Dual Graphics Card Order Question
duckbomb
Posts: 585
Hello! I have another question I'd like to see if anybody has an opinion on... I have upgraded from my Razer Blade/Core setup to an actual workstation, and I have 2 video cards to put in it. I know you might laugh at me, but one is a GTX GeForce 1660 with 6GB of VRAM and the other is an NVidia Quadro RTX 8000 with 48 GB of VRAM. I know the two are very different cards, and so my question is this:
Should I put the 1660 in PCIE slot 1 and just keep the Quadro for churn and burn when it comes to rendering in iRay, or would you put the Quadro in slot 1 and just ignore the 1660? I ask because when I started using the 1660 I noticed that the viewport and everything else is sluggish where it wasn't before (in my core I had a GTX Titan XP). I assume that this is because the 1660 is slower than the Titan, but I don't know for sure. I doubt I'll ever max out the VRAM on the Quadro, but I also don't want to place unnessary load on the big card while it just churns out renders if I can keep all of my other tasks on the 1660.
Does this make sense? What do you think?
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 Quadro 8000 uses roughly the same GPU as the 2080ti. It is going to do tasks much faster than the 1660. so if viewport responsiveness matters leave the Quadro as your video out card and just leave the 1660 as extra render power, or alternatively return or sell it as it is pretty superfluous.
I guess I should have phrased my main question better. I suppose I'm just curous to know if there's any performance hit I should be concerned about switching down to one card to do everything (run OS, in particular, as I suppose I don't do much else during render time).
Maybe I should just try switching it around instead of asking opinions...
And perhaps it is superfluous, but who cares? So is my neighbor's Tesla haha.
If you want the absolute best rendering performance you can get from the RTX 8000, keep both of them and use the 1660 to drive your displays/do whatever else you want besides rendering. This will allow you to switch over your RTX 8000 into a Quadro/Tesla/Titan GPU exclusive driver mode called TCC ("Turing Compute Cluster") which will net you slightly better rendering performance as well as more usable GPU memory headroom for rendering (not that you'll need it for the foreseeable future, but still.) The major drawback to TCC mode on supported cards is that it shuts down the display output capabilities of the card. Having a second much lower power GPU like you do for driving displays/GUIs is actually the manufacturerer recommended scenario for getting the most out of a professional card like the RTX 8000.
As to which card to put in which slot. The only situation in which it matters is when using a single GPU on a consumer-grade (non-HEDT, non-Server grade) motherboard since the only slot capable of full bandwidth communication is normally the primary (closest to CPU) one. In which case you would want your single GPU in that slot. In situations with two GPUs mounted to a consumer board's primary and secondary PCIE slots, PCIE lanes are split evenly between the two - making it irrelevant which one goes where. And on higher end HEDT/server boards, both the primary and secondary PCIE slots (at minimum) get full speed PCIE connectivity regardless, so again it doesn't matter.
PS: Once you get your system configured to your liking, mind contributing some benchmark data for your RTX 8000 to the collection? (instructions here.) Am very curious to see how it measures up to my Titan RTX. In theory the Titan RTX, RTX 6000 and RTX 8000 are all exactly the same GPU - just with different memory chip configurations (2 or 4GB per die ECC GDDR6 RAM chips versus 2GB non-ECC GDDR6 RAM chips.) But it's always fun to have some actual numbers to back things up.
AH, ok, cool, that makes sense. I think that with my motherboard, then, I can probably just keep the 1660 in slot 1 and put the Quadro in slot 2. So, even though the Quadro won't have any display connected to it, render performance will still be good, and even though the viewport is clunky this isn't representative of render performance? This makes sense, and it's what I thought, but it's good to have that confirmation.
Sure, I'd be happy to put up some benchmarks when everything is up and running correctly. Currently, I'm dealing with some driver issues I need to iron out, and then I need to re-download my library, but once all of that is done I'll put it up. I almost went with a Titan RTX myself but (you may remember from an earlier thread where I asked if this was too much overkill or not) I ended up picking up this Quadro used when a department had to sell their assets at work. If it hadn't been for that, I'd have gone with the Titan (and almost still did, it's just so damn pretty).
Yeah, although ironically the single biggest design flaw to the Titan RTX is its (admittedly pretty) cooler. Design and performance wise it is identical to the 2080Ti Founders Edition one. Which, while not a bad design per se, is already pretty much stretched to its limits with the 2080Ti. And since the Titan RTX has a greater thermal output density than the 2080Ti (more active SMs/Cuda cores on the same size die) that limit is stretched just a little bit more.
