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

[PeterScott]

BF5 running at 4k with ray tracing enabled runs at sub 30FPS: https://wccftech.com/ray-traced-battlefield-v-runs-sub-30-4k/

Though tuning could increase this.

Sounds like lots of tuning potential exist, and they are already doing 40-50fps at 1440p, so that will probably get to 60fps in the near future.

FWIW: New RTX BF5 gameplay footage. Harder to spot effects in real gameplay.
https://www.youtube.com/watch?v=Q8V2F3NtCGk

 

[Paratus]

Not if perf/price points are chosen right. Take another look at the only performance figures Nvidia gave us.

Let's say this means RTX 2080 is 50% faster than 1080, excluding DLSS for now. If we normalize for price we need to compare 2080 to 1080Ti though, that means shaving 30% off that performance advantage. That leaves us with a 20% advantage, which can be managed by ensuring prices are lower, but not much lower as to make 2070 unattractive. DLSS and RT will make sure Turing looks more appealing for early adopters while raw performance numbers will likely favor Pascal in the eyes of perf/dollar aficionados.

Win win right? /s

One thing to note here is the performance of the 1080 at 4K generically and with HDR specifically is not great.

• 1080 Mem Bandwidth: 320GB/s
• 2080 Mem Bandwidth: 448GB/s

The 2080 has a 40% increase in bandwidth for the cases where the 1080 is likely bandwidth starved.

So the price / performance at resolutions where the 1080 is not bandwidth starved is not going to be great for 2080 unless the devs make use of DLSS and concurrent shaders. Otherwise the performance difference will be pretty close to the difference in FP16 performance I would think.

 

[Qwertilot]

Is there that much point to anything more than a 1080 beneath 4k resolution?

 

[sze5003]

So, according to Digital Foundry (the youtube video WCCF links to as their source), the RT in BF5 is limited to reflections only. So The rest of the lighting and shadowing is being done "traditionally". Makes sense now why they only focused on the reflections and nothing else during the BF5 demo.

Also, Dice isn't using AI for denoising, they're using their own custom post processing algorithm that doesn't use the Tensor cores.

Yeah they mentioned at one point they can't do full Ray tracing and games must choose one of the rtx options to implement.

 

[Muhammed]

Yeah they mentioned at one point they can't do full Ray tracing and games must choose one of the rtx options to implement.

They also are not using the RT cores properly, they were working with Volta before Turing ..

 

[Hitman928]

They also are not using the RT cores properly, they were working with Volta before Turing ..

I wouldn't say "properly" as the RT cores don't do anything a Volta can't do functionally, the RT cores just do it a lot faster. So I would say they're not using the RT cores as effectively as they could for the best quality/performance result.

 

[PeterScott]

They also are not using the RT cores properly, they were working with Volta before Turing ..

After watching the DF video, I get the impression that they are using the RT cores. They just haven't had time to optimize as they only had the cards 2 weeks before the demo. If they were trying to brute force this without RT cores, performance would be even worse than Volta.

But one thing stood out in the video. They are not using the Tensor cores for denoising. Which is surprising. It seems like it should be more efficient to denoise with Tensors, since that is what they are supposedly there for.

 

[Qwertilot]

Not yet perhaps? If they were aiming at doing it for other cards before this arrived then.....

Very hard to believe that not using them could make objective sense.

 

[sze5003]

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

 

[krumme]

Sounds like lots of tuning potential exist, and they are already doing 40-50fps at 1440p, so that will probably get to 60fps in the near future.

FWIW: New RTX BF5 gameplay footage. Harder to spot effects in real gameplay.
https://www.youtube.com/watch?v=Q8V2F3NtCGk

They can easily turn down ray resolution to get to 60 fps. No one will ever notice it. Using this video as judgement.

I think a part of the problem is as you mention in a game like this what is the actual benefit?

Imo the reflexions looks very very nice in itself but kind of tacked on and out of place. I dont think it adds quality judging from this video but i am looking forward to judge it first hand.

