Author Topic: Corona GPU rendering  (Read 84386 times)

2015-03-09, 16:52:33
Reply #30

juang3d

  • Active Users
  • **
  • Posts: 636
    • View Profile
I simply wanted to counter-argue against dismissive statements that these cards are expensive, contribute to economically unviable solution and that gp-gpu rendering is still in limited infancy.

I agree with that, the only thing is that it is not a viable solution for me, but it could be for other people of course :)

Quote
And as matter of fact, that slowly starts happening now.

The key thing here is "slowly", it depends on Nvidia and what pace do they want to keep for releasing their cards and how the technology evolves, what I ask myself is if Intel will counter attack again like they did with Embree in a future CPU family, I doubt the GPU's will stand as more powerful as is, but rather there will be some kind of leveling over the time, the thing is IMO what to use in what moment, and how much can you spend to mutate your environment and adapt it to each evolution of the state of the art.

Cheers!
« Last Edit: 2015-03-09, 17:09:09 by juang3d »

2015-03-09, 18:08:34
Reply #31

juang3d

  • Active Users
  • **
  • Posts: 636
    • View Profile
Out of curiosity, check this:

https://forum.corona-renderer.com/index.php?action=dlattach;topic=559.0;attach=28939;image

It is from the benchmark post, RobSteady took the time to do this test:

https://forum.corona-renderer.com/index.php/topic,559.480.html

It's interesting that the benchmark was 3 minutes for a 5930K and 2 minutes for a similar version (in noise level, specially in the floor and the end wallof the room, and this with 2 Titan Z + 2 Titan...

Not a "valid" benchmark at all, but a curious test :)

Cheers.


2015-03-09, 18:44:57
Reply #32

racoonart

  • Active Users
  • **
  • Posts: 1446
    • View Profile
    • racoon-artworks
Apart from the fact that I find both images a bit too identical, Octane renders pt only, Corona images are Pt + HDcache. Yes, I know, "it's the end result that counts.." etc etc, but still, if this comparison would pop up in the Octane forum or if it the places were swapped, people would complain about it.
Any sufficiently advanced bug is indistinguishable from a feature.

2015-03-09, 20:17:48
Reply #33

Coronaut

  • Active Users
  • **
  • Posts: 131
    • View Profile
If you spend that much money on any hardware things will be fast no question about it, it is 4 titans...
And he has some details missing on curtains...
Anyways, corona bench is somewhat not optimized for dual systems... 2680v3 x 2 give smaller results than 5960x OC to 4.2ghz and it ias more then double faster in corona 1.0... so if one 5960x(4.2ghz) do this bench in about 02min:22sec realistic expectations are that 2x2680v3 do bench in half that time, and this system is much cheaper than his 3k GPU vs 3k CPU, even 5960x is very fast.
In both cases you will need same stuff(except gpu for rendering node that uses cpu) but you can't place 4gpus in shit MB or with slow CPU and that make additional cost that place nail in GPU rendering coffin(for now).
« Last Edit: 2015-03-09, 20:21:35 by Coronaut »

2015-03-09, 21:28:52
Reply #34

Juraj

  • Active Users
  • **
  • Posts: 4743
    • View Profile
    • studio website
but you can't place 4gpus in shit MB or with slow CPU

No, you can actually do just that. GP-GPU rendering isn't gaming, the CPU doesn't really come as bottleneck anywhere in pipeline as it would otherwise. And 4xPCI-E come on boards as low as 100 euro 1150 ones (any Z87 for example), since you don't need all of them running at X8/16 to get full performance. Almost any CPU provides enough bandwidth (with exception of strangely crippled 5820k perhaps).
It is actually cheaper to build 4xTitan-X build, than it is to build above-average Dual Xeon build (2680v3+ and higher).
« Last Edit: 2015-03-09, 21:32:38 by Juraj_Talcik »
Please follow my new Instagram for latest projects, tips&tricks, short video tutorials and free models
Behance  Probably best updated portfolio of my work
lysfaere.com Please check the new stuff!

2015-03-09, 22:04:57
Reply #35

Coronaut

  • Active Users
  • **
  • Posts: 131
    • View Profile
I know it is possible, but i am talking of minimum 5930(40 pci lanes) cuz of memory bandwidth needed to feed those gpus, run max etc... well you get picture, it would be insane using similar setup like you would in crypto mining.

Let's say 5930k+mobo(around 700e)+memory 32gb(400e) case and all other stuff: ssd, cooling etc.(500e) + 4 titans that are around?!(1.5k piece?) + psu(that have to be at least 1.5kw(250-300e)= 7900e
This will draw 1.5kw from socket under full load...

I don't know how much is your setup(i see you have something similar) but here is mine and it is lot cheaper from this GPU build...
2x2680 v3(2300e) + 530e(asus z10-pe w16 mobo) + 760e 64gb crucial ecc ddr4 + chief case(100e) + 2 Noctua u14s(120e) 240gb intel ssd 730(200e) + (ss 850w)psu 150e =4160e
No gpu...
This draw around 500w from socket full load(checked with UPS)
This is render node and even if i am to use this as WS it will have 780ti(more than enough) so how is this more expensive, it is cheaper, faster, more versatile and above everything else more economical.


