What makes a rendering card so much better than something like a 4090RTX card? I see cards like the Quadro and the Nvidia 6000 out there and they are not cheap at all. Are they really worth that much money when it comes to rendering? Do they really make that much difference?
My assumption is they’re really designed for 3D professionals but for the average gamer/3D artist would they be worth buying? I’m trying to understand what makes them so expensive in the first place. I’ve seen them range from $2200 to $40,000 and I just picked up a liquid cooled 4090 for $1700 with tax and shipping. Thanks!
3D art creation and rendering, like it or not, tends to be fairly niche compared to the huge gaming market, and the memory that you can get with a Quadro for instance is far above what is needed for the average game title.
It is a combination of that and the relative lack of competition from AMD that Nvidia can get away with charging such prices, Jenson Haung is more or less the owner of the creator market at the moment (to the point where most DCC apps. have essentially become Nvidia-only solutions if you want something functional).
Ok, I guess that makes sense if Nvidia has the market cornered. Is the performance really worth the extra cost though? I mean, a 4090 can render pretty well and I understand the VRAM quantity definitely helps but $10k-$40k? Why not buy 10x 4090’s for example and build out multiple nodes? What am I missing?
You can stack a certain number of RTX 4090 cards to pool their performance, but what you cannot pool is the memory. Your maximum level of VRAM for rendering will be 24 gigs no matter how many nodes you use.
Though Nvidia did in fact introduce a new connection type known as NVLink which worked with the Ampere xx90 model, and that did allow you to combine the memory to obtain Quadro-like VRAM counts. However, Nvidia decided this was creating too good of a deal for what are marketed as gaming cards and now has NVLink as something that only works with their professional line.
Power consumption is a factor. And like Ace said there is no NVlink anymore, so you’re stuck with the max memory on one given card. That is, until hardware changes to allow sharing system memory more efficiently with the GPU
@Ace_Dragon wait NVlink still works on quadro cards ?
I see, it all makes sense now…and it’s a bit of a…no a huge power/money grab by NVIDIA; of course it all comes down to money lol… Ok thanks for the explanation.
I’m fairly new to Blender and 3D rendering and I dont really understand why VRAM is a factor when doesn’t the OS allow you to share system RAM with your video card? For example, when I pull up task manager I see dedicated GPU memory 24GB and shared GPU memory 31.7GB. The only thing I can think is that GPU RAM must be substantial faster then or…?
More v-ram is nice, but the key difference is stability.
If you are working on a CAD model of a jet engine, you need to be able to process all of that geometry with zero errors. You need it to work no matter what you throw at it, even if it isn’t fast. Gaming GPUs can get away with some minor rounding errors or memory glitches. If a single pixel is slightly off, no one is going to notice, but it must be fast.
If a high end gaming GPU is like a sports car, a Quadro is like a bulldozer. It can provide tons of power and is almost unstoppable.
Ah, very interesting… Final question, can you game on something like a Quadro? I’m thinking, if I pour the money into this card can I still game on this thing or what? What kind of performance should I expect? Similar?
You can race a bulldozer, but you’re not going to get any medals.
More specifically for blender, other than the increased ram for larger/more textures, the rendering performance is significantly worse than the geforce cards. I advise anyone using blender to avoid quadros, they simply aren’t for us.
From a raw performance point of view, the actual RAM speed between system and GPU isn’t that much of a factor. The issue is moving data between or accessing data across VRAM and system RAM.
That is WAY slower, so once the GPU runs out of VRAM and you basically start swapping, you are almost better off just using the CPU to render, as it may well be just as fast or faster.
This largely isn’t just a rendering thing, think AI, crypto in the past, etc. Hence VRAM is king when your workload needs more then your GPU has. Nvidia knows this and this is why ‘gaming’ GPU’s have usually had lowish RAM, while the pro cards have both much larger amounts and in many cases can be directly linked/pooled together.
At which point, Nvidia charge an arm and a leg and can pretty much get away with it.
Ok wow, that really hit it home for me. I think I get it now thanks. That makes perfect sense. I have a hard time believing though that rendering on the CPU would be faster if you ran out of VRAM though, that’s a tough one for me. I wouldn’t think the transfer time would make THAT much of an impact but I’m also a 3D noob so…lol, there’s that too.
Think of it like a game level and you move from one map to another, ie the ‘loading screen’. What it is actually doing mostly during that time, is unloading the data from GPU VRAM (textures etc) of the old map and loading in the new one.
So even if that is only half a second, imagine doing that hundreds or thousands of times as it renders a single image, due to the fact it can’t keep it all in VRAM as the VRAM isn’t large enough. So for say each 16x16 pixel block it renders, Cycles needs to work out what data it needs, load that into VRAM, render and then move on, swapping any data in and out.
The larger the scene, the more textures it has, the lower amount of VRAM you have, the worse it gets.
Compare to the CPU with direct access to 32+GB of system RAM and the difference in speed can add up very fast, even tho the CPU is technically ‘way slower’.
Yeah ok, that makes total sense. So it really is like thrashing then where you’re writing RAM to disk versus main RAM. Ugh…well, glad I bought a 4090 with 24GB of VRAM then, I guess that’s an excellent mistake if you will lol… Really I just wanted a video card that could do some flexing and that’s top dog for gaming and I guess, “general” 3D work.
There’s always SheepIt for bigger renders I suppose also, lots of those guys have their own farms of Quadros and what-nots.
Yeah, not as bad, since it’s still all RAM within the system, but still a huge overhead.
Unless you are rendering an animation and need to do a lot very quickly, or it just doesn’t fit in 24GB VRAM, then chances are the 4090 is all you will need.
Personally, unless ones rendering requirements are a one off or the amount is massively variable, I personally think just buying a GPU (or a few) is cheaper then most render services.
I see, have you seen SheepIt? It’s free…kind of, you’re letting them use your GPU/CPU to render and you get points which you then spend to render your stuff at no cost to you other than you’re paying your own power bill for the extra load they put on your own machine which is pretty negligible.
Can I install my 3080 card along with my 4090 card to get some rendering performance out of that? I have enough space, cooling and a 1200W Corsair Platinum powersupply; I used barely 560W tops right now with just my 3080. The 4090 is roughly 500W let’s say, on the high side. I mean, I’m sucking some major power but if it reduces my render time at home I might be down.
With Octane using “out-of-core” memory (basically swapping VRAM in and out with system RAM) only slows the render down about 10%. Although I think there might be limitations with it, like you can only do that with textures and geometry? I’m not sure and it’s been a while since I have rendered anything that wouldn’t fit in my VRAM.
Im not familiar with its usage and the only thing I know about it is it’s some renderer Nvidia came out with. Will it render Blender files and others like Maya and so on? How does it work in that regard? Is it just a render engine?
