Messages

4276

Hardware / Re: Corona - Titan black SLI or Titan Z?

« on: 2014-07-02, 14:06:57 »

Since you want "much more" rendering power, than current hexacores don't cut it. But if you would find the xeons to be over budget,
you can also wait until mid September, where Haswell-E i7 arrive. But, only the highest model 5960X is going to be Octacore @ 3.0Ghz for 999 euros.
Plus 250 euros for new motheboard (new X99) and possibly for the sake of novelty DDR4, but that will have its heavy price (for very little benefit in CGI). But this solution would bring you roughly 30perc. more performance
of i7 4930k. i7 5930k is even bigger dissapoitment, it's not even worth waiting for, unless you are extreme overclocker/game enthousiast, and you really want DDR4 memory.

But since you were looking for super-overpriced monstrosity like TitanZ, then you probably can afford Xeons and should :- )

4277

Hardware / Re: Corona - Titan black SLI or Titan Z?

« on: 2014-07-01, 21:53:07 »

If you don't like overclocking, the Xeons are even better for you, as they don't give you that choice of luxury :- ).

Anyway, Opterons are out. They haven't been updated for very long time, and AMD had withdrawn from the high-end market since, this includes server platform.
If you mean the last high-performance architecture they had, the 6xxx range, than 64 cores equalled up to 30 in Cinebench score. Similar to Xeon range but with following drawbacks:

4xCPUs draws 4x140W. Compare it to 2xXeons at 115-130W, that's twice power.
They require server based OS system to run. Not a big problem, but inconvenience for workstation.
The architecture isn't so universal as the Xeon range. Even though syntentic tests shows potency, this didn't prove in practice with many renderers. Vray for particular, had big trouble feeding the cores and recorded subpar performance. You
can search Vlado's answer to this on Chaos forum. Advice:Stay away.
Extremely poor single-threaded performance. We're talking, half of that per core to SandyBridge architecture in Intel, at clock up to 2.7Ghz. That's absolutely useless for any workstation task. It's like buying Celeron.

tl,dl: It looks nice on first sight, but if something looks too good to be true, that it probably isn't. Opteron range isn't.
You already bought the worst AMD ( I really like the company...but, you basically have to ignore their cpu range recent years) cpu out there, don't repeat the mistake.

4278

Hardware / Re: Corona - Titan black SLI or Titan Z?

« on: 2014-07-01, 20:35:23 »

Well, if you had money for double 780, I figured out had atleast one hexacore i7.

I was wrong though :- ) And you have very slow AMD cpu so. But why do you need second PC ? Why not just swap motheboard + CPU ?
Keep you dual 780 for gaming and profit ;- )

I presume you have the highest bulldozer, i.e FX8350 ? If so it has roughly 60perc. overall performance compared to i7 4930k, so while switching to top i7,
you would render faster, but not crazy much more.

Relevant cinebench (11.5) scores (non-overclocked):
FX-3850=7
i7 4930k=10.6
i7 4930k at 4,4Ghz=13

So if you keep your i7 clocked pretty hight, you would be able to render at roughly twice the speed compared to your current setup.

For example my Xeon nodes:
2x2680v2 at 3.1Ghz(turbo)=28

You can buy Asus Z9PE-D8 + 2x Xeon 2667v2/2680v2/etc.. for about 3400 euros. [or buy much cheaper ES versions but...that's risk]
This would give you pretty ultimate machine.

4279

Hardware / Re: Corona - Titan black SLI or Titan Z?

« on: 2014-07-01, 18:11:22 »

So much bro-science in this thread.

1)"Xeons not worth the money". Not true, people just regurgitate common nonsense. If you do your math correctly and know about hardware bit deeper, you will understand it will come to same. Without having multiple machines. Of course, up-front investment is larger, but overall performance for value
will be close to identical.
2)"Suggestions of outlier boards (Sabertooth/ROG/WS). Currently, any LG2011 boards support almost identical stability under over-clock. Even the cheapest old "Vanilla"grandpa P9X79. The overclockability is much more down
to how lucky you are with your Ivy-bridge chip sample. You can run all of them on 4.4GHz if you wish and don't care about voltage (for some you will need to go over 1.4+ and will sacrifice stability and longetivity, but this doesn't affect the board you choose !!).
You need ROG if you need advanced feature set for overclocking ,and by 99perc. chance you don't. Sabretooth is hardproofed pseudo-army gimmick that will give you longer warranty. WS is E-ATX size Vanilla P9X79 with possibility of 4 GPUS. If you don't plan on 4GPUS (for Corona ?) then you don't need WS board. Buy regular old vanilla P9X79 and update bios, or buy newer Deluxe X79 (only "X79" not P9X79).
3)"Close-looped water cooling system". Not that amazing choice. Water cooling=/= Water cooling. Unless you plan to cool few (2 to 4) GPUs, you don't need water cooling in your PC which is the only benefit of it. Close looped systems that connect only your CPU to
outside fans like Corsair Hydro i100/i80/etc... Kraken XTZ/etc...are mediocre (Kraken) to trash (Corsair). The cooling capacity is same level as top air cooler like Noctua NH-D14/15 but it's louder (cheap pump, cheap fans you have to position at top of case) and more complicated. There is zero reason to go this route for pure CPU based workstation.
4)Save your money and don't buy 1TB SSD to your workstation. 256 or even 512 is more than you will need and store the rest on network. If you already plan to spend so much money, the PRO is better choice the EVO, it's small price difference only.
5)You don't need 1200W PSU ! If you plan to use GTX760 and overlocked i7 4930k than even 600W will be more than enough. If you want to be "safe" for future, ok, buy 720-750W but know it's overkill. Silent, passive, platinum grade are better atributes to look for.