2015-03-09, 23:06:02
Reply #36

Juraj

  • Active Users
  • **
  • Posts: 4743
    • View Profile
    • studio website
Oh, we're going so off-topic, and I have to work :- (, but I have to keep my part :- D

If you're commenting on my 2680v2 setup, it was bit more, I think about 5500 euros. including VAT., and almost exactly the same spec, just different brands (samsung ssd, fractal case,) somewhere. Anyway, the Xeons were pricier, where did you get 2680v3 for 1150 euro ? Even without VAT it would be cheap. Current average price is about 1800 euros including VAT. You could have gotten good deal (if so, congrats), but don't change your math because of it. Please be objective.

4xTitan-X build wouldn't be 8000 euros with VAT either... The price is also expected to be 1000 euros, not 1500...
So my math comes to be very similar 5500 +/- euros.

The watt loads would be sort of like you described, 500 vs 1500.

Now, I am not using any actual GPU rendering engine, so I can't and don't want to go into comparisons at the moment, but don't try to persuade me we're comparing equal performance....4x3072 Maxwell cores is absolutely different level of performance than 40 Haswell cores, even if no direct comparison can be made {note}. But really 2x2680v3 =/= 4xTitan-X, that's literally just....wrong. This gets very abstract though and the discussion looses point.

{ http://www.cs.virginia.edu/kim/docs/ispass11.pdf  "Where is the Data? Why You Cannot Debate CPU vs. GPU Performance Without the Answer"}
« Last Edit: 2015-03-09, 23:17:23 by Juraj_Talcik »
Please follow my new Instagram for latest projects, tips&tricks, short video tutorials and free models
Behance  Probably best updated portfolio of my work
lysfaere.com Please check the new stuff!

2015-03-10, 00:12:21
Reply #37

Coronaut

  • Active Users
  • **
  • Posts: 131
    • View Profile
Yes i agree, there is no real cross reference here apart the fact we are keep seeing more and more RS that are quite capable in providing same results on cpu/gpu, it is more of a Bike vs car stuff, someone can go around the world with bike and someone can't drive to the store it all depends on many factors.
I can't discuss publicly but this retail price for 2680v3 processors is insane... cheapest i did find was 1400e so this was very good deal(no it didn't fall of the truck :))
I can't find titan-z(i was discussing them since those are the ones used for octane RS) for less then 1.5k i know titan-x will be killer and cheaper but when it gets out, right now price is big mystery someone is saying 1k others 1.3k.
It is matter of some time(years) when gpu(or some other form of more compact architecture) will take over, just 2-3 years ago you could fry egg on mid range gpu, now you have beast that don't even get warm(over exaggerating :))
I used and tried few of GPU RS they are not as nearly "flamboyant" types you are willing to jump in bed when you lay eyes on them for the first time, but when you get them home... :D More of a plastic fantastic types that keep their teeth on night table and garter belt they bought from Top shop... but they know how to talk dirty and sometimes it is exactly what you want, but they are dyslectic so do not expect them to read you bed time story...

I tried 2x780gtx and it doesn't even come close to let's say 2x4930k, i am talking not only one gpu/cpu RS but several tests. I don't have titans but 2680v3 is beast and Vray and Corona go well with it.
12gb is joke, it cramps style of(few months in project) i don't care anymore(yes when you are listening to Ramones), no one will be willing to be extra careful with texture size, meshes etc. so it could fit in 12gb of ram... I am sure that 99% of people would rather save few days there as they wouldn't have to clean scene meshes and different assets... and have those extra days for longer rendering(but it isn't).
Already saw those papers, they are quite old... Anyhow lots of new stuff is coming soon, nvidia is already failed to fulfill their promises(last couple of years) things are rolling slowly and i am sure CPU giant Intel wouldn't let P4 fiasco to happen again same as AMD claim to have ace in their sleeve for 2016.
Things weren't so openly secretive for quite some time and who knows what might happen, but i am sure in one thing that GPU isn't only thing right now that wants piece of cake.
I think it is more of standardization of industry that will win at the end, that is why you can see UE starting to take of, it is based on something old more or less standardized, usually to change things you need more revolutionary approach and for now i don't see anything else except Corona to do this(no this is not fanboyism) i hate the fact i have to learn new RS again, i hate the fact i had to spent 300e on another piece of software, i hate lots of shit but i have to take bow to Ondra(and puke on his new yellow later pattent shoes while taking that bow and thank him at the end) as standardization based on monopoly is worst type...
All this seems to me like i saw it before, oh yes i did...
Sorry ondra i puked on your new shoes i will buy you new ones next year :D

2015-03-10, 07:11:21
Reply #38

philippelamoureux

  • Active Users
  • **
  • Posts: 218
    • View Profile
I have currently no fan in GPU renderers. Octane is moving at speed of....no speed, all attention is given to cloud, Brigade or some other funny gimmicks they are doing.

