Two Cuda Cards For Iray
poorman_65
Posts: 140
I'm seeing a lot of discussion about using two cards, but they all seem to get into comparing a single Higher End card to two Lower End Cards. I know you need to install them without SLI and some of the other info on setting them up.
The one thing that I am not seeing, which is what I want to know is:
If I have a GEFORCE GTX 1070 Ti already, will adding a second TI 1700 roughly halve the Rendering time?
Post edited by poorman_65 on
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 TI 1700 doesn't appear to be an Nvidia product. AMD Ryzen cards do not support CUDA-based rendering.
As for "2 lesser cards equals 1 big card", to an extent that's true, until you get to the VRAM. There are no lesser cards with 12GB of VRAM, and two 6GB cards will not give you 12GB of VRAM. You'll get double the CUDA cores, but as soon as your scene exceeds 6GB, it's dumped to the CPU.
I actually specified that I wanted information on two Cuda cards and that I have no interest in comparing them to a single card.
My current card is a GEFORCE GTX 1070 Ti
https://www.nvidia.com/en-us/geforce/products/10series/geforce-gtx-1070-ti/
I am considering adding a second one.
Granted, but a TI1700 is not a GTX 1070TI; hence the comment by DrNew. Giving advice on Forums is fraught enough without presuming we know the card a poster is asking about. :)
Presuming your motherboard supports two cards, and your PSU has sufficient headroom, then there is no reason why you can't add a second card.
Just to emphasise the PSU comment; don't push your PSU to it's max, you will need some headroom.
Actually, you didn't, and you stated a totally incorrect Nvidia card ID. You merely mentioned that the discussions you've seen swing that way. The reason they swing that way is because of the common sense assumption that 2 6GB cards should equal 12GB, had Nvidia set them up properly. Unfortunately, since only end-users see the sense in that arrangement from a cost standpoint, the discussions usually begin with "can I get a Titan X Pascal's power for half the cost by going with a couple of 1060s?" and we go from there.
But to answer your clarified statement, yes, by adding a duplicate card your render times will increase, but not by 2. It's something between a +50% and +75% increase, if that. Scenes you currently render in 10 minutes will take between 6 and 8 minutes. Bigger cards have better throughput (base clocks, bus speeds, all that) but it all comes back to what your mobo and CPU can do with them. If your CPU can't push 40 PCI lanes, adding more cards merely means a higher light bill and a noisier rig when the cooling fans kick in.
That is what I wanted to know was how much adding a duplicate cuda card would improve the rendering time. I am doing a ton of animations and need to cut the time down. WIth what you are saying, I would probably be better off putting a GEFORCE GTX 1070 Ti in another PC that I have and just rendering seperate Animations or splitting an Animation between the two Systems. Does that sound like a better option than adding a second card to the current system?
Sorry about miss-stating the Video Card. That is why I clarified in my response.
FYI - I put pretty large PSUs in all of my systems when I built them so that shouldn't be an issue.
@DrNewcenstein ". . .But to answer your clarified statement, yes, by adding a duplicate card your render times will increase, but not by 2."
I know you know better. You might want to correct the mistatement.
For others not familar, render times will decrease, but not by a factor of two. Caution: folks estimating speed increases tend to assume both GPUs will be identical. This is not necessary and partially leads to some of the varying estimates of increased rendering speed (decreased rendering time).
Yes, effectively you will half your times. (Minus "loading" and "processing of non-cuda stuff", which is what accounts for the non-linear results.) However, your limits will still remain within the memory limits of the cards. (Less if you are on windows 10. Until they "fix" the memory issue that consumes 1-3 GB of VRAM.)
For Iray, if it is all Iray and all Cuda-based rendering, it will be exactly twice as fast, except for loading. (That loading "lag" is due to CPU threads, hard-drive speeds, memory speeds and CPU decoding ability.)
When/if you get into batch-rendering and/or animations, it will realy show more gains towards being 2x faster. Less when you are just spot-rendering individual scenes, "on demand", as you work.
JD isn't the only one pointing this out, but I'm going to wave the YMMV flag. I currently run a 1070ti and 980ti and don't see anything like that running Windows 10 Pro. My 1070 which runs an WUXGA and UHD monitor sits at about 700MB on start up (via GPU-Z) and my 980 which runs an HD monitor starts up at about 500MB. Interestingly, here at work the NVS 510 on my machine starts up at 1732MB*, running two HD monitors. There's also onboard Intel P530 running an HD TV using only 132 MB. ¯\_(ツ)_/¯
*using Windows Enterprise drivers for the card.
Start-up Daz3D and set the viewport to anything other than IRAY-Preview.
Open the LOG file. Should be in notepad. Will be full of log garbage.
Keep the log-file open, and shut-down Daz3D.
Erase the log-file contents and save it, so it is blank and does not have any "closing log info".
Start Daz3D again, and immediately open the LOG file.
This is what mine shows... (Down near the bottom.)
The thing you are looking for, is this line...
12 GiB total, 9.98798 GiB available
Windows "Sysinfo" reports the windows settings correctly... Shows that I have up to 34GB available, total. (Not including my system RAM, which is nearly that same value. With only ~1GB of VRAM actually being used, at the time Daz3D is running and the log is created. But IRay and Daz3D don't care what windows says is available. They both listen to what the GPU says is available, which is the total VRAM, minus what windows reserved as a "soft-maximum", which windows will treat as a hard-max for windows graphics, for itself.
Windows 10 uses something new, called "unified shared GPU memory". It allocates a percentage of VRAM from all installed cards, for "windows performance" (itself), since windows uses DX9/10/11/12 for the GUI. (They no longer use old GDI calls to draw everything in the OS.) Every screen, pop-up, overlay, icon, image-preview, font, etc... It "should" be using a soft-allocation, which auto-reduces to needs. Which it does, sort-of... For windows, it works, but not for actual software that directly uses hardware, bypassing windows. In these instances, it still sees the "reserved space" as a "hard-allocation", as if it is all "being used, at the moment", not as "potentially expandable to..." (Something about the new WDDM 2.0 drivers is the culprit.)
https://social.technet.microsoft.com/Forums/windows/en-US/15b9654e-5da7-45b7-93de-e8b63faef064/windows-10-does-not-let-cuda-applications-to-use-all-vram-on-especially-secondary-graphics-cards?forum=win10itprohardware
Similar posts are on almost every "rendering board", and have been for about 2 years.
It is not hard to hit the limits of your VRAM. Load three full models with iray materials, subdivisions turned-up to 4, and using 4096 as your image compression quality size. (Largest size allowed, before it forces compression to a lower quality image, for all your material textures.) All standing in some kind of decorated room, as most renderings would be setup. Now, render it as a 4K or 8K image, so it can be touched-up, without much quality loss, with an external editor. (Remember, 1080p is "TV standards", not PC standards. Though it is also a comon format with PC's, because of TV's and BLU-RAY and X-BOX and PlayStation.)
I know, that is hardly a "common use", but it is one that will eventually be used, and demand some crafty work-arounds to render. In a professional use, that is actually a mild setup. I dare not ask you to add dynamic-outfits or "where is my hair", as that would be slightly unfair, though, more realistic to a professional use. Honestly, professionals of that level, are not using Daz3D and IRay for production... Yet...