Titan Volta and Daz3d
Deadly Buda
Posts: 155
As much as I search these forums and search online, I am still confused about the following:
1. Does a Titan V work with Iray and Daz3d currently? If so, is it about 2.5x faster at rendering than a 1080ti?
2. I see nVidia says Titan V uses nVlink, but I read conflicting reports. Is it possible to nVlink 2 Titan Vs in order to get 24GB of VRAM in order to render a Daz3d Iray scene?
3. For a large scene that used about 20GB textures and geometry, mostly close-up objects, would 2 Titan RTXs be faster than the above theoretical 2 Titan Vs in nVlink?
Daz 3D is part of 
Connect
DAZ Productions, Inc.
7533 S Center View Ct #4664
West Jordan, UT 84084
Licensing Agreement | Terms of Service | Privacy Policy | EULA
© 2025 Daz Productions Inc. All Rights Reserved.
Comments
The way Iray works with multiple cards is that each card renders the scene at the same time and then Iray combines their output to form the image. Having two or more cards does not increase your overall vram as each card will use what it has and no card shares with another.
I dont personally know about the nVlink but for cards that have a SLI bridge connected, Nvidia states that SLI should be disabled in Windows before rendering with Iray as it can cause issues.
Right, but I thought the whole idea of nVlink was so that applications could pool their memory, including iRay? SLI we know doesn't work, but I seem to hear conflicting things about nVlink in relation to Daz and iRay.
...the Titian Volta has neither the capability to use for NVLink or SLI.
https://www.techpowerup.com/239519/nvidia-titan-v-lacks-sli-or-nvlink-support
1) It does work but I have no idea about its render speed. You might want check the benchmark thread.
2) Titan V hasno NVlink connector.
3) RTX Titans are faster than the Titan V. For a scene that requires 20Gb you'll need either the RTX Titan or one of the Quadro's with more than that much VRAM.
On Nvlink and combining VRAM, it is theoretically possible for cards with Nvlink to do so but it requires that the software involved support it. Iray at present doesn't. Nvidia has never said whether it will be changed to do so once the RTX enabled Iray comes out sometime later this year.
Hi everyone, thanks for the info.
Yep, After more research, it looks like the Titan V has no nVlink support. The confusion comes in because Nvidia touts "Volta" series as having a super nVlink, but it was disabled in the Titan V and appears no way to mod it to do so.
After scouring those threads I'm still hard pressed to find any solid numbers on Daz3d + Iray. Other programs running Iray seem to show Titan V is the best except for one small thing...
...without nVlink you are stuck at 12GB of Vram, whereas Titan RTX has 24GB VRAM on board, and possibly more if nVlink is enabled for Daz Iray and you get more Titan RTXs.
So, if you are doing large scenes, it seems that the ideal situation is multiple Titan RTXs right now. I'm guessing the Quadros are too fussy for DAZ? I would just think that peformance just wouldn't scale well with those cards and Daz, and might be more problematic on a day to day basis?
...the RTX Titan also can take advantage of NVLink like the Quadro series does for sharing memory.
Not in DS they can't. The software being used has to enable that capability and DS/Iray doesn't.
If you're doing large scenes then you're probably far better off with a RTX Quadro 6000 or 8000. There ar some features enabled on those cards that make them far more suited for professional level work, for instance if you are using something else as a display adapter then Windows reserves no VRAM on the card which Windows does for all other graphics cards even when they are not connected to a monitor.
The RTX Quadro 6000 is functionally identical to the RTX Titan but the RTX Quadro 8000 has 48 Gb of VRAM for the same price, roughly, as 2 RTX Titans.
Interesting. Let's use the example of a 20GB scene. What renders faster today, 1 Quadro 8000 or 2 6000s or Titans? I'm guessing the latter because of the additional CUDA. And, that begs the question of the 6000 vs. the Titans. Even though it lacks some of the functionality, wouldn't the Titans be just as good as 6000s for Daz? If nVlink gets make for iRay, I'm under the impression the Titans would be able to use it. How much would you really notice that 1or 2GB of space Windows is hogging?
I guess we should keep in mind DS./Iray still refuses to incorporate Image sequences on Iray Server for no conceivable reason. So maybe nVlink is like a ridiculous hope as well.
That 1 to 2 Gb could make a difference or not. To the rest, if you don't need or want the professional features then the Titan might be the better choice but if you're spending $5k+ on graphics cards I assume that you'll do more than Daz. Being able to use Photoshop and other applications in 10bit would be very important.
Are you saying the setting is unavailable or ineffective usinf Iray Server?
...but Octane4 does.
Fyi Iray in Daz Studio has had full in-software support for NVLink with capable cards for more than a year (since Volta support was officially added.) So yes - Titan RTX's can (and could since launch) take full advantage of NVLink for pooling memory during rendering.
Yes; theoretically yes - as of yet, no one (that I know of) with a Titan V has bothered making any sort of DS/Iray performance results public (it's primary userbase is academic researchers, so...)
Yes. There's a driver-level setting (only available for Quadro/Volta/Titan cards) called "TCC Mode" that puts the card into a state where it and other cards (also in TCC Mode) automatically pool their resources together and act like a single massive compute unit.
