OTOY Forums

I had a fully watercooled system with 3x1080tis built for my workstation around 5 years ago…Getting fed up of Vram limits and Octane blowing up and crashing when I add a tree into my scene and copy it a few times..

So I just wondered, it’s insane to pay like 5k for a single graphics card…But do Nvidias ‘profession’ line of cards really have benefit over the RTX cards…I mean the 48gb of vram in the sounds very appealing, I could definitely push my work further with that huge amount of breathing room.

But are there any other benefits to these cards? Bare in mind I could get 2x4090 cards which would equate to almost 3x more CUDA cores than a single A6000…

So is this just a marketing trick to make you think you’re getting stability when in reality it’s just the same card as an RTX with just a bigger vram? Are there other components in them that make these better cards for GPU rendering if money is no object? And if so does Octane take advantage of it?

Would you get less crashing? Would it load heavy scenes a lot faster? I’m getting to a point where I’m fairly happy with the speed of my 3x1080s…My issue is purely stability and Octane crashing when I’m trying to improve my work and tired of having to compromise by losing details.

overall, A line cards differ very little compared to gaming cards, however there are few small differences that might make (or not) for the extra price You pay:

* the main one, for A6000 You already mentioned - that's amount of VRAM. OctaneRender and other apps, do get a bit more crashy if You push card to the edge, so having the extra space does help. Question how often You need it and if it's worth to pay hefty extra for that.

* another thing is that A (and other pro line cards) are clocked less aggressively. They run al slower speeds, use way less power and allow GPU to be less hot. That effects the cooler dimensions and for A line cards You can get them as dual slot blowers, while 3090 and 4090s will be 3-4 slots, require extra space to breath, unless You are into hybrids or watercooled cards.

* as for Pro cards and Titans they also have TCC (Tesla Compute Cluster) driver mode, that allows You to disable screen outputs and run them purely for rendering, meaning You will have full amount of VRAM, that will not be eaten or reserved for screens/windows, etc.

* stability wise,..it's subhective thing - sure NVIDIA will claim, that those cards will run a a bit more stable, and such, but in reality, properly cooled 3090 machine (we do not have too much data on 4090s yet) could be as stable as dual A line card equipped cards (with the same amount of VRAM).

* sure, for some applications that does benefit from ECC memory on GPUs, there is a benefit, same as as running ECC RAM on Your system, however..as far as I know, that's not important, not being leveraged in OR.

now for the same amount of money, You can buy 2x A5000 or 3x 3090s - that's around 50% faster if You choose gaming cards, on the other hand, now You can get 4090s, that will be around 2x faster than 3090. New ADA A line will be out soon, but the price will be higher.. Haven't seen official MSRP yet.

Few things to mention if You are thinking about whether to get 24 of 48GB GPUs would be:
* think if You need that? how big are Your scenes, how close to the edge of VRAM do You get? having extra free space will help You to have better stability for sure. But is it worth if most of Your scenes would fit into smaller card easily? maybe getting multiple of those would lead to better life as it would render multiple times faster..
* keep in mind that we do have out of core. But once You start leveraging that, RTX boost will get disabled. so there are cases where heavy scene rather than running multiple times faster on 3090s will actually be slower compared to single A6000 that would keep RTX boost (same will go for ADA based A pros and 4090s).

Overall, You can simple use gaming cards, but plan Your build according to Your needs, expandability, noise and such. There is no one answer fits all, so recommendations could be slippery here. Understanding what You actually need and what You get is the way to go. There are benefits to get multiple 3090 or 4090s, even slim them down using watercooling, but there are also benefits to get single A6000 or newer one and enjoy - it's all about Your projects and future plans.

for instance, if You are planing to dig deeper into more serious projects, might be wise to spend a bit extra now, rather than risking have stability issues, dropped speed on the long run once that bigger dream projects comes to Your studio.. however, if have streamline workflow and 24GB cards are enough, maybe simply opt for multiple 4090s for the same amount of money and enjoy cutting times multiple times..

Thanks for the detailed reply Tom!

Yeah it's pretty much what I thought...It's a tough one, if it was the sake of an extra grand for the stability it would be a no-brainer for me..However there is a significant cost difference meaning that my budget is pretty much blown on the one card and I do other tasks on my workstation, so the other components need to be high performing too.

I think the 48gb of VRAM would be such a luxury..I get alot of dense CAD models from clients so I don't have a lot of control over the mesh so I guess this leaves me stuck sometimes, especially when they want the objects rendered in a scene like a forest or city, I hit my ceiling really quickly then it's just me scrambling spending HOURS sometimes days trying to find ways to reduce the scene to stop the crashes...Sometimes even rendering things separate and comping together which isn't ideal...I guess 48gb would give me over 4x more freedom

I've had to leave it for a few years after investing in 3x 1080ti's around 5 years ago and they only have 11gb, and seems a few GB's are used before you've even started leaving me with like 9 or so..Then when you start adding textures/cad models it VERY soon hits the limit, it's almost laughable sometimes adding like 1 more object uses 1gb , to the point that even in my mind 48gb wouldn't be enough for the freedom, but it's definitely better. Just a shame you have to take such a hit financially just to get that much VRAM and compromise speed.

There's also an element of going in blind as there's not alot of content covering this line of GPU so less to compare, especially for specific use cases like Octane/GPU rendering.

The heat thing you mentioned is interesting, even though my main workstation is water-cooled it makes my office unbearably hot.

Anyway thanks again for the reply, I'll keep pondering!