I’m also confused with NVlink. If I understand correctly SLI HB bridges are not NVlink.
Also, NVlink has a graphics (SLI) and compute (TCC) mode, where at least the latter is capable of sharing memory. However, GeForce RTX cards seem to support only SLI mode. Bummer.
… and I’m really unsure of the capabilities of the RTX Titan
My primary workstation is an intel/AMDGPU (Hades Canyon) with Wayland. It’s an iGPU, but a powerful one enough to drive multiple displays for coding purposes. Laptop is Windows/GTX 1060. My Nvidia linux box, powered on when I need it, is running a minimal Ubuntu server setup. It has a pair of 1080s in it. No sli bridge. I use it for nvidia-docker. X and CUDA are not installed – just the drivers. I haven’t attempted Wayland on Nvidia. The last I tried was when I had 900 series cards and I don’t recall it working.
RE: NVLink. The marketing materials make it sound like shared memory is supported on the Titan RTX you mentioned, but I am not sure. You shuold wait for an Nvidia rep to answer for sure before making any purchase, of course.
100 GB/s interface. The result is an effective doubling of memory capacity to 48 GB
Thanks. Yes, I was wondering if one can run, e.g. Gnome Wayland on the iGPU and keep the Nvidia cards for computation (as suggested).
I tried Debian Gnome Wayland on the Nvidia GPUs with Nvidia proprietary drivers (needs to be force-enabled). It works but crashes quite a bit (especially with multiple monitors).
You’re getting a bit confused I think by the capabilities when running Windows. With Windows, two drivers (driver models) exist. The “Game Ready” driver and the compute (TCC) driver. The latter is not available for GeForce type cards, only for Teslas (and maybe Quadros, don’t know)
Linux only has one unified driver so there’s no such hassle. Take a look at the Link you provided, last paragraph:
Answer: works.
I see
I tried Debian Gnome Wayland on the Nvidia GPUs with Nvidia proprietary drivers (needs to be force-enabled). It works but crashes quite a bit (especially with multiple monitors).
We have similar taste. I run vanilla gnome (no Canonical customizations) on Wayland on Ubuntu. Haven’t had it crash yet but there are some issues with visual corruption from time to time. I am running multiple monitors and the connection to one of them is flaky, so that may be part of it.
You can indeed keep Nvidia cards in the same machine just for computation, but many motherboards will disable the iGPU when a PCI Express GPU is present. I am not aware of any offhand that don’t and you’ll likely have trouble Googling for one with all the Optimus info out there. In my case, I’m using two separate machines and tend to power down the gpu box when not in use.
There is a studio driver as well on Windows and that works with GeForce. I run it for stability since Adobe apps and Resolve regularly choke on the Game Ready drivers. I think this is a recent change as the drivers used to be Quadro only, IIRC.
Yes, I forgot about that though technically it’s not really a different driver (model), it’s just “certified to work” i.e. tested with pro graphics applications.
Asus MBs can enable the iGPU in the advanced settings.
I just switched to the iGPU for diplay. Works great with Wayland and the Nvidia cards are now reserved for compute. This is pretty cool - thanks for the suggestions.
Nice to know. You mind sharing your motherboard reference? Could be useful for others looking for a similar setup.
Sure thing, see
https://www.asus.com/support/FAQ/1017796/
I set also the primary display for good measure. Otherwise, no configuration is needed
lspci | grep VGA
00:02.0 VGA compatible controller: Intel Corporation HD Graphics 530 (rev 06)
01:00.0 VGA compatible controller: NVIDIA Corporation GP102 [TITAN Xp] (rev a1)
02:00.0 VGA compatible controller: NVIDIA Corporation GP102 [TITAN Xp] (rev a1)