Comment on Looking to build my first PC in almost 30 years; What should I be on the look out for?
CosmicTurtle@lemmy.world 4 months agoTwo GPUs? Is that a thing? How does that work on a desktop? Honestly, if it wasn’t for my curiosity into AI, I’d just go with the onboard video though given my need for specific resolutions, I find comfort in having a dedicated card.
I’ve been using ubuntu exclusively for 10 some years and don’t use snap at all. tbh, not even sure what snap is.
If it’s not apt, then I don’t use it.
You do use it as a bunch of snap packages automatically install the snap instead.
For Nvidia I still think passthough is the best option as it isolates the Nvidia issues to a VM instead of the host. If you aren’t going to spend a bunch of time on AI then you can just use a CPU as long as you have enough ram.
Two GPUs? Is that a thing? How does that work on a desktop?
GPUs these days aren’t like your old Voodoo, with its daisy-chained VGA port and one-way, fixed-function graphics pipeline. They can actually send the results of their calculations back to the CPU over the PCIe bus instead of only out to the monitor!
(In all seriousness though, you don’t actually need two GPUs.)
Two* GPUs? Is that a thing? How does that work on a desktop?
I’ve been using two GPUs in a desktop since 15 years ago. One AMD and one Nvidia (although not lately).
It really works just the same as a single GPU. The system doesn’t really care how many you have plugged in.
The only difference you have to care about is specifying which GPU you want a program to use.
For example, if you had multiple Nvidia GPUs you could specify which one to use from the command line with:
CUDA_VISIBLE_DEVICES=0
or the first two with:
CUDA_VISIBLE_DEVICES=0,1
Anyways, you get the idea. It’s a thing that people do and it’s fairly simple.
SNAP is just a proprietary packaging by Canonical. Basically the same as a flatpak, but fully controlled by Canonical, store and all. Integrated graphics will give you as much resolution as most GPUs, albeit they won’t be able to render at dedicated GPU speeds. But unless you’re actually rendering very heavy videos, integrated, matched together with 1 or more CUDA TPUs, and YOU set the limits.
If you don’t need a lot of GPU horsepower besides the AI stuff then you could just use the integrated graphics and have a dedicated GPU for the AI stuff.
Having multiple GPUs in your system isn’t really that special. Plug HDMI into GPU1 to make GPU1 drive your display/play games. Plug HDMI into GPU2 to have GPU2 do stuff. If you’re doing AI work then you don’t need anything connected to the GPU, the program just needs to know it’s there and to use it.
The only thing to look out for when using the iGPU and a dGPU is that the bios doesn’t turn off the iGPU if it detects the dGPU. If you have 2 dGPUs then it shouldn’t matter outside of maybe the bios wanting to use the first one.
catloaf@lemm.ee 4 months ago
You put one GPU in one PCIe slot and one in another. Just be aware that just because a PCIe slot is full-length, it doesn’t mean it’s a full-speed x16 slot. Check the manual to be sure. Most cards will work with fewer lanes, but not all. And of course it’ll be slower. (Fun fact, you can put a long card in a short slot. Some have open backs to allow this, but if you have an oscillating Dremel tool and a steady hand, you can make your own.
But personally, for basic 3D games and work, I’d just use the integrated video (which is on the CPU these days, not the motherboard) and give the discrete card to the VM.
CosmicTurtle@lemmy.world 4 months ago
With the conversations I’m having here, I’m leaning in the direction of integrated video (assuming I can get one with display port) and a discrete card just for AI work.
I use VirtualBox for VMs. I’m assuming there are instructions on how to give the card to the VM? My cursory google search came up with dubious results.
catloaf@lemm.ee 4 months ago
Looks like PCI passthrough was dropped in virtualbox 6.1.0. You may want to use Qemu or KVM.
jjlinux@lemmy.ml 4 months ago
That.
jjlinux@lemmy.ml 4 months ago
Most new boards will have at least a display port and an HDMI port. Add that most also have Thunderbolt4, plug in an HDMI or Display Port dongle. The sky is the limit man. On the VM panorama, VMWare is now all fucked with their forced subscription model, Virrualbox is still a thing, but GPU passthrough (I’ve heard, can’t really confirm) seems to have turned into a real shitshow. KVM / Qemu seems to be the only alternative that makes sense right now.