access to LC/IM files?

Did you perform BB session before running DR with V-Ray?

[TomasEsperanza]

My thinking is that you are hitting your RAM limit on the Laptops and thus they are quitting out early. Since this is a bucket to bucket issue and assuming you are using the the Adaptive Subdiv Sampler (which is maybe not that typical).

 

Bucket issues are obviously more commonly due to Network problems, especially when DR-ing

 

Cheers Corey. We can rule out RAM, as it's a tiny little test scene, and all the machines have at least 32GB. Adaptive DMC is being used. I've tried Adaptive Subdivision which makes the disparity less apparent, but the artifact is still there.

[TomasEsperanza]

From my understanding of DR, the render node will crash (or will show NODE-NAME is not responding in the log) when hitting RAM limit.

 

This looks like a LAN issue to me - are you sure your node have access to LC/IM files?

Did you perform BB session before running DR with V-Ray?

 

The LAN cabling has now been checked and improved (cat 7 into the switch for all machines).

 

I have successfully run BB jobs from both the WS and the Laptop, experimenting with different BB configs; Manager on WS, then Manager on Laptop, yielding the same results each time; (consistently only the buckets pertaining to the Laptop are lighter).

 

When the same file (located in a shared folder on the laptop) is opened on the WS, and DR rendered with just the WS and the RNode, it's fine.

 

I'm beginning to suspect that the VRaySamplerInfo pass is illustrating the buckets incorrectly, perhaps using some kind of algorithm that doesn't take into account some significantly varied CPU threads. Still it doesn't effect the other passes so, I'm going to let it go... spent enough time trying to understand this. When I originally posted, I was hoping someone would go "oh yeah it's just this..." sorted! Nevermind, Thank you all.

[elipan]

The LAN cabling has now been checked and improved (cat 7 into the switch for all machines).

 

I have successfully run BB jobs from both the WS and the Laptop, experimenting with different BB configs; Manager on WS, then Manager on Laptop, yielding the same results each time; (consistently only the buckets pertaining to the Laptop are lighter).

 

When the same file (located in a shared folder on the laptop) is opened on the WS, and DR rendered with just the WS and the RNode, it's fine.

 

I'm beginning to suspect that the VRaySamplerInfo pass is illustrating the buckets incorrectly, perhaps using some kind of algorithm that doesn't take into account some significantly varied CPU threads. Still it doesn't effect the other passes so, I'm going to let it go... spent enough time trying to understand this. When I originally posted, I was hoping someone would go "oh yeah it's just this..." sorted! Nevermind, Thank you all.

 

I suggest you contact chaos. I not experience anything like that.