GPU - hopping on the bandwagon / best bang for buck?

[Vizioen]

Hi,

if I would be interested in RT rendering and I need to upgrade my setup (Quadro K4000), should I do it now, or is there some new architecture on the way. Or should I wait for Vlado to finalize IPR and just stay with my dual xeons. I don't like to be limited (RT doesn't support a few plugins I like and also having to fit everything in the VRAM is quite a downside for me). Currently I have one Quadro K4000 and it's insanely slow, for about anything even simpel materials, progressive rendering is faster. SO GPU rendering on this thing is a no no. My budget is around 1000$/€, I don't know how far that will get me. But I want to know if I'm in the right ballpark or if I should wait for a bigger budget and/or IPR.

[savage309]

Currently the best GPU for rendering is Titan X Pascal. It is near the price as well.
How it will compare for your needs against the dual Xeons (they can be really powerful) is hard to say. I would suggest to wait for and test the IPR. You can use RT CPU with the current stable builds as well.

[Vizioen]

Currently the best GPU for rendering is Titan X Pascal. It is near the price as well.
How it will compare for your needs against the dual Xeons (they can be really powerful) is hard to say. I would suggest to wait for and test the IPR. You can use RT CPU with the current stable builds as well.

Using either GPU or CPU as an interactive render brings my computer almost to a halt, the feedback is very slow and everything just starts lagging. And that's not because of Vray, I know because I tried them other renderers that are supposed to be very interactive (corona, fstorm) and they are just as slow so must be my GPU.

I would test the IPR already but I don't have access to max 2017 and I don't think Vlado found the time to make it work for 2016 yet.

[Nicinus]

Another alternative that I'm contemplating is 2x1080s instead of the one Titan. That gives another 2000 cores (about 5,000 compared to 3,072 on the Titan) and better interactivity in the viewport with RT. It is however 'only' 8GB compared to 12GB. Price about the same.

I haven't seen any benchmarks comparing these alternatives yet though.

[Vizioen]

Another alternative that I'm contemplating is 2x1080s instead of the one Titan. That gives another 2000 cores (about 5,000 compared to 3,072 on the Titan) and better interactivity in the viewport with RT. It is however 'only' 8GB compared to 12GB. Price about the same.

I haven't seen any benchmarks comparing these alternatives yet though.

That would be my next question, only one of the GPUs counts for the VRAM?

[Morne]

That would be my next question, only one of the GPUs counts for the VRAM?

Yes with the "gaming cards". Apparently later on with the Quodros the VRAM will be added together as many cards you have. But Quodros are much more expensive

[glorybound]

Over on the Autodesk forum they are talking 1080 and MAX compatibility issues.

[Nicinus]

Hmm, that doesn't sound so good. However, I saw one thread without a response. Have you seen many of them?

[glorybound]

Hmm, that doesn't sound so good. However, I saw one thread without a response. Have you seen many of them?

well, I mentioned getting one and the moderator mention that there are comparability issues. I'll find the king and post it.

[mitviz]

over on the gpu forums it seems to be working just fine with renderers like octane and fstorm in max

[savage309]

Just to note that it works very well with V-Ray GPU. We even didn't had to make new builds to support it and it works even we as old build as V-Ray 3.30.

Best,
Blago.

[Nicinus]

Yes with the "gaming cards". Apparently later on with the Quodros the VRAM will be added together as many cards you have. But Quodros are much more expensive

There is always later on. I'm hoping to see RT supporting out of core rendering (Blago doesn't like that term but I don't what else to call it) where it can use regular RAM for paging, which seems to work in Fstorm or Redshift, can't remember which.

Right now I have 6GB, so doubling that to 12GB would of course be nice, I'm just wondering how long that would keep me happy. Right now I seem to be able to fit most things as long as I don't scatter proxies or accessorize excessively, and I doubt 12GB will help those situations anyway. Besides, at least on my current card, when scenes is starting to become that big it takes such a long time to start on RT that I might as well use progressive in Adv.

[vlado]

There is always later on. I'm hoping to see RT supporting out of core rendering (Blago doesn't like that term but I don't what else to call it) where it can use regular RAM for paging, which seems to work in Fstorm or Redshift, can't remember which.

It's Redshift; Karba says he has no plans for out-of-core rendering for FStorm (I'm pretty sure he'll change his mind though). However even with Redshift, if it gets to paging out to CPU RAM, it slows down quite a bit to the point where CPU rendering might make more sense anyways.

Besides, at least on my current card, when scenes is starting to become that big it takes such a long time to start on RT that I might as well use progressive in Adv.

We would be very interested to take a look at such a scene and try to optimize the loading times.

Best regards,
Vlado

[Nicinus]

It's Redshift; Karba says he has no plans for out-of-core rendering for FStorm (I'm pretty sure he'll change his mind though). However even with Redshift, if it gets to paging out to CPU RAM, it slows down quite a bit to the point where CPU rendering might make more sense anyways.

The whole discussion is interesting and I admittedly have very little practical experience of it, but it seems to come down to choice of strategy. If I had a dual Xenon with 36 or 40 total cores, progressive and IPR would probably be very convenient most of the time, but if I instead have 3-4 1080s or Titans with something like a 5960, it would probably be worth dipping in to CPU RAM now and then, especially if the VFB could alert me of this so that I could switch off a layer or two.

[dimension3plus]

I just read the thread and was wondering what IPR is?