In stock configuration on an open test bench in a comfortably cool room, my Titan RTX was peaking at around 82c under full load. While 82c is perfectly within the GPU's spec, there was some core speed downclocking going on to maintain that temp. Plus, the dual axial fan design where heated air is exhausted within the computer case itself (rather than out through the PCIE slot bay like with the blower design on your RTX 8000) means that having one or more GPUs in addition (a common scenario in professional computing situations where the expense of something like a Titan RTX is actually warranted) is going to compound the heat situation further.
The only situation in which the Titan RTX's cooler is (close to) satisfactory in perofrmance is where it is the only GPU installed in a well ventilated system. At the same time, the only way to take full advantage of the Titan RTX's raw computing capabilities (the only task its 24GB of VRAM is good for for the foreseeable future) is to use it along with at least one other GPU. It's an inherently flawed product. Although luckily it's a flaw easily remedied with some aftermarket cooling (in my case, with some custom watercooling.) And even with that added expense, the Titan RTX is still a steal when compared to its Quadro RTX brethren when you're talknig about paying full price.
So long story short - am very happy with the Titan RTX itself. It's cooler however has turned out to be little more than a pretty paperweight.
You know you could just try it both ways and see. I'd be curious if there is any noticeable difference. I wouldn't worry about slot 1 or 2. I've actually got my display driven by slot #2 and it hasn't impacted my performance, certainly not in Iray, and my benchmarks prove that. GPU rendering Iray is not effected by x8 or x16.
I have two 1080tis, and with every render, they generally run almost the exact same number of frames. One scores just a few more because that GPU is clocked higher, so its bound to get a few more iterations over a long render. And my testing (though I did not write it all down) showed that it didn't matter which GPU the monitor was plugged into, the numbers were basically the same. But 1080tis cannot do TCC, so for you that could make a slight difference.
Also, if you are getting stuttering and stuff just using Daz Studio, maybe swapping to the 8000 as the driver would be better if it helps that. After all, rendering is only one part of the equation. If your experience in Daz Studio itself is poor, that probably hurts your productivity just as much as slower rendering (and that is assuming that renders would actually be slower using the 8000 as the driver). However, a 1660 is not a terrible card, it should have no issue driving Daz. It is also possible that your stutter may be CPU related. Daz Studio is only single threaded from what I recall, so having lots of cores and threads on a CPU does it little good for Daz performance. Daz runs better with faster single cores, and a workstation CPU can have a much slower single core speed. If we are talking Xeons, they could be running at like 2.5 Ghz or something, or even less, which is glacially slow by desktop standards. A Windows update could also be effecting you, I believe some people had stuttering after a Windows update because it kept typing up the CPU for some reason. So your issue may not be hardware at all.
Oh that's another good point! My single thread performance isn't fantastic... I'm running an Intel i9 10980xe at 3.0 GHZ. I know the Ryzen processors are faster per-core, but my buying situation favored this specific build. I honestly hadn't considered that it could be the CPU, but it makes sense.
It's reassuring to know the slot doesn't make a difference. I was honestly unsure, so this is really good to know!
I think I'll try both configs and see what works best for me, but I think first off I'll try using the 1660 to drive the display and see what I can do to speed up in-app performance.
Thanks for the helpful response!!
I dont believe anyone has mentioned drivers yet. Its something you have to think about. Geforce and Quadro run different drivers.
I looked into this a few years back when i was considering doing it, and there is a lot of wildly conflicting information regarding the viability of running both Geforce and Quadro in the same machine. As above, the main issue is the different drivers. From what i could gather, you can run them both in the same machine, however you cannot install the drivers for both in a normal scenario without horrible issues. The Quadro will run on the Geforce drivers, but the Geforce will not run on the Quadro drivers.
What this means is that while you can run both of them on the Geforce drivers, you lose all the benefits that Quadro cards have (that are enabled by their drivers) like 10bit color and TCC. For some that may not be a problem, but it does mean you are paying a monstrous price and the only real benefit you are getting for your money over a high end Geforce card is the added VRAM. In terms of render speed, on all the benchmarks i have seen across multiple GPU render engines, the RTX 8000 is only barely faster than a 2080ti, which is probably down to the drivers rather than the hardware.
