RTX Support

[Neil_Cross]

Nvidia recently announced their new Quadro and GeForce RTX cards which ray tracing and tensor cores, the major thing being real time ray tracing support on the GPU. 

If you're able to comment, do you guys have any plans to implement this into Vred, it could potentially be a game changer for visualisation software? If so, any idea on a timeline for support?

[michael_nikelsky · Accepted solution]

Hi,

sadly I am not allowed to say much about it so let´s just say we are doing active research on it 😉

Kind regards

Michael

[Neil_Cross]

Gotcha 😉 Thanks!

[michael_nikelsky]

Since not everybody was able to attend GTC Europe: We just showcased an early beta version of VRED running path tracing on two RTX 6000 at the Nvidia booth the last two days. So I guess this will answer your question a bit more clearly now 😉

[Neil_Cross]

Is there any footage/material from that event showing it?

[michael_nikelsky]

Nvidia had a film crew there but I have no idea when or if or how they are planning to release the footage.

[Neil_Cross]

I can't even begin to explain how frustrated I am with literally everyone over this RTX product launch... I pre-ordered a 2080Ti on 5th September, it's paid for, but the ETA is now 30th November...!!! Because lack of stock, somehow.

Whilst in the mean time I'm seeing gaming Youtubers with 2 or 3 of them messing about with pointless synthetic benchmark battles complaining there's nothing they can do with them. 

If, and it's a big IF I do actually get this thing on the 30th November, I'll have paid for it for 3 months already.  I refuse to believe that demand is outweighing supply to this extent for a product at this high of a price point.

/RantOver

Soz, that had nothing to do with what you said but it's really p*****g me off now.  But congrats to you guys though for getting this technology implemented so quickly, even if there'll be so few end users able to actually adopt it.

[Anonymous]

I have a ASUS RTX2080Ti and CAD2019 don't work. its the problem because the graphic card is not compatible.

[richardlevene]

---

@michael_nikelsky wrote:

Since not everybody was able to attend GTC Europe: We just showcased an early beta version of VRED running path tracing on two RTX 6000 at the Nvidia booth the last two days. So I guess this will answer your question a bit more clearly now 😉

---

Blimey I am only just seeing this comment. Very interesting! indeed!

Nice to see some love to the offline rendering. Not that we can afford to buy 2 of those RTX cards.

Are we able to send you scenes to benchmark render speeds against compared to our CPU render times?

Also sorry to hear Neil about your long wait for the card. Those youtube gamers are a big deal I guess. Much more important to get the views than you to get get your work done. Gotta get the word out so people who will actually buy them can't 🙂

Best,

Richard

[Anonymous]

Can someone help me with Autocad's error, and I think it does not work with the Asus RTX2080TI

Thnaks for your help.

[michael_nikelsky]

You are in the wrong forum, most people around here don´t know anything about AutoCAD. Try this one: https://forums.autodesk.com/t5/autocad/ct-p/8

Kind regards

Michael

[Neil_Cross]

I got my 2080Ti in the end, I had to cancel the order and buy a founders edition... I've actually got two as I had to RMA the first one and need to send the second one back when I get back from AU.  Can't say I've opened AutoCAD on it yet but I see no reason why it would conflict.

Anyway, on topic... I went to the Nvidia booth yesterday at AU and they have the RTX demo set up in Vred with the Mercedes car real time ray traced.  My first impressions are to be slightly impressed, but also very deflated.  I know it's a very early software preview, but Nvidia have TWO of their Quadro RTX 6000 cards powering that demo - that's like $20,000 worth of just graphics cards, and it's struggling for decent visual fidelity and smoothness.  Based on that, even one top tier card won't do the job.  I'm hoping that optimisations in Vred and the drivers will improve prior to RTM of the software that but if two RTX 6000's aren't capable of 60fps clear sharp ray tracing then there's not much hope for anything less!

[richardlevene]

 Cool that you saw the demo and gave some feedback.

But was that last sentence a joke Neil? Because obviously it still gives a lot of hope for the future no?

They were in raytrace mode right? And in Full Global Illumination Mode? Do you know if it was set to lower trace depth levels? What resolution?

What fps was is getting then?