This running up and down a train because there is reflexions is kind of pathetic. Perhaps we will have a map with shiny stuff for the new rtx cards? Lol.

 

[IntelUser2000]

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

This is believable. If we assume RTX 2080 is 10% faster than GTX 1080 Ti. RTX 2080 Ti has 35% more shader firepower and 50% more bandwidth compared to the non-Ti version, comparing FE to FE.

In fact, a RTX 2080 Ti being 35% faster than 1080 Ti would mean the non-Ti RTX 2080 would be something like 0% faster than the 1080 Ti. 45% faster would mean RTX 2080 may be 5-10% faster.

 

[PeterScott]

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

Good Video, it talks a LOT about DLSS. Which leads to my next reply.

 

[PeterScott]

I think enabling DLSS for a game requires training a neural network on all the maps and characters involved in that game. They use one of those ~$100,000 supercomputer-of-video-cards to do the training in a reasonable amount of time. So you can't just "force select" it.

It looks like you were right and I was wrong. It sounds like they train DLSS per game. NVidia will actually work with Devs, and run the game code on their super computer to optimize it specifically for that game.

Talks about training DLSS for one game here:
https://youtu.be/YNnDRtZ_ODM?t=24m4s

So that will definitely slow rollout.
But with supercomputer training for each specific game, it should look really good.

 

[Jaskalas]

Meanwhile, all I want is a GTX 2060. I want a "cheap" card that's a solid upgrade from my own. Maybe then I'll keep a solid 144hz in modern games at 1080p.

 

[DooKey]

Thanks for the video link, sze5003.

 

[Paratus]

It looks like you were right and I was wrong. It sounds like they train DLSS per game. NVidia will actually work with Devs, and run the game code on their super computer to optimize it specifically for that game.

Talks about training DLSS for one game here:
https://youtu.be/YNnDRtZ_ODM?t=24m4s

So that will definitely slow rollout.
But with supercomputer training for each specific game, it should look really good.

NV has to pay devs to use those fancy new features so you’ll know that extra $400 for a 2080Ti is going to good use!

 

[sze5003]

I got bored and started browsing youtube for videos regarding the new cards, that's how I found it.

I'm ok with 35-45% performance increase over my 1080ti. I will still wait on benchmarks though as I want to see how the 2080 Ti does in 4k traditional gaming. This will most likely answer for me if I should go to a 4k monitor or stick with my 1440p gsync.

I know for a fact I'll get great fps at 1440p if i stay with this monitor.

DLSS is interesting for me as it seems it will fix performance issues that msa would cause when cranking that setting up. I wonder what this may mean for VR too if it ever is used there.

I found it funny how the guy in BF 5 shot at his own reflection but to be honest I wouldn't drop down to 1080p just to play with Ray tracing. Specially in BF5 where it's all about speed but it would make the single player environment more realistic.

Edit: Also looks like Acer announced a xb27 4k 144hz gsync monitor that's $1200 instead of $2k like the Asus model. Interested in seeing reviews of that too whenever it will be out. Although I'm not sure what type of hdr support it will have.

 

[gdansk]

Edit: Also looks like Acer announced a xb27 4k 144hz gsync monitor that's $1200 instead of $2k like the Asus model. Interested in seeing reviews of that too whenever it will be out. Although I'm not sure what type of hdr support it will have.

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.

 

[majord]

This is believable. If we assume RTX 2080 is 10% faster than GTX 1080 Ti. RTX 2080 Ti has 35% more shader firepower and 50% more bandwidth compared to the non-Ti version, comparing FE to FE.

In fact, a RTX 2080 Ti being 35% faster than 1080 Ti would mean the non-Ti RTX 2080 would be something like 0% faster than the 1080 Ti. 45% faster would mean RTX 2080 may be 5-10% faster.

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)

 

[IntelUser2000]

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.

Considering the lack of information we have, its all within the range. 35-45% could also mean most end up being close to 35% than 45%, rather than being at the middle for average.

