Nvidia ,Rtx2080ti,2080,2070, information thread. Reviews and prices September 14.

[Kenmitch]

That must be a Best Buy or Nvidia fulfillment center.

Seeing how the chic in the background is wearing blue and she's not cute my moneys on Best Buy.

The box art is lacking....I hope it's not a indication of performance.

 

[SteveGrabowski]

What a load. These cards are less than a week out and we get unboxing videos instead of benchmarks. These cards are going to suck so much.

 

[Malogeek]

What a load. These cards are less than a week out and we get unboxing videos instead of benchmarks. These cards are going to suck so much.

Releasing benchmarks before the 19th would be a breach of their NDA. With reviews out the day before, it's likely very difficult to cancel pre-orders before they're charged and shipped.

 

[MrTeal]

Architecture deep dives are coming out.

https://www.techpowerup.com/reviews/NVIDIA/GeForce_Turing_GeForce_RTX_Architecture/

 

[SteveGrabowski]

Releasing benchmarks before the 19th would be a breach of their NDA. With reviews out the day before, it's likely very difficult to cancel pre-orders before they're charged and shipped.

The review NDA expires on the 19th? Holy crap Nvidia is pulling a Ubisoft. These are going to be such crap.

 

[crisium]

First real benchmarks
https://videocardz.com/newz/nvidia-geforce-rtx-2080-3dmark-time-spy-and-fire-strike-scores-leak

In timespy rtx2080Fe is 6% faster than my MSI1080Ti gaming(default).But FE edition is this time also overclocked version...
In firestrike i have 32000 after oc vs 27000 rtx2080.

I expect FE vs FE to favour the 2080, and max OC vs max OC comparisons to favour the 1080 Ti. Unless there is yet unheard of breaching of the 2.0-2.1GHz barrier that both Pascal and Volta have.

 

[Malogeek]

Now Nvidia have added another restriction for reviewers that they're not allowed to disassemble the cards?

 

[MrTeal]

The review NDA expires on the 19th? Holy crap Nvidia is pulling a Ubisoft. These are going to be such crap.

If the leaked Time Spy Extreme score is to be believed, that puts the graphics score 54% higher than what AT has in bench for the 1080 FE and about 15% higher than 1080 Ti FE. That's in line with what nvidia said, but it's pretty disappointing to get a 15% increase in performance from a card shipping at over reference clocks vs the old reference FE for a 23% increase in launch price.

 

[crisium]

This will also help the performance-per-TFLOP of Turing. 1080 Ti still has 8% more bandwidth than a 2080 so this is a bold claim by Nvidia.

 

[Hitman928]

If I'm understanding it right, Nvidia is saying DLSS takes 1/2 the performance hit of TAA but should provide higher IQ.

TAA is already a small performance hit so I don't think you're really going to see much of a performance difference between the two, but if it works like Nvidia says, DLSS should provide higher image quality with temporal independence (does not cause blurriness when in motion).

 

[crisium]

^Then that makes Nvidia's charts from before with DLSS complete rubbish.

edit: I still think it's great tech, but not the 50% improvements those slides showed.

 

[Malogeek]

This will also help the performance-per-TFLOP of Turing. 1080 Ti still has 8% more bandwidth than a 2080 so this is a bold claim by Nvidia.

It's effective bandwidth due to the better memory compression in Turing over Pascal according to the whitepaper. It's not black and white of course and highly depends on the title and what's going on and I'm sure the marketing numbers are best cases.

 

[arandomguy]

This will also help the performance-per-TFLOP of Turing. 1080 Ti still has 8% more bandwidth than a 2080 so this is a bold claim by Nvidia.

Really the same claim and situation as Kepler to Maxwell and Maxwell to Pascal.

https://cdn.videocardz.com/1/2016/05/VC-GTX-1080-105.jpg

https://cdn.videocardz.com/1/2014/09/NVIDIA-Maxwell-GM204-Press-Slides-25.jpg

 

[dlerious]

Gamers Nexus
https://www.youtube.com/watch?v=fpTzkrBUnGo

 

[crisium]

From Tom's:

"Set to 2x DLSS, Nvidia says it can achieve the equivalent of 64x super-sampling by rendering its inputs at the target resolution, while side-stepping the transparency artifacts and blurriness sometimes seen from TAA."

https://www.tomshardware.com/reviews/nvidia-turing-gpu-architecture-explored,5801-5.html

So the old DLSS charts were likely comparing to SSAA after all.

Again though, I'm still very interested in this tech if it can provide results close to SSAA but performance hit closer to TAA. Hopefully.

 

[arandomguy]

FYI if anyone is wondering the actual whitepaper that the tech sites are publishing articles regarding is public if you want to peruse it in it's entirety and interpret on your own - https://www.nvidia.com/content/dam/...ure/NVIDIA-Turing-Architecture-Whitepaper.pdf

 

[PeterScott]

From Tom's:

"Set to 2x DLSS, Nvidia says it can achieve the equivalent of 64x super-sampling by rendering its inputs at the target resolution, while side-stepping the transparency artifacts and blurriness sometimes seen from TAA."

https://www.tomshardware.com/reviews/nvidia-turing-gpu-architecture-explored,5801-5.html

So the old DLSS charts were likely comparing to SSAA after all.

