sli is dead-ish. but multi-gpu vulkan and dx12 is technically not. and that is what we need developers to support. just that almost no one is doing it because there’s not enough multi-gpu instalations because no one is developing for the them. and that goes double for vr.
also it is not really in nvidia’s interest to have people sli’ing up two 3070’s and getting (inferior stuttery) 3090 performance.
There could be one with 12 also, at least there is still two slots for memory unused in the 10gb 3080 board.
I guess I will cry once and try to go for the 3090. Price is hard but it’s the cheapest titan you can get and focused on 8k won’t hurt us 8K+/x unseres and the like.
I guess that depends on the implementation each Gen gets a bit better but it is definitely out of the game devs focus.
Some games really get near 2x the performance, some even worse. In the end the games need to made with SLI in mind. It’s too bad so many APIs like that are so rarely used, lucky RTX is going to console so that will make its way to stay at least.
That is true, sad it’s still so little main stream accepted as of now. But hey thats where you come in! Great to see that Nvidia keeps pushing stuff like the new VR Dlss now, that’s Awsome! Hope some Titels work well on it.
Dont know if this has been posted already, but quite an interesting article regarding the cuda core count and terraflops on 3000 series. As @mirage335 previously mentioned, its not really pure 10495 cuda cores, but reading the article, it seems that depending on game and how much (or little) integer math is needed, these second half of the cudas CAN indeed be used as real cuda cores for floating point operations, which theoretically could double the performance (when up to 10495 cuda cores are used as fp32, instead of only half of it)
So, half the Ampere cores are dedicated to floating-point, but the other half can perform either floating-point or integer math, just like in Pascal.
Yes I will buy likely two of them, but not to try SLI. I need them for two PC. However will try SLI anyway… I dont expect anything…
Go VoodooDE! 
.
Hope you make a video from the SLI attempt.
I heard that Igor from Igors lab said that the tables shown from the show do come true, but of course that won’t be the case for alle games. Like you said, use loat of floating point operations, which modern games do, and the 3000’s really go running!
The latest Multi GPU things from Nvidia works nice, but the game devs must be integrated this. For VR also (VRwork) that’s the problem 
For flat games as example, Star Citicen with NvLink/SLI 2080ti, give you up to 80% boost in arena commander.
Just like you drowhunter, my 2080Ti will work just fine until some benchmarks convince me to upgrade.
I don’t believe doom eternal has ray tracing, but it does have DLSS.
Thanks VooDooDE, I’d really like to see how the SLI performs. By the way, before you try your next horror game, I think perhaps you should drink an adult beverage to calm yourself a bit. Your startled yells are hurting my ears. 
All the big companies will be going mutli gpu or ‘chiplet’ next gen (intel already have with their enterprise card). They have to do really, can’t keep going smaller. But the equivalent to infinity fabric is a hard problem to solve.
Wow didnt know they supported SLI in SC! I need more kidneys 
Yeah, have read, the MSFS 2020 should also use Multi GPU in the Future. But, when is this future? 
In the generation after this one. So what, 2-3 years or thereabouts? Unless Intel get there sooner. Navi 3 and Hopper.
Haha, I give my best 
Are you buying a Founder Edition ?
The question is also for @VoodooDE
that is not quite the same as as working with multiple graphiccards (which is often referred to as gpu’s) which for now could work fine but there is no incentive for developers. I hope they find an efficient chiplet design because that means higher smaller silicon and higher yields and just maybe cheaper cards. but it does not necessarily translate to efficient multi-card implementations.
Yeah, I kind of agree with that. But this is the way the industry has decided to go, and the way they think is the best to solve the same problem. Ie making more than one Graphical Processing Unit work scalably with another, whether it is on the same card or not, and without developer support. Obviously the further away they are in space and the more components it has to go throught the harder the solution is. So on the same card should be easier.
They are all working on it because they have no choice. They need to keep pushing advances and node shrinks are getting harder and harder and more expensive. They need to go parralel on chips as well as within them. And the benefits are obvious. Current monalithic designs hit hard performance limits linked to size (not to mention costs). If you can just scale sideways then the sky is the limit.
They have all given up on traditional SLI implementations whether that’s right or wrong. But I don’t think anyone is going to complain how it is implemented if they can get the equivalent to 4x a rtx 3090 on one card and it works just the same.