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

[Geegeeoh]

Curious, how does HDR gimp the GPU?

HDR needs tons of memory bandwidth, thats how the new cards pull ahead.

 

[Head1985]

Actually its 37% more bw. Far less than the delta between original 1080 and ti.

Shader delta is a bit higher than 1080vs ti.

TDP (power limit) delta is much smaller, only 19%

So with all that in mind it actually doesn't make much sense that it would be as much faster as claimed in the video.

That said, I don't think the 2080 in particular will be power limited when RT/Ts cores are not being utilized. hence its boost clock could potentially be capped (which would be great for overclockers)

2080TI will be 3-5% more faster vs 2080 than 1080TI was vs 1080.
Simple because 2080 is cutdown this time.Everything else is same like 1080TI vs 1080.
1080ti vs 1080-3584/2560 1080Ti have 40% more SP
2080TI vs 2080-4352/2944 2080TI have 47% more SP

 

[TheF34RChannel]

They are saying here that 1080ti to 2080ti is about 35-45% performance increase on regular games. I didn't watch the whole thing but skipped around.

https://youtu.be/YNnDRtZ_ODM

Epic find, thank you!!

So training for DLSS. Good that it's free. I do wonder if that directly translates into Nvidia titles can have it but AMD titles definitely won't?

Again, what a useful video!

 

[sze5003]

The XB273K is HDR10 and DisplayHDR 400 certification. Acer released their X27 monitor, which is about $2000 and has DisplayHDR 1000 certification, at the same time as the Asus PG27UQ. The XB273K may be a more interesting monitor, especially if it doesn't have a fan.

I'd be interested in getting one but I hear for 4k it would be nice to get a 32. I think I could fit a 32 on my desk, but if it has those side panels too, maybe not. There's also a 32 or 34 model by Asus or maybe it's still Acer that should be out 2019 or 2020.

 

[crisium]

35-45% gain in traditional games seems great seeing as it's (arguably) basically on the same node. See 35%-43% gain for 980 Ti over 780 Ti at 1080p-2160p at TPU. If true it's also more than the 15-25% (or less) that many including me were expecting.

I will of course have to mention that the 980 Ti was cheaper and had more VRAM than the 780 Ti, as compared to a 71% price increase and same VRAM here. Maxwell was widely hailed as a wonder generation and sold like hot cakes because it moved the performance-per-$ barometer upwards. That appears to be gone now and in its place is raw performance gain and features. People are paying a lot of money for those RTX and TPU cores, so they would have to deliver greatly for me to be interested in this generation at its current price points.

More than the RTX cores (in very demanding games I am not one to shy from turning down shadows or lighting), it's the TPU cores and DLSS that are most interesting to me. But I really wonder what those DLSS slides are comparing. Is the 2080 doing 4k + DLSS compared to 1080 doing 4k + SSAA or MSAA? Or is the 2080 even rendering at a lower resolution than 4k and using DLSS to mimic it? If it's either of those, is it a fair comparison? DLSS is the wildcard left for me.

 

[PeterScott]

More than the RTX cores (in very demanding games I am not one to shy from turning down shadows or lighting), it's the TPU cores and DLSS that are most interesting to me. But I really wonder what those DLSS slides are comparing. Is the 2080 doing 4k + DLSS compared to 1080 doing 4k + SSAA or MSAA? Or is the 2080 even rendering at a lower resolution than 4k and using DLSS to mimic it? If it's either of those, is it a fair comparison? DLSS is the wildcard left for me.

There has never been any sign that DLSS is running at lower res and upscaling.

DLSS is post process frame buffer AA. Watching the previous linked video reiterates that in more detail.

But given that it is NN trained specifically for each game, It could rival SSAA at negligible overhead. So probably that is what is claimed/compared, especially given the huge performance deltas. They can't be comparing to cheap AA methods like FXAA, because you don't see that much cost for running those in the first place.

I am trying not get my hopes up that this might be SSAA quality at essentially no overhead like they seem to imply. But that would be a killer feature for many, especially on the lower end cards like the 2080/2070.

The 2070 in particular. It will be behind a GTX 1080ti, but with DLSS vs SSAA on the 1080Ti, the GTX 2070 will likely be ahead.

DLSS is the one feature I want to see most detailed and analysed in reviews.

For myself, I can turn down a lot of features. But I find aliasing jarring, and usually it's a must have, and further, I hate a lot of the modern post process AA method because they blur everything.

If DLSS is low overhead, high quality AA without blurring, it's the biggest feature of the the release for me. It might have me stretch my budget to a 2070 (I doubt 2060 will have TPUs so no DLSS).

