NVidia Ada Speculation, Rumours and Discussion

[Flappy Pannus]

That's kind of pathetic.

Yeah that's not really impressive for 2 node shrinks really. 4090 sure, but you're gonna be paying a hefty premium for the 4080 over the 3090's current prices, get 8GB less vram and maybe ~30% performance in non-DLSS 3 titles? Not great.

 

[Maillog]

HUGE BS from Nvidia.

I think in the 4000 this unit has been upgraded. The speed Optical Flow block in 2000 and 3000 are not very high

 

[Dampf]

I think in the 4000 this unit has been upgraded. The speed Optical Flow block in 2000 and 3000 are not very high

But then DLSS 3 should be available on these cards, but at a lower performance and/or quality. If the hardware is there, that should be possible.

 

[Maillog]

DLSS 3, но с меньшей производительностью

What's the need for it then? The whole point of DLSS3 is interpolation frames, if there is no doubling then there is no point

 

[Flappy Pannus]

@DegustatoR Yah, I'll worry about the pricing when I can actually see the performance relative to the 30 series with DLSS off etc. If you get 3090/3090 ti performance at $900 USD, that's a good deal.

But that's very close to what 3090's are priced now. And that's with 24GB.

 

[yamaci17]

What's the need for it then? The whole point of DLSS3 is interpolation frames, if there is no doubling then there is no point

Well, let's say it slows down average framerate by %30 somehow, but it will end up, theoritically, achieve %100 more frames. Don't you think it is not worth it to bring it to older cards in that case?

 

[Maillog]

theoritically, achieve %100 more frames. Don't you think it is not worth it to bring it to older cards in that case?

What if it doesn't give 100%? Would it give 50? 10? Or will interpolation completely slow down all pipeline?
As far as I understand all stages before interpolation in DLSS3 are exactly the same as in 2, you will not get higher quality with 3

 

[DavidGraham]

AMD should understand now that it's not just about classical raw specs anymore, for too long they have ignored AI in graphics, now they are outclassed twice, if DLSS 3 works as good as DLSS 2 then I can see every developer under the sun embracing it, who would hate an instant fps boost without too much of a fuss? No more gamers complaining of bad optimization or low fps!

 

[Scott_Arm]

@yamaci17 There is a new hardware unit in the RTX 40 series called the optical flow generator. It is a requirement of DLSS 3.0. For some context:

https://twitter.com/x/status/1572249913005641730

 

[Picao84]

Wait what?! Is that the optical flow hardware that is described in the RTX 4000 marketing?

If that's the case and the hardware for DLSS3 is really in Turing and Ampere, then that's HUGE BS from Nvidia.

If this is true, they might be using it to push RTX 4090 as faster than they actually are. We'll see when reviews hit. This presentation looked like a lot of FUD.

 

[DegustatoR]

But that's very close to what 3090's are priced now. And that's with 24GB.

New vs used, the same old argument.
I mean, a new 3090 is a good catch right now if you can get it for <$1000 for sure but this won't last.
We'll see at which actual price range 4080 12 will land too.

 

[pjbliverpool]

3090-3090Ti were $1500-2500 though, 4080 12GB moves this into $900.
It is a sizeable perf/price improvement albeit the prices themselves are anything but mainstream.

It is an improvement but not by the standards of previous generational shifts where a x070 class product could be expected to achieve similar performance to the outgoing king at a mainstream price.

 

[Picao84]

@yamaci17 There is a new hardware unit in the RTX 40 series called the optical flow generator. It is a requirement of DLSS 3.0. For some context:

https://twitter.com/x/status/1572249913005641730

Like someone linked a few posts ago, Turing and Ampere already have that hardware unit or one with the same name.

 

[DegustatoR]

It is an improvement but not by the standards of previous generational shifts where a x070 class product could be expected to achieve similar performance to the outgoing king at a mainstream price.

We didn't really have a 3090/Ti products until Ampere generation.

 

[hurleybird]

If this is true, they might be using it to push RTX 4090 as faster than they actually are. We'll see when reviews hit. This presentation looked like a lot of FUD.

Not FUD (fear uncertainty and doubt) but definitely a strong smell of BS.

 

[Scott_Arm]

Like someone linked a few posts ago, Turing and Ampere already have that hardware unit or one with the same name.

It's likely the the hardware capabilities are different as they are between turing an ampere. It would be pretty scummy if they artificially blocked ampere, for example, from using DLSS 3. It's possible that's what's happening. it think it's more likely there's a hardware limitation. We'll find at out some point in the future.

https://twitter.com/x/status/1572243600469295106

new hardware accelerated optical flow

 

[TopSpoiler]

Hardware-based optical flow acceleration - NVIDIA Corporation

An optical flow accelerator (OFA) which provides hardware-based acceleration of optical flow and stereo disparity determination is described. A system is described which includes an OFA configured to

www.freepatentsonline.com

 

[LordEC911]

We didn't really have a 3090/Ti products until Ampere generation.

True, but they were still on their Titan push before that.

 

[BitByte]

https://twitter.com/x/status/1572257012263657475

vewy not fast

Read the footnotes. It says at 4k with DLSS performance mode. This is embarrassing from Nvidia. I’m not optimistic about the real raster performance improvements.

 

[DavidGraham]

Like someone linked a few posts ago, Turing and Ampere already have that hardware unit or one with the same name.

Turing's Optical Flow block has less pixel granularity than Ampere, Ampere also has less granularity than Ada.

Also clocks play an important role here, high fps at 4K requires higher clocks.

Ada's Optical Flow has been beefed up and is also more capable to suit this job, which was never required from eithef Turing or Ampere.