Reviews Nvidia GeForce RTX 2080 Ti and RTX 2080 Founders Edition Reviews

Don't feel too bad, I doubt anyone who ordered one is going to be upset by the cost / performance gain (from 1080 series) ratio.
I have a 2080Ti on order and am comfortable with the cost, if i wasn't I wouldn't have risked ordering the card plus cancelling is always an option I'm sure.

I do realise it's no where near a sensible decision but i'm upgrading from a 780ti that i bought on the day of release and that has lasted me very well so if this card lasts 5 years It's not too bad an "investment".

 

Difference being, the 780Ti didn't have new fixed function hardware that could be underpowered if ray tracing takes off - all unknowns at this point. Probably would have been better off going for a 1080Ti and putting the ~£450 towards an earlier replacement.

 

Well it's about 7% faster then a 1080Ti, until you apply DLSS when it destroys it.

DLSS is basically a resolution jump for free (e.g. a 2080 can run a game with DLSS effectively at 1440p at get an output at 4k which looks slightly better then actually running the game 4k with TAA and no DLSS).

This is obviously dependent on games supporting DLSS but as it's a post process (so easy to add to a game) and Nvidia are doing the deep learning work for free right now I'm guessing it's going to be supported by most new games soon.

Not that I have any intention of buying one, but it does bode very well for the 3xxx series which will probably be my time to buy.

 

I've not spent more than ten minutes reading about any of it because of the LOL price so I may not have grasped it correctly, but is DLSS a type of FSAA? in other words, you need to apply it to make a difference?

The only reason I am now sort of interested is because I don't use any sort of AA.

 

Yes you need to apply it, it has to run on each frame and Nvidia have to make the algorithm it runs individually for each game so it won't work for existing games.

It's not a type of AA but it can be used to apply AA. It's an image post process that uses machine learning to effectively take the image that was rendered and then improve the quality of it (which will probably apply AA, but could also upscale it too). Because it uses the tensor cores not the cuda cores (which are the ones used to render the frame) it has practically no cost. Because the algorithm the tensor cores run for a game was generated by a machine learning super computer for that game and that purpose it does a much better job then traditional image post process.

 

Most likely down to both Intel and AMD being horribly stingy when it comes to pci-e lanes.

 

more bandwidth externally with Nvlink than through x16 PCIe, and so less contention.

 

We didn't get SLi bridges for free. They used to come with motherboards only. I bet we paid for them

Crossfire bridges were "free". You got one with a board and when you bought a GPU.

 

SLI bridges only had to handle 1GB/s, and could almost be run over a bit of string (you could do such fancy larks as soldering a several together for odd GPU arrangements). NVLink is 50GB/s (100GB/s on the dual-lane 2080Ti), which puts very harsh constrains on signal integrity (the same constrains that have kept everyone on PCIe 3.0).

 

Was just watching the DF video on DLSS and I noticed they zoomed in a one point and the DLSS image seemed lower quality compared to TAA. Looking at B-Ts photos it seems odd. Most of the time DLSS looks less blurry that TAA, but looking at the ground below the car the TAA image does seem to be a bit sharper to me. DF did note that TAA in FF does seem to be a less than ideal implementation. So it may not be a fair comparison, regardless, DLSS does look pretty good at taking a 1440p image and making it 4k.

What has got be thinking though, is that Nvidia have mentioned 4k when talking about DLSS, the benchmark is locked to 4k, does this mean it won't be great at anything lower? If you were going to use it at 1080p the resolution of the image it'd be upscaling would be lower, perhaps the image quality wouldn't be so good. Will be interesting to see when we have more games that can be tested!

 

IIRC Nvidia aren't training any NN outputs below UHD.

 

So no good if you aren't running a 4k screen then? @bit-tech Someone pointed something interesting out on Anand forums, the last photo with the blue car there's a difference between the blue cars. Is the TAA one missing rust or has the DLSS one added it?

 

I'm surprised no ones mentioned DRM what with having to download algorithms for DLSS.

 

I think that's mainly because there is nothing available to the public yet. It seems like somebody (nvidia?) may be blocking developers from handing out work in progress builds using RT or DLSS to reviewers. Except that one FF XV benchmark. There is no push to show off da powah of teh arrr-teee-eggs' new features... apparently because there just isn't anything to show yet.

 

Performance boost is nice... price is utterly ridiculous. I'll be looking forward to buying a 1080Ti when they come down in price.

 

Yeah if I was in the market right now I honestly think i'd look to a 1080Ti.

Then again I am hopeful of AMD's next offerings, I really want to try out my freesync

 

You could render at X, use DLSS to upscale to 2X, then scale back down to your desired resolution (e.g. using DSR's kernel-based approach). Probably just makes more sense in that case to go for regular MSAA or SSAA though.

No DRM, just using an already existing updater to deliver and update the trained NNs used (in order to update them separately from full driver updates, unlike the situation with SLI profiles). You can bet that if Nvidia created a seperate downloader & updater for it people would be complaining about "I'm already using GFE, why do I need yet another program?!".

 

I guess some people don't get sarcasm even when you include a wink emote.