You already have PC with dual780 ? What's the problem with that for Corona ? If you need purely more performance than get budget of 5000 euros and buy dual-xeon. But you don't seem to need a new workstation. Maybe a simple node would make you happy.

Save your money, think more and do more research.

4280

[Archive] Alpha Builds / Re: Alpha v7 released!

« on: 2014-06-30, 13:51:53 »

That's impressive list of additions !

Also, thanks and gratulations are given but I wanted to also say it's great that you bundled DeadClowns' script, this is the feature that got me to try Corona first without having newborn pains that often comes from switching renderer.

4281

Work in Progress/Tests / Re: "Ascending" - Corona for C4D 2nd impressions

« on: 2014-06-29, 15:00:57 »

Looks really interesting.

I hate those AXYZ models, such stupid poses. Don't you want to try animating a moving character ? And giving him strong motion blur ? Could give it a really nice feeling.

4282

[Max] General Discussion / Re: GGX - best way to recreate it for all metals, problems translating to Corona

« on: 2014-06-25, 18:14:09 »

Very quick test of mine, Blackened steel.

Layering works fine. Although, slightly differently than VrayBlend, but I can't put my finger on how.

Some observations: Corona keeps really nice highlight energy in corner almost unnaffected by glossiness. This seems almost strange, but maybe more correct actually. This can be seen on page 1 too, on the lion,
which looks better in Corona even with single layer. So I can confirm this. Same curve and eye-balled glossiness behaves differently in Vray.

The glossy curve in Corona is so much more linear than in Vray that it's almost hard to produce 'rough' (low gloss number) surface haha. With Glossy parameter 1.0 in Vray, driving only with textures, I kept the material contribution
in Vray 0.2 to equal almost 0.8 in Corona. I sort of don't know how to convert these to even remotely match. Not bad, just different, and puts point that if the material matters, just build it straight up, don't convert stuff.

Anisotrophy is so different. Seems like what is by default in Vray (Z-Axis, 0 angle) equals to Corona 90 angle (no axis can be chosen), but it still behaves differently. Can't stay bad or better, just, different, so I couldn't manage to match it well.

Overall, I think I am satisfied and can't wait for more advanced BRDF like GGX !

4283

[Max] General Discussion / Re: GGX - best way to recreate it for all metals, problems translating to Corona

« on: 2014-06-24, 22:19:38 »

Thanks Daniel, I totally forget about HDRLabs existence. The way the one you chose shows the highlights completely rocks.

4284

[Max] General Discussion / Re: GGX - best way to recreate it for all metals, problems translating to Corona

« on: 2014-06-24, 19:58:48 »

Nice :- )

I kept his ball too btw, no reason to change it. It's the nicest ball I've seen to date so why even bother.
Seems like great HDRi you used there, mind say what is it ?

4285

[Max] General Discussion / Re: GGX - best way to recreate it for all metals, problems translating to Corona

« on: 2014-06-23, 15:01:58 »

You deserve many beers Gestalt !

4286

[Max] General Discussion / Re: GGX - best way to recreate it for all metals, problems translating to Corona

« on: 2014-06-22, 20:55:51 »

Hi Gleash, I didn't doubt you :- ) I didn't try to translate multiple blended materials with VrayBlend into nested material in Corona. I will do so as soon as I can ! Maybe it really doesn't work, which would be sort of odd.
I create all my materials in Grant's way since I follow him early and very much enjoy the results. I did it before too, blending metals, but not with additional curve to glossiness as he shown. Not scientific, but nifty !

Also weird that you get different energy for metals using 0 diffuse between Vray and Corona. I think I found it to be consistent but I didn't check.
Did you try comparing both unclamped ? Vray with Subpixel OFF and Corona with MSI 0 ? Just to see if you're not loosing some highlights.