 

[Headfoot]

I'm cautiously optimistic on DLSS too, depending on whether it 'just works' on every title or if it basically depends on nVidia to push a good trained NN on a game-by-game basis. But like you say nearly zero overhead for very high quality AA would be an incredible feature, especially for 1440p and 4k where you still can see aliasing in motion but no card is fast enough to crank the AA

 

[MrTeal]

I am trying not get my hopes up that this might be SSAA quality at essentially no overhead like they seem to imply. But that would be a killer feature for many, especially on the lower end cards like the 2080/2070.

I almost sprayed my coffee out my nose, and now have a weird taste in my mouth.

This isn't meant as an attack on you at all, but this is where we're moving too. We have informed people making reasoned arguments, while calling the $800 and $600 options lower end. The market shift with this generation is remarkable.

 

[cmdrdredd]

There is a performance hit (which varies) when enabling HDR on Pascal cards.

Some people are assuming that Turing cards will have no performance hit (or at least much less) and so the performance difference is higher with HDR enabled for both but this will not reflect the performance difference for SDR.

Is that right? I’ve never noticed a performance difference with it on or off. I haven’t done extensive testing of course and I always use vsync on a 60hz HDR TV so there’s that.

 

[RichUK]

BF5 has been delayed by 1 month - this may bode well for further RTX optimisations at launch.

 

[PeterScott]

I almost sprayed my coffee out my nose, and now have a weird taste in my mouth.

This isn't meant as an attack on you at all, but this is where we're moving too. We have informed people making reasoned arguments, while calling the $800 and $600 options lower end. The market shift with this generation is remarkable.

To clarify: Lower end of RTX, is all I meant, as RTX 2070 is likely the lowest end RTX card of this generation.

Overall, I don't even consider GTX 1060 to have low end pricing anymore. IMO those terms have lost all meaning now with regards to NVidia naming schemes.

When discussing place in the overall market, we should just refer to dollar value price points. Like ~$200 range, which is normally about where I buy. I paid about that for my 8800GT.

For me, It's a huge stretch to RTX 2070, even at the promised eventual $500 pricing. I'd be happier if the 2060 could do DLSS but no Ray tracing..., though I half suspect the GTX 2060 might just be a rebranded 1070, and GTX 2050 a rebranded 1060. It looks like we have a long wait to find out.

 

[DeathReborn]

I almost sprayed my coffee out my nose, and now have a weird taste in my mouth.

This isn't meant as an attack on you at all, but this is where we're moving too. We have informed people making reasoned arguments, while calling the $800 and $600 options lower end. The market shift with this generation is remarkable.

They could be the low end of RTX cards this generation...

Oh and try some mint for that taste, clears it right up.

 

[arandomguy]

Is that right? I’ve never noticed a performance difference with it on or off. I haven’t done extensive testing of course and I always use vsync on a 60hz HDR TV so there’s that.

As far as I know this is the only site that has tested it extensively and only done so with the GTX 1080 and at 4k -

https://www.computerbase.de/2018-07...force/2/#diagramm-call-of-duty-wwii-3840-2160

As you can see it varies game to game, some of negligible impact some with pretty large impact. I suspect this means it might vary scene to scene as well (as in where they test in the game).

If I had to speculate raww memory bandwidth did not increase much relatively for the Pascal generation and so they had to rely more on other design changes (eg. delta color compression, cache structure, etc.) instead to make sure the rest of the execution units can be fed. It could be that at the time optimization for HDR was not a priority. In particular with the GTX 1080 which has a very low memory bandwidth ratio relative to other hardware (only the 1070ti would have less) so it may be the most affected. Whereas perhaps if you tested the GTX 1060, which has the highest memory bandwidth ratio for 1xxx series, it may be the least impacted. Similarly with testing at lower resolutions.

With Turing of course raw bandwidth has increased greatly. There is obviously other changes to the memory subsystem and cache structure. The compression they use could also be more optimized for HDR.

For me, It's a huge stretch to 2070, even at the promised eventual $500 pricing. I'd be happier if the 2060 could do DLSS but no RTX..., though I half suspect the GTX 2060 might just be a rebranded 1070, and GTX 2050 a rebranded 1060. It looks like we have a long wait to find out.

Well the interesting thing at the moment is that rumors seem to point to RTX 2070 not being a further cut down TU104 chip but actually a TU106 chip. Which makes it potentially interesting how they fill out the rest of the lines.

They could price the segment lower more competitive as well. The speculation points to AMD targeting that segment with their 2019 product stack and their existing product stack does cover it as well. So the competition that exists is higher. Price sensitivity would also be higher as well when we move further downwards.

 

[sze5003]

BF5 has been delayed by 1 month - this may bode well for further RTX optimisations at launch.