I had thought of this, and I came to the same basic conclusion as you have... both cards can run at the same time, but that I might be giving up some Quadro benefits. Honestly, the only reason I went this direction was because I got a pretty good deal, and it came down to VRAM. I'm not concerned with speed, in fact I think the only thing that would be substancially faster than a 2080 would be two 2080's, but it's a "luxury" to really not have to worry about memory managment as much as I have had to in times past. I know that sounds sloppy, or lazy, but the truth is that if I'm rendering out a bunch of shots for a comic page or something it's easier for me to just hide giant parts of the set than to delete to save memory only to have to load up an old file to set up a new shot. This way, I can just keep cranking forward and "saving as" without really worrying about the size of the scene. No matter how you slice it, rendering on this one card will be faster than dropping to CPU, and that's really my only goal here.
It's a good point, though, about the drivers. I'm up and running now, and so far I don't see any conflicts, but I don't know how to tell which drivers are installed and working. In device manager, it says that both are in there and running on certified drivers, and in the NVidia Control panel it sees both, so mabye I got lucky and this issue is now a non-issue... I'm gonna try to fill up the VRAM, though, just to see how much it will take, and then see what happens haha.
I don't know how it used to be, but presently all Nvidia GPU drivers are based on exactly the same code base (indicated by the XXX.XX part of the version number) regardless of the target model line (Tesla, Quadro, TItan or GeForce) or release channel (Quadro, Game Ready, or Studio.) And rather than removing driver features for more consumer-oriented cards, Nvidia goes about the opposite tactic - omitting game-specific optimizations and updates for non-GeForce driver releases to aid in productivity use of its products.
For example, TCC mode functionality in Geforce release drivers is identical to what's found in Studio or Quadro release drivers. Nvidia enforces per-card TCC compatibility via reading of hardware model IDs - not by supplying different driver files (if you check, the related binary files included with the driver installations are identical from version number to version number regardless of whether your driver download is listed as Quadro, Studio, or Game Ready.) The only thing you stand to lose using GeForce drivers on a current generation Quadro card in non-gaming applications is some performance due to having more going on at the driver level than what's strictly necessary.
I've found that you want the video card driving your monitor in slot 1 as most computers will only use that slot before Windows is up.
Hmm....very interested in this topic as I just managed to pick up a used Quadro M6000 (Maxwell) 24 GB card for $806 (another server/workstation upgrade cycle seems to be going on). I decided to do this rather than trying to save up $2500 for a Titan RTX. I would like to use it in TCC mode to make the most memory and features available (particularly if I end up running Windows 10 with its parasitic Vram reservation). My goal is the same as Duckbomb's: not having to worry about dropping to CPU rendering. So, the question is what card can I use to drive the display without driver issues?
I have available the used Titan X (Maxwell) 12GB that I originally got for that machine, but I am concerned about divers. I also have a couple of AMD Radeon cards which I could use as a display driver. GIven what I just read above about the differences between Quadro and Studio drivers (Thanks folks above! Very helpul info!) maybe my best approach would be to put the Titan X in slot one to drive the display and also render what scenes it can (likely anything I would do right now) and put the Quadro in slot 2. Run both in non-tcc mode with perhaps a small loss in rendering performance for the M6000, which would be more than made up for most of the time by the Titan.
I will probably run some experiments comparing different configurations. Maybe Duckbomb's configuration with both types of driver will work with Maxwell GPUs also.
Duckbomb: Congratulations on getting a RTX 8000! Did not even know that they made a 48 GB card yet!
Thoughts, experiences, and suggestions are welcome!
If I were you I'd go with the Titan X (left in WDDM mode - the alternative to TCC mode on Quadro/TItan cards) in slot one as your display driver/secondary rendering GPU and the M6000 (set to TCC mode - TCC/WDDM mode switching works on a per-GPU basis) in slot two as your dedicated rendering GPU. I'd suggest leaving the Radeon cards out of things entirely. Haing both AMD and Nvidia GPUs active in a single system means having both AMD and Nvidia drivers installed/running at the same time. Which can lead to some very difficult to troubleshoot buggy behavior. This way you get the most possible out of all the hardware you hav available while still maintianing full system usability.
