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

[Kenmitch]

It was on Reddit. Posted a link a few messages up.

The link you posted was to videocardz

 

[sze5003]

https://www.reddit.com/r/nvidia/comments/99fg7b/nda_ends_in_15_mins/?sort=new

The link was from Reddit.

 

[Malogeek]

https://www.reddit.com/r/nvidia/comments/99fg7b/nda_ends_in_15_mins/?sort=new

The link was from Reddit.

NDA on the slides... yawn. Videocardz is click farming.

 

[LTC8K6]

One would hope, being that 1080 ti is cheaper.

Cheaper and lacking lots of features of the 2070.

 

[Despoiler]

That chart might as well be measuring KekRays. It doesn't declare exactly what is being measured. The 1 line is below the max 1080 performance. /clap Nvidia

 

[sze5003]

Yea I'm not sure how to understand that graph. Not a good comparison at all.

 

[Karnak]

So 40-45% in cherry picked games (and even 4 with HDR where we know that Pascal isn't good at). Reviews will show a ~35% increase IMO, a few titles here and there with more/less.

But besides that we don't even know what is being measured there. Sigh.

 

[Kenmitch]

That chart might as well be measuring KekRays. It doesn't declare exactly what is being measured. The 1 line is below the max 1080 performance. /clap Nvidia

It's better than trying to justify the pricing based on the die sizing....I'll give it that much.

Not sure what to make of it otherwise other than the new cards should be more powerful than the ones they're replacing.

 

[LTC8K6]

Yea I'm not sure how to understand that graph. Not a good comparison at all.

It's a very common graph style. You put the standard at 100% then rank everything else accordingly.

 

[LTC8K6]

So 40-45% in cherry picked games (and even 4 with HDR where we know that Pascal isn't good at). Reviews will show a ~35% increase IMO, a few titles here and there with more/less.

But besides that we don't even know what is being measured there. Sigh.

35% increase for the 2080 over the 1080 in existing games would bode well, imo.
It will likely only get better as drivers improve.

 

[jpiniero]

35% increase for the 2080 over the 1080 in existing games would bode well, imo.
It will likely only get better as drivers improve.

A 1080 Ti is already 30-40% faster at 4K as it is though. Adored is probably right, it's 10-15% faster than the 1080 Ti.

 

[Kenmitch]

It's a very common graph style. You put the standard at 100% then rank everything else accordingly.

It's one of those graph types that just leaves it up to ones imagination in the end. You don't even know what settings were used, what part of the games were cherry picked and used for comparison, if the rigs used were identical, etc.

Waiting for the real reviews before passing judgement is best.

 

[sze5003]

Interesting but we would also want to look at the games and settings. Which means the 2080ti should be about 40% or more in traditional rendering.

35% increase for the 2080 over the 1080 in existing games would bode well, imo.
It will likely only get better as drivers improve.

 

[crisium]

Well that lends credibility to the 1080 Ti +8% rumour. They saw thoseslides before hand maybe?

https://tpucdn.com/reviews/NVIDIA/GeForce_GTX_1080_Ti/images/perfrel_3840_2160.png

108 / 74 = 46%, which seems about the average of the non-DLSS games on that chart. Whether or not we actually get that in non-cherry picked games is less assured. Consider not only the cherry picking, but it's probably FE vs FE and 2080 will throttle less due to open air dual fans.

 

[LTC8K6]

If we are already at a decent performance increase, in a month it will probably be even better, and as drivers mature it should get a lot better.

 

[USER8000]

Just to note something,there is a reason Nvidia used HDR,as apparently the GTX1080 drops over 10% performance when HDR is switched on:

https://www.computerbase.de/2018-07/hdr-benchmarks-amd-radeon-nvidia-geforce/2/

So its drivers or even bandwidth being a problem.

Another consideration - Wolfenstein sees a giant leap. It uses async compute and FP16 right?? So we might see better DX12 and Vulkan titles now onwards.

Pascal is crap at FP16 and HDR,so Nvidia is benchmarking games which will run crap on Pascal.

 

[crisium]

Even if I believe NV, let me ignore DLSS for now since those are few games.

1080 MSRP is $500
1080 Ti MSRP is $700
2080 is $800/$700

Outside of DLSS, 2080 would still be a regression in performance-per-$ based on FE prices (which dictate the AIBs for now). And even based on fairy-tail MSRP, it's barely an improvement.

DLSS is the wild card of course, and ray tracing if it actually is doable at good frames.

