Nvidia 30 series cards

[GregCooper]

These seem like a massive generational jump. Do Chaos have any test results from these new cards?

[Nicinus]

I wonder when this generation will find its way to laptops.

[squintnic]

I wonder if NVLink will work with the 3090s
48gb will be nice and make Vray GPU more usbale on complex scenes until they can make OOC work like the RS one

[Morne]

There's NVlink only on the 3090, not on the 3070 and also not on the 3080

[miro.digital]

Not sure if I'm the only one, but not really keen on the 30xx series. Excluding NVlink on the 3080 and 10Gb RAM only is an obvious middle finger for the prosumer market, kinda surprised :/

Stacking 4x 3090s in a single workstation is out of the equation being 3 slot wide, not even sure if 4x 3080 blower will work due to the new fan design - hesitant to watercool a $10k workstation. Other than these my main concern is power...I mean you'll need at least a 2000W PSU to power 4x 3080s and the rest of the system, but the selection is rather limited, only a few manufacturers have 2000W PSUs.

Really puzzled right now which way to go, might try 2x 3090, but will be obviously slower than 4x 3080. Would love to hear your thoughts guys!

[MDM]

Im really disappointed that the 3080 only has 10 gb Vram and no NVlink It seems to be a BEAST 8704 Cuda cores vs 2080ti 4352.

Currently running 2x2080ti with nvlink and I really need that shared memory as it is now! Wondering if the 3090 will be as fast as 2x2080ti~ considering the insane 10496 Cuda cores, waiting for some more news about that down the road!

Thinking about maybe upgrading the GPU on my other machine an old Xeon workstation with a 3090 if that's even possible with custom connectors.

Hoping a 3080 ti with 16 or 20 gb with NVlink will surface at some point but that could be a year down the road or not at all.


I guess if the AMD Big Navi and RDNA 2 GPUs are fast cards(Gaming) with good amount of Vram - Nvidia will Add more cards to the lineup faster in 2021!

[Alex_M]

Same here. Looks like Nvidia cheaped out on the VRAM on every card below the RTX 3090. My 1080 Ti from 3 years costed the same as 3080 and has 11 GB. 3080 Should have had at least 12 GB.

[Macker]

I was pretty impressed with the whole thing tbh. I'd trade my dual 2080ti's for a 3090, or better yet dual 3090's NVLinked.

[spencerrayfitch]

Wonder how well the 3090 would compare to 2 RTX Titans in speed. I dont even think I can hold two 3090's in my machine.

[GregCooper]

It’s a massive update. At least on paper a single 3090 looks like it could possibly be faster than 4x 1080tis and it will easily outperform 2x 2080ti, possibly even 3 with the increases to the speed on RT cores in the 30 series. Looks like they haven’t released specs yet on the numbers of rt cores so that will be interesting to see. Though they did show this image I’ve attached which looks like they’ve made a lot of progress on their rtx tech

[bildform]

I think there will be a 16gb 3070 and a 20gb 3080 in the future.

[Neilg]

I'm finally starting to think GPU rendering might be worth it.
Still needs a $4k machine though, in this age of ryzens all our workstations are under 2 grand.

[squintnic]

Id be interested to see how a typical DBOX scene goes on GPU Neil. Arch vis scenes can get horibly heavy very fast.
Im looking at a renderbox https://renderboxes.com/ and trying to work out the NVlink and price performance of a 4x 3090 setup with them.

i dont do much architecture anymore but still have heavy scenes so the extra vram very useful, pity NVIDIA scamming people (except check their share price go up!) at the prosumer level.

the power draw is concerning given im forced to work at home and I dont think my circuit will take that kind of load. stduio space has much better power.

[Neilg]

Worth it is a pretty loose phrase... We'd have one box running Lavina or something.

We've got 2 films on the go that after extensive optimisation smash up to 128gb and max it out the whole time - we definitely can't use it for real, but I could see it being useful to spin off a set of scenes with a material override to render out storyboard animatics on gpu. I would not be running typical scenes through it - I already know it would let us down just as things are getting hairy.
the big issue with gpu is still that to get decent results you have to commit to working in gpu the whole time. That's such a risky move for us, so many projects have scope creep over a couple of years. Got an old project which was set up as nice self contained views and now we are doing 16k 360's with displacement on, and the only reason we got the extra scope was how quickly we can turn it around.

[Morne]

I'm finally starting to think GPU rendering might be worth it.
Still needs a $4k machine though, in this age of ryzens all our workstations are under 2 grand.

What kind of specs are those sub 2K machines Neil?