Yes, you are right about the bandwidth. Forgot that its at 352-bit not 384-bit, at same GT/s. I wouldn't call the difference far less though. The GTX Ti's have 10% more relative bandwidth than RTX Ti's. 10% may result in few single digit % difference.

For Pascal, Ti was low as 19% faster(from TPU results) at 1080p, but extended to greater than 35% at 4K.

The potential unknown for Turing is the impact of TDP. The increase is only 16% with the FE versions. However, TDP is sustained worst-case scenario, and with Turing that's with RT and Tensor cores being used. With non RTX use 2080 might have headroom, and Ti's advantage may be greater than TDP alone suggest.

 

[moonbogg]

Don't know if it's been mentioned, but the FE cards were rumored to be overclocked. That would mean Nvidia's slides are showing an overclocked 2080 with HDR enabled (which gimps 10 series cards) vs a stock 1080 with gimped out reference cooling and clock speeds. I wouldn't be surprised to see real benchmarks vs a proper aftermarket 1080 showing as much as a 15-20% difference vs these ultra cherry picked ones. Anything to trick as many people as possible into preordering, eh? That's how I see it.

I'm hoping Nvidia is just price trolling us so they can help their partners get rid of those surplus 10 series cards resulting from their bad crypto betting. The hilarious (milk through nose) prices were supposed to stop people from buying them all up, but silly us 'cause that didn't work.

 

[Hail The Brain Slug]

Don't know if it's been mentioned, but the FE cards were rumored to be overclocked. That would mean Nvidia's slides are showing an overclocked 2080 with HDR enabled (which gimps 10 series cards) vs a stock 1080 with gimped out reference cooling and clock speeds. I wouldn't be surprised to see real benchmarks vs a proper aftermarket 1080 showing as much as a 15-20% difference vs these ultra cherry picked ones. Anything to trick as many people as possible into preordering, eh? That's how I see it.

I'm hoping Nvidia is just price trolling us so they can help their partners get rid of those surplus 10 series cards resulting from their bad crypto betting. The hilarious (milk through nose) prices were supposed to stop people from buying them all up, but silly us 'cause that didn't work.

There's no rumor about the FE overclocking, Nvidia's own specification page lists the FE clockspeeds as being overclocked +90 MHz from reference clocks.

That leaked Time Spy score is telling. 10,030 compared to a (presumably) 1080ti FE stock at ~9,500. Well, I looked up the very first time I ran time spy when I installed my 1080ti FTW3, bone stock right out of the box, and it scored 10,559.

 

[moonbogg]

There's no rumor about the FE overclocking, Nvidia's own specification page lists the FE clockspeeds as being overclocked +90 MHz from reference clocks.

That leaked Time Spy score is telling. 10,030 compared to a (presumably) 1080ti FE stock at ~9,500. Well, I looked up the very first time I ran time spy when I installed my 1080ti FTW3, bone stock right out of the box, and it scored 10,559.

I can imagine the Nvidia guy installing those cards for the benchmarks and saying "whoopsie" as his gum falls out of his mouth and right into the 1080's blower fan.

 

[cmdrdredd]

Don't know if it's been mentioned, but the FE cards were rumored to be overclocked. That would mean Nvidia's slides are showing an overclocked 2080 with HDR enabled (which gimps 10 series cards) vs a stock 1080 with gimped out reference cooling and clock speeds. I wouldn't be surprised to see real benchmarks vs a proper aftermarket 1080 showing as much as a 15-20% difference vs these ultra cherry picked ones. Anything to trick as many people as possible into preordering, eh? That's how I see it.

I'm hoping Nvidia is just price trolling us so they can help their partners get rid of those surplus 10 series cards resulting from their bad crypto betting. The hilarious (milk through nose) prices were supposed to stop people from buying them all up, but silly us 'cause that didn't work.

Curious, how does HDR gimp the GPU?

 

[arandomguy]

Curious, how does HDR gimp the GPU?

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.