Running multiple Graphics cards Pro's and Cons, Dos and Don'ts
So, I asked in another thread, but I didn't want to clog their thread up with my many questions. Basically, I don't know much about computers, I learn on and off when I buy a new one, and promptly forget afterwards.
My question is: Can I run multiple graphics cards in my current rig, with the aim of increasing my VRam from the current 8gb it has, my goal is to be able to render more things in one scene. Currently there are scene's I own that I can't even open for fear of getting the black screen and blank renders. I can hardly include multiple characters with quality outfits. Having a; Character, Outfit, and a scene? Impossible. Specifically I'm wondering about adding an RTX 3060 that has 12gb of VRam, to this set up:
CASE: PHANTEKS P300 BLACK
MOTHERBOARD: ASUS B550-F WIFI STRIX
CPU: AMD RYZEN 9 3900 12 CORE 4.3GHZ
GRAPHICS CARD: MSI GEFORCE RTX 3070 8GB
DDR4 MEMORY: ADATA SPECTRIX RGB 32GB (2X16GB) 3600MHZ
OPERATING SYSTEM: MICROSOFT WINDOWS 10 64 BIT
The way I would imagine is that adding an additional 12gb Graphics card to this set up is like 12gb + 8gb = 20gb = good stuff. But I'm guessing it's not that simple? Someone alluded to the idea that it would actually be worse for Daz specifically (which is the only reason I want it, so that would be an issue), but I didn't want to ask them to ellaborate there. But is that true? And if so, what would be a better option for my situation? And possibly, do they mean, worse than just having a better card, or do they mean worse than what I'm currently running?
And lastly: if adding the card works the way I hope it does, is there anything else I need to upgrade to accomodate it? Thanks in advance for your time!
Comments
No, they dont stack.. so whatever the smallest amount of VRAM you have, that's what you're limited too.
You do not get 20GB's of VRAM by adding a 12GB RTX 3060, what you get is having 2 GPU's one with 8GB's of VRAM and another with 12GB's of VRAM.
If your scene will be small enough to render on the 8GB card, with the two cards the rendering will be faster, but if the scene is too big for the 8GB card, only the 12GB card will do the rendering, unless the scene is too big for that too.
Ouch, fair enough! Next question then: so the answer seems to be to just upgrade my current card, is 3060 12gb considered a downgrade? Or should I be looking at getting a different card entierly? I was hoping to keep the price sub £600 but it's looking less and less likely.
For DS, the difference in rendering speed is negligible
So the speed between 3070 and 3060 is negligible so for the increase in VRam the 3060 would be the better choice?
They don't stack, but they are handled independently - so if one drops out due to too little memory the other(s) may keep going.
I upgraded my 3060 to a 4070 and saw significant speed bumps on rendering and dForce calculations times
Actually, you'll have 20GB physical VRAM from two cards but only the VRAM are not pooled (without a NVLink + SLI...). So you'll just render with "VRAM on multiple GPUs" independently rather than with a "VRAM pooling". Scene-based VRAM consumption will be allocated to each card on average.
As an example, I have three cards, two of them are old Qaudro 6000 (24GB VRAM each). I can render a pretty complex scene that consumes appr. 42GB VRAM with two Quadros without VRAM pooling.
Well, in your case, 3060(as Display card) + 3070, no memory pooling, you still have chance to render bigger scenes with both of them... For instance, with a rough and conservative calculation, a scene that consumes 12-13GB VRAM.
Now, with latest DS 4.22.1.58 Beta (Iray 2023) plus latest Nvidia Driver 551.23, the calculation of VRAM consumption is a bit more complex than before, I'm still testing...
The 3060 is slower. How much that matters depends on the user, and yes that would technically be a downgrade because your renders will be a good 20% slower than you get from your 3070 now. However the 3060 can render scenes that are larger than what the 3070 can render, so in that regard it can actually be an upgrade, because the 3070 is a paper weight for anything over 8gb. So that 4gb can be quite a big deal, but only if you really make scenes large enough to use it.
You can use multiple GPUs, but each GPU must be able to fit the scene. If by chance you bought a 3060 and kept your 3070, they would both run together as long as the scene fits each one. Meaning all scenes under 8gb would run on both cards, and when that happens, yes, you get a massive performance boost because Iray scales pretty well. However, if your scene is over 8gb, the 3070 would no longer render, leaving the 3060 by itself. And if the scene is over 12gb the 3060 would no longer render and you would be on CPU (let's watch grass grow) mode.
So CUDA cores can stack, but memory does not. The only way to get more memory is to buy a card that has all the memory. If you want more than 12, your options are limited but they do exist. There is a 4060ti with 16gb, the brand new 4070ti SUPER with 16, and the 4080, plus the also brand new 4080 SUPER. And that is kind of it, unless you find some professional Nvidia cards which are not worth bothering with because of their prices.
Thanks guys! I was definitly one small step away from making a mistake, I'm glad I came here first aha! Based on what I'm reading here it's looking like I might want to look into a 4060ti with 16gb, when upgrading, do people usually have to upgrade other aspects of their computer? Like does anything about the card put too much stress on other components?
Hello,
i have 2 RTX4090s and in Blender i managed to use them both, when i render i see in the task manager both at a certain equal percentage. But with DAZ although both are checked, only one works. Cuda 1 (wddm) and cuda 2 (wddm - display). is there a way to make them both work at once, i don't mean adding Vram, which is not possible, but to share the work for speed?
Also were do i see in daz the project size in vram?
thanks
R.
What makes you say they're not both used? Task manager is not really a good indicator of when the card is used to render, you should check the log in DS to see what devices it uses.
As for the amount of VRAM the project will use, there's no way to see it before rendering. I'm not even sure it's still in the log when you render.
Nvidia removed the VRAM usage information from the log after DS 4.15
One can monitor the complete VRAM usage with GPU-Z, https://www.techpowerup.com/gpuz/
But, that is total usage out of which some 4GB's is used by the OS, DS, the scene and the necessary 'working space' (non-RTX cards use an additional 1GB for emulating the RTX functions)
What is available for Iray rendering is the amount of installed VRAM on the GPU minus 4GB's (minus 5GB's on non-RTX Nvidia GPU's)
You could do a test in DS and render a scene with both checked in the Advanced render tab and then render with only one checked and comapre the times. Also, you could easily install HWMonitor - https://www.cpuid.com/softwares/hwmonitor-pro.html - and it will tell you if both are being used