Again though, I'm still very interested in this tech if it can provide results close to SSAA but performance hit closer to TAA. Hopefully.

No, that this visual quality comparison. The performance comparison is to TAA. From the PDF Arandomguy linked above:

"Figure 22. Turing with 4K DLSS is Twice the Performance of Pascal with 4K TAA"

Basically much faster than TAA, with visuals of high multiple SSAA.

If this is true, that is the killer feature of RTX, at least for me.

 

[crisium]

FYI if anyone is wondering the actual whitepaper that the tech sites are publishing articles regarding is public if you want to peruse it in it's entirety and interpret on your own - https://www.nvidia.com/content/dam/...ure/NVIDIA-Turing-Architecture-Whitepaper.pdf

Thanks. Straight from the horse's mouth then:

"DLSS allows faster rendering at a lower input sample count, and then infers a result that at target resolution is similar quality to the TAA result, but with roughly half the shading work."

"DLSS learns on its own to produce results that closely approximate the quality of 64xSS, while also learning to avoid the problems with blurring, disocclusion, and transparency that affect classical approaches like TAA."

"In addition to the DLSS capability described above, which is the standard DLSS mode, we provide a second mode, called DLSS 2X."

"[DLSS 2X] produce an output image that approaches the level of the 64x super sample rendering"

So in summary, straight from NV:

DLSS is similar quality to TAA (while avoiding TAA problems) at half the performance hit and closely approximates the quality of 64xSS. Meanwhile DLSS 2x has unknown performance hit, and approaches the level of the 64x super sample rendering.

Try reading that 3 times. Either way regular DLSS sounds great, and depending on the performance hit DLSS 2X could be revolutionary.

 

[Malogeek]

64x supersampling at faster than TAA rendering? That's a big claim. Sounds incredible and can't wait to see extensive tests on DLSS. However since reviewers won't have access to any games with DLSS implemented, it will be awhile before we can really see those claims in real games. As with RT, Nvidia are expecting everyone to simply take everything as claimed and purchase day 1.

 

[PeterScott]

Try reading that 3 times. Either way regular DLSS sounds great, and depending on the performance hit DLSS 2X could be revolutionary.

Yeah, I guess it was too good to be true that you would get 64x SS quality and half the performance hit of TAA.

My reading is similar.

DLSS is approximately TAA quality, hints that it is better than TAA, but the approximation of 64x SS sounds more like a marketing thing, because there is huge difference between TAA 64X SS. Maybe it splits the difference.

DLSS 2X sounds nearly indistinguishable from 64X SS, but no performance hit is given. I expect performance hit is quite significant.

I expect that if these are available, you wouldn't want to use anything else. DLSS will be the best low overhead AA method, and DLSS 2X will be the best high quality option.

 

[dlerious]

Teardown from Hardware Unboxed
https://www.youtube.com/watch?v=s2o1rdh_LZk

 

[Muhammed]

DLSS is similar quality to TAA (while avoiding TAA problems) at half the performance hit and closely approximates the quality of 64xSS. Meanwhile DLSS 2x has unknown performance hit, and approaches the level of the 64x super sample rendering.

DLSS is simply an upscale solution similar to checkerboarding on consoles. It's more accurate as it uses AI. It's faster than running native resolution + TAA.

So 2160p+TAA is slower than DLSS. As DLSS works with a lower resolution like say 1800p+DLSS. This will be the same quality as 2160p+TAA, but at a much faster fps.

 

[PeterScott]

Teardown from Hardware Unboxed
https://www.youtube.com/watch?v=s2o1rdh_LZk

Looks like nice build quality. If this really is cool and quiet, with nice build quality in a compact package. It really makes life difficult for AIBs that want to sell in that price range.

Previously the refrain was don't buy the FE cards unless you need a blower because of small case.

Now the FE cards might be a VERY good all round option.

 

[crisium]

DLSS is simply an upscale solution similar to checkerboarding on consoles. It's more accurate as it uses AI. It's faster than running native resolution + TAA.

So 2160p+TAA is slower than DLSS. As DLSS works with a lower resolution like say 1800p+DLSS. This will be the same quality as 2160p+TAA, but at a much faster fps.

Huh? Where in the 87 page paper is this covered? I didn't see anything like this in the DLSS section pages 35-37.

 

[PeterScott]

DLSS is simply an upscale solution similar to checkerboarding on consoles. It's more accurate as it uses AI. It's faster than running native resolution + TAA.

So 2160p+TAA is slower than DLSS. As DLSS works with a lower resolution like say 1800p+DLSS. This will be the same quality as 2160p+TAA, but at a much faster fps.

No it is NOT an upscale.

It's a post process frame buffer AA technique that operates at the given resolution, using a DL network to properly smooth out the aliasing.

DLSS uses a smaller faster DL network to achieve ~TAA quality with less performance hit than TAA.
DLSS 2X uses a larger more complex DL network, to achieve ~64XSS quality at an unknown performance hit.