Did a promt pop-up for enabling SLI ?
Nope. A Titan Z is 2 GPUs on one frame, so the SLI connection is some kind of internal thing, not the external cable as you'd use with 2 physically separate GPUs in 2 separate slots. I got a "this system is multi-GPU capable" message, and NVCP lets me choose "Maximize 3D" where all GPUs work together for rendering, or Disable Multi-GPU for running multiple monitors and each GPU runs independently.
This is a known bug with SLI configurations, as long as both GPUs are selectable as CUDA devices in the Nvidia Control Panel it's fine.
Yes, I can select All, the K4000, or either half of the Titan Z. I usually leave the K4000 out of it.
Sounds like it could be a driver bug similar to the one reported here?
https://www.techpowerup.com/forums/...ond-card-missing-cuda-tooltip-in-gpu-z.244757
I would suggest you contact NVIDIA directly regarding this issue
and please share their response here so that others with the same issue can find the answer.
That thread reports the issue was addressed with the 4xx drivers. However, no one in that thread explicitly mentions a Titan Z, so I don't know how different the SLI operation is between that and physically separate GPUs.
However, I'm now running into a more pressing matter: for the longest time, I ran the 387.92 drivers because they were stable. The Titan Z had issues with 389, IIRC, so I went back and stayed there until I got a 1080ti (about a year after getting a Titan X (Pascal), which I also ran on 387.92 since all I played was Skyrim and Daz Studio Iray worked just fine).
With all the reconfiguring I've been doing, as noted earlier, when I put the Titan Z and K4000 in the same machine, then did a DDU-cleaned install of 387.92, I got nothing from the Titan Z, even after several reboots and checking the simple stuff. No Device Manager appearance, not even "Other Devices" with the generic icon and triangle.
DDU'd again and put in 392-something and got the same thing.
DDU'd again and put in 411 and it worked, but came right back to the present issue where Side 2 doesn't contribute, and THEN started getting "This device cannot be ejected", as if it were a USB device, followed by "Display Driver stopped working and recovered", along with "Daz Studio encountered and needs" errors. I use Daz Studio on this machine to the almost exclusion of all else, so for this to happen is a major issue.
Reboots usually hang after these faults, forcing a hard reset and a Safe Mode load and then a reboot, and it's 50/50 as to whether the Titan Z will reappear once Windows loads.
Another dance with DDU and put in the latest 417 drivers changes nothing.
(to clarify: my primary machine uses the Pascal and 1080ti, so putting one of those in place of the Titan Z is not where I want to be - I did it for about 2 days last week, and while I had 2 machines that were roughly equal, the loss of rendering speed in the one machine was excruciating. I went from one fast and one slow to two "ok I guess, but not as good").
To make sure it wasn't the MOBO going out, I replaced the Titan Z with one of my 780tis, and everything's fine. Swap that out for the other 780ti, everything's fine. Swap that out for a 980, everything's fine. Swap that out for a 1080ti, everything's fine. Put the Titan Z back in, and it's a dice-roll.
Magic 8-Ball says the Titan Z is having a terminal issue, but I'd prefer as many second opinions as I can get. And a Magic Cure. if there is one.
I keep telling myself I shoulda sold it 6 months ago during the GPU shortage when miners were paying $1500 for them used.
I've got a couple of other things to try before I take a hammer to it.
================================================================
EDIT:
Ok, so I went back and got the latest 417 drivers loaded *without installing GeForce Experience* this time. Card shows in Device Manager, still the same issue with GPU-Z (except PhysX capability check mark comes and goes - check it once it's checked. Refresh and it's unchecked. Refresh and it's checked. Or not).
I should note this exact GPU worked previously in this exact machine some few months ago, except for what I believe was an overheating issue that caused BSODs. But then, it was right next to the PC's PSU. I've got things shifted around a bit so there should be more clearance and less heat. Still, it is possible the heat issues have bjorked something.
At any rate, I got it loaded and stable. Disabled Multi-GPU in NVCPL, then check GPU-Z and it shows SLI Disabled, so that's how that gets done on a Titan Z. I had no idea.
Fire up Daz Studio with a basic single-figure scene and select one Titan Z for Photoreal device (it shows 2, obviously). Renders just fine. Run the device monitor function of Nvidia-smi as that's going on and it's clearly Side A of the card. Side B shows no activity.
Render completes in 30 seconds, device monitor shows everything at rest.
Deselect Side A and select Side B, render completes instantly without actually doing anything. No figure in the image, just a transparent background. Clear sign the card is being skipped completely.
I doubt Nvidia will want to see the debug dump from a product they stopped making.
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
EDIT again because replying to your own post inserts it as an Edit:
I may have found the issue: In Device Manager>Properties sheet, on the Details pane, if I scroll down to Bus Reported Device Description, it says 3D Video Controller for Side 2 (the bad one).
Side 1 (the good one, that has all the outputs on it) shows as Video Controller (VGA Compatible).
Both are listed under Display Adapters, as they should be, so at least that part's right.
Shouldn't both of these show as Video Controller (VGA Compatible)? Anybody got a Titan Z that works can check this?
MORE EDITS:
DDU doesn't show a vBIOS version for the 2nd half of the Titan Z. Shows it for the 1st half. Shows it for the K4000. Hmmm. Fishy.