This is something that saddens me. I think Octane is a formidable renderer but the absolut lack of documentation/support/communication/tutorials makes it less appealing. The community may be small but the team is not putting much emphasis on communication to help it. Same thing for their cloud services, we don't know much details, don't know when it's coming, etc. They don't talk for god sake lol. The amount of times I've asked questions on their forums only to get no answer...

Another thing is that maxwell is (might be fixed now tho, haven't checked recent patches/bernchmarks) not performing as well as Kepler with Octane yet. It kind of suck that every time a new gpu architecture will come out, the renderer may not be updated fast, or even not at all...

2015-03-10, 10:59:09
Reply #39

juang3d

  • Active Users
  • **
  • Posts: 636
    • View Profile
Wasn't Otoy under agreement with Autodesk... There you have why they don't speak.

Rest assure that with that agreement all the Otoy business will be revenue oriented without thinking in the customer at all :P

In fact it's possible that in a near future you won't be able to acquire any octane license but you will be forced to go under the forced-saas model paying 100€/month per license (Autodesk thinks that 100€/month per node is nothing for everyone, who don't have a few thousands in their pocket to rent software monthly? Come oooon!)

Cheers.

2015-03-10, 11:21:29
Reply #40

RobSteady

  • Active Users
  • **
  • Posts: 45
    • View Profile
Hey there, CPU-competitors ;)

Some notes:
  • Power draw is 860W max for 2 x Titan + 1 x Titan Z + System (for big scenes with proxies etc.)
    I've heard that peaks could be much higher but I've never seen that so far (watching the power meter while rendering).
  • I got my Titan's for 900€ each and the Z for 1400€
  • CPU and GPU ram consumption behaves different: If you organize your scene a little bit you can get away with 6gb very easily.
    There is also the option to use system memory with a little bit of speed loss. Have not tested so far.
  • Support is......ok. Requests get implemented...and sometimes not. I think that's a pitty because the Max plugin could be much better (material converter/editor, light lister etc)
    We have a long request list running on the forum with very little feedback

Overall I think it's a great engine and a big relief (coming from V-Ray) but usefull requests are getting ignored for whatever reason.
I do like Corona as well but had no time to really test it.


« Last Edit: 2015-03-10, 11:45:42 by RobSteady »

2015-03-10, 11:33:54
Reply #41

RobSteady

  • Active Users
  • **
  • Posts: 45
    • View Profile
For the benchmark scene: are there any light-planes/portals in front of the windows? The Octane scene is only lit by daylight.
I will try to optimize the scene a little bit and also render with "Direct Light" which is not unbiased.
« Last Edit: 2015-03-10, 11:45:06 by RobSteady »

2015-03-10, 13:00:40
Reply #42

cosbu

  • Active Users
  • **
  • Posts: 15
  • civil engineer
    • View Profile
Hi all, I am user of both octane and corona, but not a high quality visualizer as most guys in the forums. My opinion is that both engines are great and very fast. But you can't do direct comparison, it's just a matter of choice for the user, depending on more parameters, like the modeling application. If someone have the money for let's say 4xtitan cards or more  I would bet on octane because I think the setup may be easier than a local render farm and you can have almost instant results . Octane's real time is looks a bit smoother but corona is new on this. On the other hand corona is very fast without the need to spend a fortune for GPUs+PSUs and there is the solution of cloud render farms for high end results.

Octane on my single 580 with 1.50gb ram was far outdated in comparison with corona on my i7 but then otoy did two things: the one was "coherent ratio", a mode that almost doubled the render speed and the other one was "out of core rendering" for textures only (they are loaded on system's ram if there is a need) . So now they are almost equal because I can do a clean 2k image in about 2hrs I think, in both engines and that gives me flexibility.
One funny thing is how less ram memory a scene consumes when you remove it from 3dsmax. I did a scene first in sketchup and then in octane standalone. It was almost 700mb in my gpu's ram. 3dsmax for corona took 7gb!
civil engineer

2015-03-10, 13:24:35
Reply #43

Ondra

  • Administrator
  • Active Users
  • *****
  • Posts: 9048
  • Turning coffee to features since 2009
    • View Profile
Sorry ondra i puked on your new shoes i will buy you new ones next year :D

do not worry, I stopped reading this flame war/fanboy outing-from-closet long time ago ;)
« Last Edit: 2015-03-10, 13:28:02 by Ondra »
Rendering is magic.How to get minidumps for crashed/frozen 3ds Max | Sorry for short replies, brief responses = more time to develop Corona ;)

2015-03-10, 13:49:38
Reply #44

borisquezadaa

  • Active Users
  • **
  • Posts: 614
    • View Profile
Another priceless post. I expect more to come now that Corona is going comercial. Is always refreshing to read this comments.
What i do with Corona My Corona post of random stuff rendering
WARNING: English.dll still loading...