PP$ essentially stagnating with a whole new gen is a symptom of a lack of competition. So I'll curb my enthusiasm and wait for reviews (I'm sure I'm getting on somebody's nerves again lol).

 

[IllogicalGlory]

If we are already at a decent performance increase, in a month it will probably be even better, and as drivers mature it should get a lot better.

Sounds like what we were hearing about Vega.

 

[Head1985]

Well that lends credibility to the 1080 Ti +8% rumour. They saw thoseslides before hand maybe?

https://tpucdn.com/reviews/NVIDIA/GeForce_GTX_1080_Ti/images/perfrel_3840_2160.png

108 / 74 = 46%, which seems about the average of the non-DLSS games on that chart. Whether or not we actually get that in non-cherry picked games is less assured. Consider not only the cherry picking, but it's probably FE vs FE and 2080 will throttle less due to open air dual fans.

1080TI is also 66.6% faster than 1070 at 4k and i dont think 2070 will be anywhere near that faster vs 1070.
There is also alot HDR ON games and 1080 sucks at it.So maybe 2080 will not be even 10% faster than 1080TI in more games and with HDR OFF.

Lets say 40% avg perf gain in 20+games with HDR off vs 1080?

 

[antihelten]

It's not Amdahls law. GPU load is essentially 100% parallel.

No they are not. Nothing is ever 100% parallel unless the individual tasks are 100% independent (which they aren't in a rendering pipeline).

Besides, you seem to be of the belief that Amdahl's law only refers to the compute load. It doesn't, it refers to all the loads and bottlenecks involved in delivering the final result, so it also includes stuff like I/O load.

It's the memory bandwidth which I pointed out for the 2080ti comparison, which you neglected here.

While Shader performance could have jumped 80% for GP104, memory bandwidth only went up about 40%, thus limiting performance.

2080 shader and memory bandwidth both increase around 30%, so it wont be bandwidth starved.

Bandwidth being the problem with GP104 relative to GM204 is a nice theory, but it is also quite clearly an incorrect theory. How do we know this? we know this because Anandtech clearly showed that there was little to no performance scaling from increasing bandwidth (1.8% average improvement from 10% overclock).

We can only go by specified boost for clock speed, we don't know how high 2080ti boost in real loads but going by the much improved cooler design, we can bet that will be much higher as well.

We can bet all we want, but then we are no longer talking about baseline estimates (which obviously don't include optimistic betting/guessing like that).

30% is a reasonable baseline.

If you include a bunch of optimistic bets, sure, but then we're not talking about a baseline anymore.

Baseline without including any sort of optimistic assumptions (i.e. a proper baseline) is 15-20%.

 

[PeterScott]

No they are not. Nothing is ever 100% parallel unless the individual tasks are 100% independent (which they aren't in a rendering pipeline).

Besides, you seem to be of the belief that Amdahl's law only refers to the compute load. It doesn't, it refers to all the loads and bottlenecks involved in delivering the final result, so it also includes stuff like I/O load.

Which aren't part of the actual GPU performance and are disingenous to include. The point is if the GPU is x% faster, not how the other bottlenecks in your sytem are holding the GPU back.

If you include a bunch of optimistic bets, sure, but then we're not talking about a baseline anymore.

Baseline without including any sort of optimistic assumptions (i.e. a proper baseline) is 15-20%.

Give the leaked slide above show ~50% gain with DLSS OFF for a 2080 vs 1080, with only 15% more shader units, my 30% baseline estimate is likely on the pessimistic side for 2080Ti vs 1080Ti.

 

[realibrad]

https://www.techradar.com/news/the-...ally-twice-as-powerful-as-the-nvidia-gtx-1080

Twice as fast as the 1080? Its twice in a few areas if you turn on DLSS.

 

[LTC8K6]

Sounds like what we were hearing about Vega.

Vega didn't get better over time?

 

[sze5003]

https://www.techradar.com/news/the-...ally-twice-as-powerful-as-the-nvidia-gtx-1080

Twice as fast as the 1080? Its twice in a few areas if you turn on DLSS.

I hope they show 1080ti vs 2080ti. Sounds promising and I should finally be able to go from 1440p to 4k.

 

[PeterScott]

35% increase for the 2080 over the 1080 in existing games would bode well, imo.
It will likely only get better as drivers improve.

And an easy update will probably be DLSS support. It looks like DLSS must be nearly free to implement. Hope it looks good. Sick of the general blurring from post process AA.