PS: The secret to managing TCC/WDDM mode switching and other advanced Tesla/Quadro/Titan GPU functionality is the commandline only Nvidia System Management Interface tool (Nvidia-SMI) found in the following location after the installation of ANY recent Nvidia driver package:
C:\Program Files\NVIDIA Corporation\NVSMI\The manual for it is also located in that same directory:
C:\Program Files\NVIDIA Corporation\NVSMI\nvidia-smi.1.pdfBtw this is also the tool used on the enterprise level to control GPU clock speeds, power draw limits, and the like. I highly suggest throroughly going through its functionality if you want to get the most out of Quadro/Titan hardware.
The information i provided above was outdated i think, judging by what RayDant advised. I was only going off what i found a number of years ago. If RayDant is correct, then by the sounds of it you can use both and you will get the benefits of the Quadro.
Maybe a bit more research is required though, because i do remember reading that Nvidia were limiting it in that way on purpose, to discourage major studios from buying just the one primary quadro for their machines to get the benefits of Quadro (like 10bit color) and then a number of cheaper Gforce cards for the added power.
Maybe duckbomb can run some tests for us. Should be easy enough to confirm via attempting to enable TCC on the RTX 8000 (assuming it is being run headless). Or by enabling 10bit color if it is being used to run displays. If those settings cannot be enabled, then i guess that would mean the Quadro is being limited by the Geforce also being in the machine.
edit - and RayDant beat me to it :)
Hi all! Sorry, long day at work... sure, I can run some tests. It is running headless, so I can try enabling TCC. I'll have to do it tomorrow, as it is late here, but I'll do it and post back. So far, from what I can tell, there don't seem to be driver issues, although I understand the limitations might not be immediately apparent.
I don't exactly know how to enable TCC, but I'll YouTube it. I'll also post benchmarks in that thread for ya'll.
I'll get back to ya tomorrow!
I use a gtx card for rendering and a vega for the viewport and no issues here.
My experence with old generation card mix (Quadro K6000 and GTX 1700) is:
- installed the Quadro Driver following the standard procedure of the installer -> got all Quadro features (mainly 10bit output and optimized Open GL driver)
- downloaded the GTX driver package
- used 7zip (free zip / unzip software) to extract the GTX Driver software exe to a temp directory
- went to Windows Device Manager -> it indicated that the GTX is not working (which was expected as the Quadro Driver installer does not take care of the GTX)
- opened context menu (right click) on the GTX entry in the Device Manager and used the "update driver" entry
- selected "Browse my computer for driver software"
- browsed to the temp dir to where i had extracted the GTX driver files (I think it had a subfolder called driver)
- let Windows find the driver files and install them
Result: The quadro driver plus Nvidia Settings are still offering all Quadro features and the GTX is also running well as its own driver was installed by Windows
I don't know if this is working for the new generation cards as well.
-------------
Regarding your slots.
As you have an i9-10980XE your mainboard likely has min. three PCIe slots and likely two of them are full x16 bandwidth while one is x8. To my experience it does not matter in which slot you put which card as the bios finds out on which card a display is connected (but this might be different in your system). You might want to take care that the Quadro is in a full x16 slot (likely slot 1 and 3) because the "big" RTX cards are the first ones ever which are slightly limited by the x8 bandwidth (but the effect is very minimal).
Regarding the single core performance you mentioned. The i9-10980XE has a base clock of 3.0 GHz BUT an all core turbo of 3.8 GHz. That means as long as you cooling and power delivery is sufficient it can run at 3.8 GHz on all cores. If only one core is under heavy load (I think there are a lot of functions in Studio that are single threaded) the max. clock is 4.8 GHz for this one core.
By the way, nice Quadro you have there :)
(Correct me if I'm wrong) your setup is external Nvidia GPU + INTEGRATED AMD graphics though. Based on what I've seen, external Nvidia + external AMD is where the trouble tends to manifest (which was one of the scenarios @duckbomb was contemplating.)
Thanks for all the excellent info and suggestions. I will do some experiments as time permits. First with Win 7 (current installation) then Win 10 on a separate drive (I run both on some machines, and just switch drive cables to select).
My motherboard in this case is a Supermicro X9DR3-F server board with two E5-2680V2 CPUs (10-core) and 64 GB 4-channel DDR3 RDIMM. The motherboard was puchased new, the CPUs, RAM, and GPUs are used/surplus. The choice of hardware was due to my goal of running VUE Infinite 2016 renders on several dual-Cpu machines (this version was the last outright purchase version, and relies on network rendering for decent throughput, with GPUs only used for anti-aliasing and preview). This also gives me a good number of PCIE lanes. I also need to run Daz Studio and Carrara, and eventually Blender. It has taken me several years to buy the parts and put them together, so I have been kind of upgrading on the fly. I never thought I would be able to get an M6000.
