NON-SLI vs SLI problems/render time?
AnonTurtle
Posts: 31
I understand that people say not to render with SLI enabled.
However, I'm thinking about purchasing 2 Titan RTX to pool the VRAM together for 48GB of VRAM.
For this example lets consider 2 of the exact same GPU.
What would be the render time difference between SLI and NON-SLI?
What issues/problems would having SLI enabled incurr vs NON-SLI?
If you have any additional answers or information to give about this situation please feel free to tell me.
Thank you and have a great day!
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
From memory there are a few factors involved in how SLI affects rendering. It is slower, but how much slower probably varies significantly.
As for pooling the VRAM, that only works if you are using NVLink. SLI and NVLink are different. Not only that, even though it is supposedly possible with NVLink, it still wont work if the application (in this case iRay) does not support it. Which i believe at the present time, iRay does not.
Looking into this further now, it appears that full NVLink capability is not available in non Quadro cards. The RTX cards have the NVLink connector, but im reading that it is only there as a substitute to SLI and does not support the full NVLink capabilities. Sort of like NVLink 1.0 and NVlink 2.0.
RTX cards don't precisely do SLI. They connect by Nvlink. When used with non Quadro/Tesla cards the connection is labeled SLI but it is a new version.
Nvlink with Quadro/Tesla cards and RTX cards in enabled software can pool memory and door some other things. DS andiray is not enabled in this way.
Therefore 2 RTX Titans will not pool memory at this time. That may changein the future but Nvidia has never acknowledged that this is possible in the non pro grade cards so it seems unlikely that they will enable this for the consumer grade cards in iray.
So at this time 2 RTX Titans will function just like any other dual card setup. If the scene fits in boths cards VRAM both cards will render the scene together roughly doubling render speed. There is no need to connect the cards by Nvlink for this to work.
@kenshaw011267
I am calling it SLI as it is still being called SLI and I don't think anyone is calling it NVLink 2080 Ti or NV 2080 Ti or any other form.
Titan RTX is different from the 20 series RTX cards. It is already confirmed that it can pool VRAM. It has been confirmed by many users and Nvidia. This statement is from the official Titan RTX product page.
TITAN RTX NVLINK™ BRIDGE
Double the effective GPU memory capacity to 48 GB and scale performance with up to 100 GB/s in total bandwidth of data transfer utilizing the NVIDIA NVLink™ technology.
Without dwelving into this issue, can I please get an answer to the 2 questions above which aren't related to Pooling VRAM.
@joseft
While I don't quite understand the technology behind it, if it is anything like SLI performance in games where the game needs to be coded for SLI GPUs then a program may need to be coded to support VRAM pooling too. Do you have any articles for your statement above that I can look at?
Call it whatever you like, the fact remains that SLI and NVLink are two different things.
The Titan RTX being able to pool VRAM still does not mean it will work in iRay. They need to implement support for this before it will work, and as i mentioned above, i do not believe this has happened yet.
The NVLink replaced the SLI bridge. The 20 Series NVLink (not the Enterprise NVLink) is a higher bandwidth bridge than the SLI Bridge which allows it to do more things than the SLI bridge was capable of. If both the SLI Bridge and the NVLink were on the same 20 series card then yea they are different. However, since it replaced it it is relative. You understand that there is no SLI bridge on the 20 series cards thus you understand what I mean. There was no reason to make such a big deal about me saying SLI instead of saying Titan RTX with NVLink SLI, Titan RTX with NVLink, NVLink Titan RTX, NV Titan RTX, NVLink SLI Titan RTX. You have to understand there's no universal agreed on saying at this point. A Google search will show you many different sayings.
Additionally, I stated "However, I'm thinking about purchasing 2 Titan RTX to pool the VRAM together for 48GB of VRAM." Which does not state SLI or NVLink, I simply said I wanted to pool it. Finally, when you use 2x 2080 Ti with an NVLink the Nvidia Control Panel still states SLI not some sort of NVLink (Google up NVLink Control Panel). Therefore, everything is relative and you shouldn't be making this a big deal considering that you FULLY understand what I am trying to say.
