Going from 8GB to 11GB of VRAM - how much of an actual difference does it make?
dennis.d.t@outlook.com
Posts: 81
in The Commons
Hi All,
How much of an actual difference would going from 8 GB to 11 GB of VRAM make? I currently have 8 GB and DAZ switches over to CPU rendering when memory load is about 7 GB (I understand that Window 10 uses up that 1 GB of VRAM). Can I expect DAZ to GPU-render up to 10 GB or so if I switch to an 11 GB video card?
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
About that. What's more is that cards with more memory of the same generation also have more CUDA cores and are faster at rendering and simulations, as well.
Usually, the cards with 11GB of VRAM are a bit faster than their 8 GB counterparts (say 2080 to 2080 Ti or 1080 to 1080 Ti). If you have a 1070/2070, it'll be a bit more of a jump.
Here's a few rendering benchmarks:
https://www.pugetsystems.com/labs/articles/V-Ray-NVIDIA-GeForce-RTX-2070-2080-2080-Ti-GPU-Rendering-Performance-1242/
So at least a 30% boost in rendering speeds/reduction in render times. Two 8GB cards may still be faster though, vs a single 11GB Ti card.
Also, depending on how involved your scenes are, as you've noted, you'll hit the 'CPU only' barrier less often. I find it significantly easier to work within the 8.9GB or so limitation (after Windows VRAM tax). NOTE that supposedly with the latest Windows and Nvidia drivers which have a 'hardware GPU scheduler', you should have a bit more than that amount available for rendering. My rendering system has been deliberately separated from the internet for a long while now, so I have no easy way to download the new Windows drivers to verify that... others can comment based on their recent experience with the latest drivers.
It sounds like you just need to optimize your scene, getting rid of any 8K textures, and using 4K only for things close to the camera. A single 8K texture can easily use up a gig, and a 4K can use up a quarter of that. Unless it's a closeup, you probably will not be able to distinguish it from a 2K texture at all, certainly not if you're animating and use motion blur.
I think thats a problem for me too^^. I have a RTX2070 Super, with 8GB Vram and it get so quick overloaded... I assume the textures are the problem, does HDRI also write down to the Graphiccard, like when I use a 8-16K HDRI?
I also think about how much better would the system run with the RTX2080.
But there is also this baby:
https://www.nvidia.com/de-de/deep-learning-ai/products/titan-rtx/ (I just have the german url in my cache, sry..)
With 24GB Ram... AND there are rumors that the rtx 3070/3080 Series are also coming early 2021.
So Maybe the upgrade to another card is more worth if the next Series get releaased :/ And maybe "we" have to manage better textures :D
I tested one time a texture resizer script, but the problem was, that it saved all textures in a TEMP Folder and when you reload the scene, all is gone.. and you have to find ALL textures for ALL items again -.-....
And like me, I love to make comics, so I use a Scene 10-30x to create storys, with little changes from Frame to Frame.
Well, one is 8 and the other is 11. Sure Windows takes a certain amount, but you are looking at around 3GB more. I don't know what else to say. VRAM is VRAM, you either have enough or you don't. How much you need is specific to your own creations, and nobody else can really help with that because everybody uses Daz and Iray differently.
I would certainly recommend trying to optimize your scenes before spending money. There are also products in the store to help with that.
What is your GPU? The 2080, 2080 Super and 2070 Super all support Nvlink, and the recent version of Iray add support for it. Nvlink allows them to pool their VRAM together. If you own one of these cards, then you have the option of buying a second one along with Nvlink, assuming your PC can support this set up. For Iray, using Nvlink allows them to share texture data, but not geometry, at least currently. I doubt that will change, but it is possible. Still, texture data is one of the biggest culprits for eating up VRAM, so just being able to share texture data is a huge win. We have had someone in the forums successfully Nvlink two 2080tis, proving this is possible.
Note, the original 2070 does not support Nlink, only the Super version does.
So you have an option. Ampere will be releasing soon, but we have no specifics, just rumors. The 3080 might have 10GB, their may be a 3080ti with 12GB, and a Titan 3000 or 3090 that has 24GB like the current Titan RTX does. But these are all rumors. Odds are these will have their own Nvlinks, too, but again, that's speculation. I think Ampere will be great, but regardless of its performance, you can bet that the 2000 series will drop in price after Ampere launches.
