Hacker Read top | best | new | newcomments | leaders | about | bookmarklet login

Sorry, I think you misunderstood. This is installing both types of card in the same computer: one for desktop/Wayland (AMD), and the other for ML/CUDA (Nvidia).


sort by: page size:

Couldn't you use the Nvidia card for CUDA and just have an AMD card for the display. It's a little dumb having two graphics cards, you need a spare PCI slot and graphics cards are expensive (but you could just get a cheap old and used card, if it's just for your desktop environment).

I'm asking because it seems like no one relying on Nvidia card for number crushing and who want Wayland seem to have opted to just get a second card.


I think that, for my use case, it would be a good combo to put both an Nvidia card and an AMD card in the same Linux machine, using the AMD one on Linux and spinning up VM guests using the Nvidia card for Windows gaming or for ML work. This way I can use Wayland (especially the Sway window manager) for my development workflow while being able to use CUDA and stuff.

Two separate Nvidia cards, not SLA'ed.

The article mentions 48gb workflows. Presumably two cards attached? What does this look like in the AMD world, I'm used to what Nvidia offers

The two cards show as two distinct GPUs to the host, connected via NVLink. Unification / load balancing happens via software.

Ah I see, I have no experience with dual-GPU setups.

You can get Nvidia and AMD cards to live together in the same machine: https://www.youtube.com/watch?v=YqkI7bOfRkA

Very, very cool!

I've set up a dual-GPU system in the past using two nvidia GPUs and whilst I found the trek towards PCI passthrough to other virtual machines rewarding when it finally worked, I also found the arrangement to be inconvenient.

What you've achieved here, seems the ideal. Well done :)

I will either patiently wait for an Arch Linux version of the install, or I'll eventually end up impatient and see if I can rustle up something - an install script is an install script, it should be just a matter (famous last words) of altering the install script/procedures to suit.


What do you mean? The two GPUs or the part where you need to give the VM a GPU if you want that VM to have a GPU ?

The article said NVLink. Wouldn’t that mean this GPU module instead?

https://www.microway.com/hpc-tech-tips/comparing-nvlink-vs-p...


Well, not exactly. AMD calls this Dual Graphics and it only works with certain graphics cards. In the case of Kaveri this is like 2 entry level R7 series cards that use DDR3 memory instead of the traditional GDDR5 for more powerful graphics cards.

This makes it almost pointless in my opinion as you can get a discrete card that is more powerful on its own then the combination for not much more money.


So I need two machines to used this GPU? Thats even worse. I know some trickery with an onboard CPU maybe works and thought you meant that.

> Unless you are using multiple GPUs or another PCI device

You mean like having a GPU and one or more PCIE NVMe storage-devices?


Dual GPUs should be considered normal/consumer grade setup, hopefully they'll add it soon, on 4bits it's enough with plenty of space for context.

This whole thing is a fork of llamacpp, also hoping it'll all go upstream sooner or later.


I have a hackintosh with dual 280x's running in crossfire. Not sure what exactly you meant by that sentence.

I believe this is mostly for memory capacities. PCIe access between GPUs is slower than soldered RAM on a single GPU

You can plug nVidia GPUs in AMD boards...

I only have one card. I swapped the Titan V with my 1x1080 in my old, cheap motherboard and it just worked. I had to hook up the water cooling, but I did that because it came with a block, not because I optimized the thermal design. To verify motherboard compatibility, I looked up in the nvidia specs how many PCIe lanes and at what speed I should expect and confirmed in HWinfo that they were active in that configuration -- much like I'd look at a network interface to make sure my 10/5/2.5GbE hadn't turned into 1GbE on account of gremlins in the wires.

I'm not using this for machine learning, so you might want to talk to someone who is before pulling the trigger. In particular, my need for fp64 made the choice of Titan V completely trivial, whereas in ML you might have to engage brain cells to pick a card or make a wait/buy determination.


I think they mean the exact same model data (for the GPU). I don't think it requires the exact same model of GPU.
next

Legal | privacy