Rt gpu with irradiance map please!

[tashko.zashev]

Unfortunately not yet. We will post here when we have some good news on that topic.

[3LP]

Thanks, hopefully it wont be long, it's a feature that would really be a game changer, as much as IR was when it came to CPU...

Stan

[Sbrusse]

Hey guys,

still no news about this I'm guessing?

Even if it's "only" loading a IR map that has been pre-calculated in Vray ADV first.
Just need to replace the BF primary bounce by IR

Really looking forward for this,
Thanks

[tashko.zashev]

still no news about this I'm guessing?

No, not yet. Will post here when we have any good news.

[Sbrusse]

Sure,

Just checking out as it been nearly a year and a half the question was asked and Vlado did mentioned that only loading a IR map would be pretty easy to implement.

Thanks

[tashko.zashev]

Sure,

Just checking out as it been nearly a year and a half the question was asked and Vlado did mentioned that only loading a IR map would be pretty easy to implement.

Thanks

Yes, that's true but this task was in very low priority though. I moved it a little bit up

[super gnu]

yay, please nudge it up a bit. as mentioned in this and other threads, irradiance map really makes rt gpu insanely fast (as can be demonstrated by setting the gi to LC/LC and checking the speed.. it would mean rendering flythroughs on a single gpu would be practical. )

if you guys didnt already do that test, i recommend it.

[Sbrusse]

I most certainly second that as GPU is great, but it still isn't as efficient as CPU with IR,
bringing IR to GPU will make everyone turn they head through RT and I have no doubts it will be one of the major advances that RT has had since LC had been introduced, and we all know how much that did!
Nowadays, with 60 fps 360 stereo animations, we need more than ever a fast way to render out with GPUs

Please bump it up as much as you can

Thanks

[padre.ayuso]

OK, so just saw this and I have to give my own input.... With Denoiser, it actually takes longer for IR to process than just go ahead and do LC+BF in GPU RT. Now that may depend on your GPU, but with the 1080 now in the market, is 15 to 20 minutes for a frame at 3K resolution too long? Now, the final image with the denoiser is pretty close to what I was getting with IR, except now I don't get those damn blotches... Just my own input here, since I have dropped IR about 6 months ago, BF definitely gives one a very nice result without too much fidgeting around with parameters and testing. And I do know that Vlado is also trying to convert everyone into BF believers... I was one of those that have been converted.

[super gnu]

you are definitely right, to an extent, but consider, in my testing, when rt GPU does not have to do the gi calculations (is using a stored lightcache for primary and secondary) in my tests, its 4-5 times faster. add the denoiser to that, to remove grain, and your looking at some *seriously* fast renders.

obviously you have to set up the imap properly, but hey, was the only way until recently, its not that hard!

also, of course for stills, or animations with moving objects, the calc time of the imap (its not been updated for years) is not much (or at all) quicker than new BF renders, despite doing much less work ( another subject of course)

HOWEVER, for flythroughs/static animated scenes. , using multiframe incremental, it is still many, many, MANY times faster than doing it with BF. especially since you can use DR to calculate your imap on multiple machines.



choose the right weapon for the job and all that.

[Sbrusse]

With Denoiser, it actually takes longer for IR to process than just go ahead and do LC+BF in GPU RT.

That's VERY interesting, can you explain me how you have been able to get ANY data about how long it takes to process a image with IR/LC against BF/LC as IR has not been implemented in RT GPU AT ALL, YET?

Concerning the 1080, it has been proven by Blago himself that it's ONLY about 3% faster than a Titan X or 980 Ti, so it's not like that card revolutionized anything.

Your conception of what is "long" or not is very personal and completely irrelevant. How can you judge what our needs are?
You might have few stills to do overnight and you will be happy with your render times using BF/LC, that's great for you.
We have minutes long stereo 360 degrees 60fps 4k animation to render, that's somehow a little bit more demanding that your 3k render...

From my early test with what I CAN actually test, is using LC as primary bounce (LC/LC) and from those tests, I found that LC/LC is on average 10 times faster than BF/LC.
Now, tell me, is denoiser REALLY bringing your render-times by 10x down?

If it's 10 faster for you with the denoiser, assuming IR as first bounce will be the same as LC as first bounce (as they are both precalculated GI solutions) don't you want to be able to render 10x faster with IR and then again 10x faster using denoiser on top?
Jesus, that would be 100 times faster, wait a sec, 15min for a 3k render, that's 900 seconds, so divide that by 100, oh that would make your 3k render done in 9 seconds, sure you don't want that...

Mate, IR has proven over the last 10 years to be a great solution and is still widely used. If it's well set up, you hardly see any difference with BF, and it's not for those 1-2% pf difference in shadows that you can afford getting 10x more render times.
Yes people like BF, and I do too of course, but there are constrains in the business that force you not to be able to work as a little freelancer that does 1 still frame every two weeks and render overnight his 3k render on his home computer where 15-20 minutes for a 3k render is more than good enough.

Having said that, again, I would be very pleased if you could show me any data/benchmarks/scenes that have been rendered with IR on RT GPU.
Thanks

[padre.ayuso]

That's VERY interesting, can you explain me how you have been able to get ANY data about how long it takes to process a image with IR/LC against BF/LC as IR has not been implemented in RT GPU AT ALL, YET?

Having said that, again, I would be very pleased if you could show me any data/benchmarks/scenes that have been rendered with IR on RT GPU.
Thanks

If it works for you, great, I was just giving my own opinion, as I said. Not meaning to make anyone upset here. But then again, IR RT GPU? I thought you said this is not ready yet, so not sure what you are asking me.

Then again, 9 seconds for a render? I wish that was true, unfortunately, the file doing all the translations, setting up, LC, calculating, saving, denoising, etc., (even in RT GPU with BF+LC) would take longer than 9 seconds, perhaps this is something that can be made to take less time, but pretty good already.

The 1080, for those that have not bought a Titan X; it is not faster than a Titan-X, I am aware of that, but it certainly is cheaper! And you get comparable speeds at $400 less, for some people that's not much money, for others that is worth the investment. Then again, as a note of warning, that comes with a loss of 4GB of VRam, for anyone out there looking into GPUs.

But hey, if IR works for you and as you said, you are knowledgable on using it for the past many years, then great. My experience, personally, has been less than satisfactory. You have also others such as Peter Guthrie and Vlado himself that promotes moving towards BF rather than staying in IR, but as said, again, it is a matter of preference. If I have upset you in any way, please accept my apologies, it is not my intention. If Chaos will move IR over to RT GPU and this makes your workflow a lot faster, then again, cheers for you guys and Chaos.

Best
alex

[super gnu]

ive been offline a while.. logged in for the first time in ages and i see.. vray 3.5 appears to support loading irradiance map for rt gpu?!


wohooo! i shall be testing asap.