But first, definitely use the full 0-1 spectrum :- ). At 90angle incidence, almost all of them (not all, some retrograzingly fall close to 0, very funny, like some fabrics,etc..) get there.

I sadly don't have much free time to test it, I am waiting for Corona's blend material, once that will be done, we can do full 100perc. comparison and I am greatly looking forward to that !

[btw,off-topic: Grant's materials are harder to render because 3 blended materials create 3 times more sampling (each VrayBlend compoment samples separately), but it's survivable. With today power.
But definitely doesn't cook your CPU, 90+C are overkill for long time, you don't need water cooling, just well sitting AirCooler. I have 40C top with simple Noctua NH-D14]

4287

Work in Progress/Tests / Re: beach house

« on: 2014-06-22, 18:10:47 »

heh I see. Sometimes trying to come with easy and clean solution we get caught up messing up something instead.

4288

Work in Progress/Tests / Re: beach house

« on: 2014-06-22, 17:16:55 »

You won't be able to correct this HDRi to sun, it doesn't have sun :- ) if you boost the intensity of bright parts (sun behind clouds) it might look bit awkward.

Since I used the same HDRi, I added CoronaSun with low-intensity to some interiors as well. It plays nicely together imho, no reason why to avoid it.

4289

[Max] General Discussion / Re: New Frame Buffer icons

« on: 2014-06-22, 15:08:41 »

Region !! Nice to see !

4290

[Max] General Discussion / Re: GGX - best way to recreate it for all metals, problems translating to Corona

« on: 2014-06-22, 14:02:17 »

Lol, dear god, I am glad to see these threads pop up :- ). Just yesterday night I told myself I will need write HUGE FUCKING BLOG why we want better shader, complex BRDF, layered speculars inside single materials, blended material, roughness affecting both diffuse and specularity or atleast complex IOR.
[rendering all those examples will give me headache probably...I wish someone talented could instead...but I really want this, like over9000 want this]

Yep, if Keymaster needs motivation, I will do so.

But frankly, Vray already has GGX in nightlies, countless of users have confirmed once starting to creating more complex materials, that this is the way to go.
I might test those two plugins from Sergey as well, for now for 3dsMax there are only GGX (Sergey's GGX has 'power' parameters for tail, Vray not yet), ABC (full peak/tail),
but he doesn't have time to convert the most impressive last one, ShiftedGammaDistribution which has all of above, plus 100 MERL Presets.

Almost every renderer has the same featureset now, or building towards it slowly. This is one thing that can mix the cards, bring back some new fun, and revolutionarize the workflow.
This should be much bigger priority imho than integrating bullshit from past.

When I have time I will show my examples of metals (layered metals such as Oxidized Blackneeded Steel,etc...), but in meantime I think Grant does awesome job (I have evolved the workflow a bit too though )
And of course...the above video !! Everybody throw some money on Sergey so he can port it to Vray for now atleast :- ) Or hire bit of his time so he can port them to Corona.
I mean come on, this stuff can definitely give brutal edge on competing renderers like nothing else can. It's already everywhere else ( Disney, Pixar, Unreal4,....).


Btw Gleash: What do you mean with the horizontal stretch ? If you refer to Refractive index, it goes from 0-90 angle of incidence, therefore it's ok to interpolate it identically between 0-1. Same stuff. Or maybe I don't understand you correctly.
I am surprised you mention regular blend doesn't work the same as VrayBlend. I will have to try mysel, did you put the fallof curve into mask slot ?
999 IOR in Corona should be equal to no-IOR in Vray. But I didn't try.
Using simple IOR (8-40) for Metals will always give shitty metals. It looked wrong to me eyes long before I've see Bertrand mix his metals 2 years ago (although he only eye-balled it, didn't give the scientific reason) and of course, Grant few months ago.
The difference can be from small, to drastic, the latter when material features quite strong "retro-grazing" reflection. So the curve (without doing the "Well, what random value will I put today into specular slot?") in reflective slot dealing with intensity from 0-90 angle is just superior. Chrome looks like Chrome, Gold looks like Gold.

Disney-PBS_BRDF white paper http://disney-animation.s3.amazonaws.com/library/s2012_pbs_disney_brdf_notes_v2.pdf

Imho the best stuff on this. On page 9/10 is shows the correlation between albedo and specular in GGX/Roughness type of shader and the various curves fitted from MERL data.
Having their version of material would be a dream. What is funny it's like 90perc. identical to what I have in Unreal4 [metalness/roughness/GGX/base reflection]. If current realtime engines share the same model high-end
companies like Disney, MPC, ILM (all have PBR/PBS materials), yet we're still stuck with unlinked specular/glossy material harkening to prehistoric era.