I updated my statement before you posted asking where you got that information from. If you had read it somewhere that you could link me? Thank you.
https://www.pugetsystems.com/labs/articles/NVLink-on-NVIDIA-GeForce-RTX-2080-2080-Ti-in-Windows-10-1253/ has something on the application needing to support memory pooling. There may be other sites with other information as well.
In their own documentation on Iray, Nvidia have stated that you do NOT enable SLI when rendering with Iray as it can damage the GPUs. Since the NVLink is simular tech, but its their tech, I would double check with them first
I do not think i was making a big deal about any of it to be honest. Your original post gave the impression that you thought SLI was what enabled memory pooling, so myself and Kenshaw were just trying to point out that is not strictly correct. When i responded the second time it was before you edited it, and in any case, in that post you implied that it everyone else was still calling it SLI, and therefore thats what it was. My response to that may have been blunt, but i was not offended/angry or trying to offend you or anyone else. I just believe that lumping different things under the same name can cause confusion and should be corrected, especially when people might be making costly decisions based on that information.
And yes, i do admit that even the information from Nvidia can mislead people on the difference between the two. Regarding them still being listed as SLI in the nvidia control panel, that is probably because at present the vast majority of PCs still use older generation cards where NVLink is not appliccable. That, or Nvidia have just been lazy in updating it so it reads NVLink instead of SLI where appliccable. Im sure eventually that will change
The article posted above by Daywalker Designs is very good, explains just about everything one would want to know about it
Titan RTX owner speaking.
SLI is Nvidia's solution for multi-GPU load-sharing on its consumer level graphics hardware. It works at a computer's CPU level by first taking a graphics workload in its totally pre-calculated state, splitting it into separate equal sized chunks, sending each of those chunks (via PCI-E lanes) to each SLI-enabled GPU in the system for processing, and then sending the results of those separate processing operations (either also via PCI-E lanes or a separate dedicated data line such as an "SLI" or NVLink connector) to a pre-designated "master" GPU where the results are combined in hardware for final output/use by the system.
The reason why SLI and Iray don't mix well is because Iray already has a self-contained multi-GPU load sharing mechanism built into it, which works by first taking a graphics workload in its totally pre-calculated state, splitting it into separate equal sized chunks, sending each of those chunks (via PCI-E lanes) to each GPU and/or CPU in the system enabled for processing, and then sending the results of those separate processing operations (also via PCI-E lanes) back to the CPU where the results are combined in software by Iray itself for final output/use by the system. Hence why the performance impact of having SLI enabled for Iray rendering is negligible at best and a hindrence at worst. SLI opens up a specialized pathway for graphics processing which goes against Iray's own internal graphics processing pathway.
Meanwhile, GPUDirect is Nvidia's solution for multi-GPU load-sharing on its semi-professional (eg. Titan) and professional graphics hardware. It works at a computer's GPU level by establishing a high speed peer-to-peer resource sharing relationship between all GPUDirect enabled devices in the system (either via PCI-E lanes or NVLink connectors) and then presenting a single logical device (eg. a single GPU made up of multiple Titan/Quadro GPUs) for specifically compatible software to make use of.
GPUDirect P2P via NVLink has been specifically supported by Iray since version 2017.1 beta, build 296300.616. I can't verify that it actually works as expected with the Titan RTX on the current Daz Studio beta (since I only have the one card to work with) but I can tell you that all the signs indicate it does (eg. "TCC" driver mode - what makes GPUDirect P2P possible on windows systems - is enable-able on a Titan RTX.)
Doesn't mean it's the same thing, and in 3D graphics rendering, it is important to differentiate. SLI is bad in 3D rendering, whereas NV-Link might prove to be a requirement (depending if shared RAM - or something else) needs it.