Richard

[michael_nikelsky]

I think until we reach 60fps there will be some more hardware revisions necessary. Even what is shown in games with raytracing is currently at about 30fps with mostly rasterizing and some raytracing effects in FullHD, it is a long shot away from full path tracing.

Since I know the demo let me answer the question about it:

So it was a dataset with about 33 million triangles in full path tracing, ibl, and i guess medium realtime antialiasing (so 4 samples per pixel) and the nvidia deep learning based denoiser running in FullHD. I am not entirely sure about the trace depth since I don´t know which revision of the file they are showing, but it should be either 8 or 16 bounces, depending on the version of the file. However, higher trace depths don´t really affect the performance that much anymore.  

The 4 realtime samples are needed for a better result from the denoiser but of course they cost a lot of performance. Especially interior shots are very expensive with this setup, so I guess the frame rate there was probably around 1-2 fps and 5-6 fps on exterieur. If you only use 1 path per pixel you are probably around 5 fps in the interieur and about 15fps on exterieur with the denoiser and 30fps without it. You do want the denoiser though because you can get final quality renders with a fraction of the samples, in many situations something like 32 or 64 samples is enough.

Still take these numbers as what it is, the current state of the development. There is no way of telling where this will end up since everything is still changing a lot, VRED, Optix and the drivers. This is a technical preview and also first gen hardware, that´s it.

Kind regards

Michael

[richardlevene]

Thanks for the info Michael.

When you say "However, higher trace depths don´t really affect the performance that much anymore" this relates more to exterior shots though right? Because when using Full GI on an interior the higher trace depths has a huge impact on render time as that seems to impact how many indirect bounces there are as well. There is not a separate control to differentiate between an indirect lighting bounce and general reflection/refraction depths.

Sorry I do not want to derail the conversation with this.

Anyway so back to the demo..I always find it hard to gauge the performance of it by words so it would be nice to see it if you could upload a screen recording demo. I just think of the vray next rtx demo as what the performance would be like as a comparison.

[michael_nikelsky]

I actually mean interior as well. But I think I was not clear about what I was trying to say. The difference between 8 max bounces and 16 max bounces in an interior is usually not that huge, the difference between 16 and 32 max bounces is even less, simply because only very few rays survive this long unless you are in a snow environment where . Also, in a normal interior you will face quite a challenge to see the difference between 8 and 16 or 16 and 32 bounces. Of course there are exceptions, headlights for example, especially with light fixtures. But with the exception of these light guidance fixtures even for normal headlights 16 bounces is often enough.

Sadly I am not allowed to post any videos, especially not from this dataset.

Kind regards

Michael

[richardlevene]

Well normally for interiors we have to reduce the number of trace depths down to around 3 otherwise render times are really high (when using FULL GI) and you need the full gi for the extra indirect bounces especially for brighter leather materials. Typically we either override the trace depth on the brighter leather materials rather than globally but it would be nice to have a global override for separate indirect bounces so they are not always fixed to the trace depth. To keep it simple the default is that they are fixed to the trace depth, but then you have an indirect bounce override in render settings that will allow their to be less indirect light bounces on all materials.

Sorry for derailing the conversation Neil

Richard

[Neil_Cross]

Hi Richard - no problem.  Regarding the future comment, I was considering the future in that I said I hope the future software & driver optimisations will allow this technology to be usable with lesser hardware configurations, as not everyone will be able to justify two hero level Quadro cards.  However after speaking with Pascal at length at AU, I'm not so concerned about that any more after better understanding who the overwhelming majority of Vred customers are...! I have a better understanding of the wider market now so my point of view on this has shifted as of today.

Anyway, regardless, I'm excited to see this bleeding edge tech develop over time.

[richardlevene]

Oh haha so you just learnt that the main clientele are huge automotive companies with deep pockets 🙂

I always see it as new cards will be affordable for us in a couple of years.

Best,

Richard

[Neil_Cross]

Well kinda yes and kinda no, there was more to it than that but either way I won't consider it much of an issue from now on.

Plus - after seeing the recent RTX gaming results, it's quite apparent that this is either a tech preview generation or an intentional seriously limited hardware package giving them scope to increment on it for the next 10 years.  I wouldn't be surprised if they already have a die with 10x RT and tensor cores ready to rock... but releasing the first gen with less allows them to spread this out over the future releases for that illusion of continued development and evolution.

Just removing my tin foil hat now.