Yea I'm guessing rtx optimisations and also good idea not to launch it at the same time as red dead 2.

 

[cmdrdredd]

As far as I know this is the only site that has tested it extensively and only done so with the GTX 1080 and at 4k -

https://www.computerbase.de/2018-07...force/2/#diagramm-call-of-duty-wwii-3840-2160

As you can see it varies game to game, some of negligible impact some with pretty large impact. I suspect this means it might vary scene to scene as well (as in where they test in the game).

Not to go too far off topic but they tested destiny 2 and saw a drop of about 10fps or so with a 1080 using 4K hdr. My 1080ti gives me 60fps either way. So I think perhaps my configuration is at the limit of my refresh rate for some titles and that’s why I don’t notice a difference. Perhaps if I tested the raw uncapped frame rate I would see a drop somewhere.

I learned something though because I didn’t realize hdr had an impact at all.

 

[IntelUser2000]

As far as I know this is the only site that has tested it extensively and only done so with the GTX 1080 and at 4k -
https://www.computerbase.de/2018-07...force/2/#diagramm-call-of-duty-wwii-3840-2160

With RX Vega 64 there's a minimal impact enabling HDR. Also, RX 480 had minimal impact while GTX 1060 had a noticeable drop.

https://www.gamestar.de/artikel/was...dr-auf-dem-pc-ausprobiert,3271227,seite2.html

This might be one of those scenarios where Nvidia's design is lacking compared to AMD. Since its a deficiency it can be expected that in the next generation it will be addressed. It could also be a trade-off Nvidia's engineers chose to make to gain in other areas.

 

[TestKing123]

Supposedly some actual 2080ti benchmarks are out.

https://www.youtube.com/watch?v=tJLzlmLHORY

 

[TheF34RChannel]

I'm cautiously optimistic on DLSS too, depending on whether it 'just works' on every title or if it basically depends on nVidia to push a good trained NN on a game-by-game basis. But like you say nearly zero overhead for very high quality AA would be an incredible feature, especially for 1440p and 4k where you still can see aliasing in motion but no card is fast enough to crank the AA

What worries me is that AMD optimized titles may not be eligible for NN training.

 

[RichUK]

Supposedly some actual 2080ti benchmarks are out.

https://www.youtube.com/watch?v=tJLzlmLHORY

If that's legit, perhaps the 2080 will also outperform the 1080ti.

 

[TheF34RChannel]

Supposedly some actual 2080ti benchmarks are out.

https://www.youtube.com/watch?v=tJLzlmLHORY

Anyone here who speaks Turkish?

 

[Qwertilot]

What worries me is that AMD optimized titles may not be eligible for NN training.

I can't think how you'd block the NN training on a technical level.

I would presume it works by analysing the images the game engine outputs with no AA vs it with a really powerful AA added on and then learns to predict where the AA option adds pixels in. That's very like known NN uses.

If it really is such a big potential gain for the RX series then NV will definitely want to do it for as many titles as possible, and I'm really not sure if AMD have the clout these days.

 

[PeterScott]

What worries me is that AMD optimized titles may not be eligible for NN training.

When it comes down to it, AMD has more interest in stopping that, than NVidia does, because NN training will create advantages on NVidia HW even for AMD optimized titles.

 

[sze5003]

Supposedly some actual 2080ti benchmarks are out.

https://www.youtube.com/watch?v=tJLzlmLHORY

So by those graphs it shows you can get 100+ fps at 4k in battlefield 5 vs 67fps at 4k on the 1080ti? Strange how some games are so close in fps vs others you see +30-40 fps improvement.

I wish google translate worked on videos.

 

[alcoholbob]

If that's legit, perhaps the 2080 will also outperform the 1080ti.

Theres something seriously wrong with those numbers for Witcher 3.

They claim 44 fps average with max settings at 4k with 1080 ti -- that is barely above 980 Ti which averages around 37fps at 4k max.

In fact a 1080 ti should be averaging around 65 fps with everything maxed out, and a Titan V should be getting around 85 fps.

 

[sze5003]

Theres something seriously wrong with those numbers for Witcher 3.

They claim 44 fps average with max settings at 4k with 1080 ti -- that is barely above 980 Ti which averages around 37fps at 4k max.

In fact a 1080 ti should be averaging around 65 fps with everything maxed out, and a Titan V should be getting around 85 fps.

Yea I noticed that was off too. I'm pretty sure with Max settings at 4k you can see around 60+ fps on witcher 3 on the 1080ti by looking at last year's benchmark results.

Then they show GTA 5 ultra at 4k getting 120 fps? Seems odd.

Edit : Crap I was just looking at max instead of average which would be the orange bar.