I am pretty sure that Cycles got NVLink support a while ago.
From the release notes of 2.90:
NVLink support for CUDA and OptiX . When enabled in the Cycles device preferences, GPUs connected with an NVLink bridge will share memory to support rendering bigger scenes .
The first scene was in Maya using Arnold 4.0.1
The second and 3rd were using V-ray next 4.3
I think those are scenes for those specific render engines for benchmark purposes (might be wrong).
Stream is live now. 3000 series supposedly has double the performance of 2000 series (including a slide specifically claiming ~120% performance improvement in Cycles)
RTX 3090 sure looks great on paper! It’s coming out next month! I should sell my 1080 ti and 1070 now before they become paperweights.
I purchased an overclocked RTX 2080 ti earlier this year and have been using it with E-Cycles and Blender 2.83 as I learn 3D.
Should I sell the card before the value significantly decreases and upgrade to an Ampere card? ( I checked and there seems to be a large value drop already )
My main concern is stability with Blender. I don’t know if it would work with E-Cycles No information available on Blender Market yet.
It’s probably best I wait for users to report on stability after they’ve tested the cards more. What would you do?
It depends why you want to upgrade.
The value of RTX 2000 will drop like a rock and it’ll also age badly. NVIDIA is very aggressive with pricing because of AMD and the price can drop even further if AMF showcase a great product.
Unless you need better performance it’s already too late to sell at good price. I doubt there will be many new sales of RTX 2000 series until we see solid benchmarks of Ampere series to compare price/value. Even then it doesn’t makes sense to buy first generation RTX cards. The technology is much better in Ampere and the pricing is very good.
I myself wanted to buy 2080 Ti badly but thank God I’ve waited.
I’m still in no rush to buy Ampere and can wait a couple months for AMD release to makes sure I get new Nvidia cards at best possible price.
Let’s be absolutely clear what that relative performance graph shows before people swallow the nVidia marketing BS without doing any thinking.
The graph shows the relative ray intersection performance compared to Turing. This does not mean you’re going to get a doubling of render speed.
There are two stages to rendering, ray intersections and shading. RTX accelerates ray intersections not shading.
In a scene that is equally limited by ray intersection and shading an infinitely fast ray intersection engine can only reduce render performance by a maximum of 50%.
Be very careful before believing that Ampere offers double the rendering performance of Turing without independent and trustworthy benchmarks i.e. the opendata benchmarks.
nVidia have a habit of choosing edge cases where their new gen GPU out performs the previous gen, they put up a slide and people interpret it as a general case. nVidia are masters at pulling the wool over your eyes. Watch back the Turing launch and you’ll see some outlandish BS.
Like Infograph said, Nvidia marketing always exaggerate the performance with pretty graphics. I would say 2080ti will be plenty good for the next three years, unless you’re rendering animations in 4K on regular basis,. When it comes to rendering, it can never be fast enough.
this aged worse than milk outside on a summer day
Haha Yeah, I think pretty much no one could predict what we’d have by now with GPUs and pricing. It’s just a complete garbage what is happening and I for once have no idea when it’ll normalize.
I think it’s a new reality most hate. Except maybe miners.
It is time to think of purchasing your own nuclear reactor if you can afford one (and if getting one is legal in your part of the world).
Nvidia is currently readying the RTX 3090Ti, which will bring a slight performance increase and an eye-popping 450 watt TDP.
New graphics card power plug for up to 600 watts is coming (12VHPWR) (guru3d.com)
It gets worse, initial rumors are talking about Lovelace having freakishly power hungry entries up to 600 watts. With a dual GPU setup alongside a 350 watt Alder Lake chip, we are talking about your machine potentially becoming a fire hazard (that is if you don’t need to rewire your house because there’s no wall outlet that can output the juice it needs).
Those AMD Ryzen APU’s suddenly don’t look so bad in terms of the likelyhood of having a PC that still runs after a few years.
Stats like those, coupled with the street price of components point towards, what Alexey described as, the new reality most will hate.
The low/entry level is dead, none of the key companies offer anything of current gen serving that end of the market. Even “mid-range” is now a couple of tiers more expensive, and a tier lower in spec. The 60 class cards come with heavily cut down dies, reduced memory bandwidth and in some cases, only 8x PCIE lanes, all while trading for 500 and over.
This “shortage” situation has been going long enough to the point where the “consumer” has grown accustomed, allowing the companies to restructure their product stack accordingly. Higher sell price means higher margins, more powerful and power hungry components for the high end, none for the low end. Good for those with deep pockets, not so great for those on a more limited budget.
Who knows, come time to upgrade in a few years time, maybe an APU from apple will look like a great deal.