I have a 1070 and I have noticed recently, while running GPU-Z while loading and rendering, that IRay seems to take as much VRAM as it can up to the 8GB limit. At least that's what I see in GPU-Z. For example, having a scene with 3 characters and then removing one made very little difference to the total VRAM being used. IRay seems to use whatever is available. Sometimes I get a black render though which means I have exceeded the 8GB (I am not able to get 4 characters in a scene - even after running Scene Optimizer - which I always use for all scenes). It renders black when you disable CPU fall-back.
i would assume from this study tha tthe 2070 Super would be even faster?
Did you close DS and re-open the scene? If not the textures of the removed character will still be in memory. Just sayng.
Yes, I tried both ways - same scene session and also close/open but the VRAM stays high. Maybe something to do with texture compression - maybe it needs to compress less when the demands are smaller so it fills up the available VRAM with textures which are uncompressed (or less compressed)?
My bet is if you have a 100Mb HDRI and lots of other hi-rez textures in the scene then removnig one character doesn't make much of a difference percentage-wise. If the overall scene is heavy already with textures and pushing your GPU limit, then removing a small part of it won't make much difference to the total load.
I've had only a few scenes that have dropped to CPU with 8Gb of GPU RAM. In that case I either reduce the texture sizes with the script available, or render it in interactive Iray mode which seems happy to render without with the GPU dropping into a hole. The resultant image is just about the same as a "full" render anyway. Good enough for me anyway.
2070 Super has 2560 CUDA, 2080 has 2944, 2070 has 2304. So a 2070 Super is between the 2070 and 2080 in rendering speed.
There are scenes e.g. collective 3d neighborhood etc. which require more than 32GB system RAM. Upgrading VRAM is similar to rat race, it is endless and after huge expenditure finally worthless because still someone will suffer insufficient VRAM. I realized it and learned to stay happy with my current VRAM.
Well however I will upgrade only if 32GB VRAM available within my budget otherwise in future I will look for an ordinary RTX with minimum 6 GB VRAM.
I'm going to come at this from another angle and I hope it helps. There is rumor that DAZ will be able to take advantage of 2 GPUs in future releases, meaning if you have 2 RTX 2070 supers, you would effectively have 16 gigs of vram. While this is not a feature in DAZ yet, it's worth waiting for. Much cheaper I think to simply buy a second RTX 2070 and double your vram than shell out for a new GPU and only get a boost of 3GB.
That's already implemented in the version of Iray that Daz is using.
NVLink is already enabled.
The only cards it should work for, the cards need to be a matched pair, are the 2070 Super, 2080, 2080 Super, 2080 ti, RTX Titan, Quadro GV100, Titan V CEO edition and the RTX Quadro line.
While 2 cards might double the vram, does it also halve the render times?
No, it's more like 80% faster going from 1 to 2, and the same from 2 to 4. But if you run an instance of Blender on each GPU, you can get a true 100% speedup in Cycles.
Still, it is worth considering 2 cheaper cards rather than one expensive. I'm not sure my desktop will fit 2x2070 (or, more likely 3070). I have 2 PCIE slots but these are fat cards and the need air.
Do you need to use NVLink to get the benefits of two cards in Iray, or is that only necessary to combine the VRAM? Last time I checked, you could tell Studio to take advantage of multiple video cards regardless of type or whether they are linked.
Only if you want to combine the VRAM.
I'm running a 1080ti and a 2080ti (so obviously no link) and Daz uses both of those when rendering.
double post
Why spend cash when you could try something else first?
Try Octane for Studio. It does out of core rendering and is supposedly very fast; there is a free version available.
I use Blender and Cycles, which also has out of core rendering.
Other options will require learning a new process (or processes), but you might find one you prefer.
You've identified the problem as being a lack of VRAM; there is more than one solution. Nvidia, of course, prefer you to throw cash at the problem - cash is often not the best option when taken in haste.
On a mATX motherboard, the ones that have only 2 PCIE slots, you can only fit 1 card without using riser cables.
Per the results in this thread, it's slightly over 1.9x faster with 2 cards, or (looking at the quad 2080 Ti vs single 2080 Ti result) 3.68x faster with four cards.
https://www.daz3d.com/forums/discussion/341041/daz-studio-iray-rendering-hardware-benchmarking/p1
So not quite 2x, but close...
My post above this already addressed this?
And the benchmark thread is in my sig.