With RTCore support (still in development on Iray) almost certainly yes. Without those extra, dedicated cores the two Titan Vs should theoretically come out ahead (see my answer to question #1.)
Have you got evidence that it works because others who have multiple RTX cards say it doesn't.
Migenius is the only place that I have seen any form of Iray benchmark with a Titan V. The results are listed in megapaths per second. While people may take issue with this, all of the results of known cards pretty much fall in line with expectation. Which would indicate the Titan V is still as of this moment the fastest single GPU on the planet for Iray and is much faster than the 2080ti or Titan RTX. That may change when full RTX features get enabled, but not by much given just how much faster it is.
Beyond my issue with made up units that are not defined here is a basic problem with these benchmark results, The Quadro GV100 is effectively the same GPU as the Titan V, there are some minor differences that should favor the Quadro in rendering, yet according to these results its slower.
Then take a look at the Turing cards, the RTX Quadro 6000's results are barely higher than the 2080ti while the 6000 is the same chip as the RTX Titan and has more CUDA , higher base and boost and more memory bandwidth than the base 2080ti. A result that that close, likely inside the margin of error of the test assuming they have real methodology underlying the made up unit, makes no sense at all.
Yes - I am a Titan RTX owner and have successfully enabled/disabled the functionality in question.
None of the people mentioned own Titan RTX cards. They own Geforce RTX cards, which do not get TCC Mode access since it is currently treated by Nvidia as a premium professional/workstation feature only available to Titan/Tesla/Quadro cards.
But you haven't actually run a render bigger than 24Gb and gotten it to render on 2 cards with combined VRAM. Simply telling a driver to flip a switch is meaningless unless you verify it does what its supposed. There are any number of things in drivers that are supposed to fail silently and many others that fail silently when that isn't documented.
Nvidia Quadro RTX 8000 Turing Compared to Titan Xp/Titan V
TCC "Turing Compute Cluster" mode isn't that sort of a feature. It's an alternate function mode for the entire driver, one of the side features of which is that it reconfigures each capable graphics card for function in a peer-to-peer framework independent of the operating system, where resources such as VRAM are pooled across all available devices. Meaning that you don't need multiple cards of the same model to verify the functionality. As long as a single card works as expected in that mode, you know that multiple will work as expected as well (ie. with VRAM pooling) since even getting a single card to work requires full hardware/software compatibility.
Until its verified to work simply declaring that it does means nothing.
BTW, because you don't seem to know this, RTX Titans aren't pro cards and use the commercial driver. So enaling TCC, which is a Quadro driver thing, isn't possible on them at all. That you installed a Quadro driver on your PC doesn't make the RTX Titan magically use those features.
Although TCC mode seems to be a prerequisite for Vram sharing in most gpu's, it is not a guarantee that your renderer will recognize the configuration. As Nivida stated, it is an application specific feature. As of 2017, Iray has not supported NVlink, although there seems to be a product called iRay VCA which was a hardware / software combo using a version of Iray that did. It has been discontinued. Link: https://forum.nvidia-arc.com/showthread.php?15605-VCA-Discontinued-Iray-Next So you could successfully enable TCC mode in a Quadro card, but Iray would not have been able to recognize Vram pools. According to the link, a Nvidia representative said that they planned to enable NVlink support in iRay by the end of 2017. A lot has changed between Nvidia and iRay since then and there is no evidence that the support for NVlink has occurred. There is no mention of it in the news and even on Nvidia's own page promoting iRay, given the golden opportunity to promote this, they have failed to do so. https://www.nvidia.com/en-gb/design-visualization/iray/features/ Notice at the bottom of their feature page, they promote the ability to use multiple gpus but fail the mention the support for memory pooling. This leads me to believe that iRay does NOT support NVlink for memory pooling at present. The only renderers advertised to do so are V-Ray and a beta version of Octane.
GTX Titans have natively been able to support TCC mode since at least the Titan XP, possibly earlier. They didn't need to use Quadro drivers AFAIK. At any rate, TCC mode did not come into existence to enable NVLink. There is also another purpose. So the enabling of it does not confirm Vram pooling is enabled in a piece of software. That software must be configured to support it and as you mentioned, it should be verified with 2 cards and an NVlink bridge.
...which will be accessible to many of us here through the Octane Subscription track once the full release is available. V-Ray unfortunately does not have a plugin for Daz but does work with the Unreal Engine, though costs four times as much as Octane per month.
TCC functionality has been verified as working on the Titan RTX as expected for more than 5 months now (by myself as well as others.) It's just not a commonly discussed feature in the context of DS/Iray functionality because there are so few people in the community who have these particular cards (afaik just me and perhaps two others.)
Yeah, TCC functionality (via NVLink or any other interconnect for that matter) has to be explicitly supported by a particular application for it to work.
NVLink support for P2P functionality (including VRAM pooling) was added to all shipping versions of Iray last summer at the latest, and officially made its way into DS/Iray on July 16, 2018 with the adoption of Iray version 2017.1 beta, build 296300.616. As per the official Daz Studio changelog thread active at the time:
Iray 2017.1 beta, build 296300.616
Added and Changed Features
Prove it. That's all that's being asked for.
See this discussion and this exampe of the testing methodology from late February.