I don't have an AMD card, my question was related to two NVidia ones, but this is a very interesting topic. I'm surprised there isnt' more "hard facts" out there from either NVidia or AMD about running either multiple classes of cards simultaneously or cards from different vendors. It's an expensive proposition to guess and check.
I'm gonna try turning on the Quadro functions later today, and I'll report back with the results.
Nvidia has actually been on a bit of a roll lately with un-arbitrarily disabling "pro" features on less "pro" hardware. Eg. as of last July, 10-bit (30bpp) color functionality is now supported across all GPU product lines (look under NVIDIA Control Panel > Display > Change resolution > "3. Apply the following settings" > Output color depth.)
As of last April or so, liveview streaming (the one feature Iray Server held back for Quadro users only) is also supported across all GPU product lines. And TCC mode (with a compatible card) works regardless of driver release channel (I know this one for a fact because I routinely use my Titan RTX in TCC mode with whatever the latest Game Ready GeForce driver release is, since that is the release channel that gets most frequently updated.)
At this point, pretty much the only thing Quadro's have to offer that Titans/even GeForce cards don't is ECC memory, quaranteed multi-GPU friendly blower style coolers, stability-minded boost clocks/power limits, and overall higher VRAM capacities. None of which are critical features unless you operate a data center or work for NASA (where ECC memory is actually useful.)
Obviously if you can get a good deal on a Quadro, then it's a steal. But otherwise for the vast majority of users/uses Titan/GeForce really is all you need.
The commands for enabling TCC mode on a GPU is:
or
C:\Program Files\NVIDIA Corporation\NVSMI\nvidia-smi -i 0 -dm 1 (for already headless cards only)
And for disabling:
or
C:\Program Files\NVIDIA Corporation\NVSMI\nvidia-smi -i 0 -dm 0 (for already headless cards only)
Where the number in red is the ID number of the GPU you wish to target. Keep in mind that you will need to execute the above commands (either via DOS window or batch file) with admininstrator permissions. And a total system reboot is required before the changes take effect. And also don't forget to not run your dispaly(s) off of the TCC enabled card once you restart the system (if you do, funny things will happen like getting stuck with 640x480 screen resolutions.)
To determine what the ID numbers are for the GPUs in a multi card system do:
in a DOS window and you should see a list.
Thanks, RayDAnt! That is really helpful!
Thanks @RayDAnt !
I followed your instructions, and this was my CMD output:
C:\Program Files\NVIDIA Corporation\NVSMI>nvidia-smi -g 0 -dm 1
Set driver model to TCC for GPU 00000000:17:00.0.
All done.
Reboot required.
Once I rebooted, everything runs exactly the same as it had before, but I assume the change was made. I tested in DAZ, and seems all good there, and I still have the GTX 1660 drivers installed, as well.
I guess you can run both at once now... sweet!
Thanks all!
And even though the 1660 isn't the most fantastic GPU out there, I would bet that using both GPUs together would render surprisingly faster render times. That is assuming the scene fits the 6GB VRAM while it drives the monitor. That's 1408 Turing CUDA cores to add in on top of what the 4608 the Quadro 8000 offers. That is 6018 CUDA cores of power between them, and these are current generation CUDA. So while the 1660 may not be the fastest, it will do its part in contributing to a render, again if the render fits its 6GB.
If you open Task Manager you should see your Titan card listed on the left but NOT your Quadro. That's the quickest way to verify your card is actually in TCC mode (Windows has no way of knowing what it's doing once it's taken out of WDDM mode. Meaning that to track your Quadro's activity you'll need a third party app like GPU-Z or the Nvidia SMI tool on the command line. On the upside there is no Windows related penalty on memory usage on the Quadro card.)
I did this, and that's exactly what I see. It's not in Task Manager at all. However, in my Device Manager, the Quadro 8000 is now in there as an "NVidia Quadro RTX 6000". So... there's still obviously some driver issue. It seems to be running just fine, and according to DAZ it sees all 48 GB of RAM, so maybe I'll try a driver update and see if I can get that to change. Otherwise, I'll probably just ignore it lol.