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

[sze5003]

....And some would call you crazy for spending that kind of money on a watch. lol

Yea I know, I don't even bother to tell my siblings or parents about the stuff I buy. Everyone would think it's crazy. But I can hand those down to my kids one day or sell them for literally almost the same amount I bought them. Very little money to be lost.

My buddy just texted me: "just ordered two 2080ti's."

I'd get something new so he would have to as well. He's nuts for buying 2 of them when yesterday he was telling me he won't buy one unless he sees it for $999 and if it can do 60fps at 4k with rtx on.

This is why Nvidia can charge what they want lol

 

[PeterScott]

@PeterScott, looks like we're not the only ones confused about DLSS versus super-resolution. https://www.reddit.com/r/nvidia/comments/99492x/techradars_ndalimited_review_2080ti/


So DLSS technically isn't "AI Up-Res", but Nvidia is working on "AI Up-Res".

AI upres is something that has been around for a while, just like I said in my previous post and even linked a site where you can test it.

It's not magic though. It mainly just improves some upscaling artifacts.

DL itself has AMAZING potential. If this was wide open to third party developers on consumer cards, I would be much more excited about that, than getting Ray Tracing in Games.

But, given history, I expect DL access will be limited to NVidia in house stuff for Deep Learning applications, to keep selling the lucrative Deep Learning Pro cards.

 

[Kenmitch]

My buddy just texted me: "just ordered two 2080ti's."

I'd get something new so he would have to as well. He's nuts for buying 2 of them when yesterday he was telling me he won't buy one unless he sees it for $999 and if it can do 60fps at 4k with rtx on.

He didn't lie to you in the end. He didn't buy one he bought two....His way of saving face.

 

[crisium]

Peter, you're overestimating the chip size effect. You're shocked it's only $1,200? You honestly think Nvidia is barely selling this at a profit? The same Nvidia that has been increasing margins every year?

The record margins (above Intel even when Intel competed against Bulldozer) were posted in this thread already. I know you read it.

Yes it is the biggest chip and it costs more than any other GeForce. But you forget about what true competition can do. Nvidia did not sell Fermi at a loss. Yet they offered a cutdown 520mm2 GPU for $289 MSRP. That's $324 today.

If a 520mm2 GPU could sell at $324 in 2018, then a 754mm can make profit far below $1200. Undoubtedly. You cover the R&D costs with workstation and Quadro cards, if necessary. But it is not neccsary, since we have 0 competition and thus Nvidia can gouge consumers as well as professionals.

You slipped and fell down Nvidia's gradual pricing slope.

2011: 520mm2 $290 (560 Ti Core 448)
2013: 561mm2 $500 (780 price cut)
2015: 601mm2 $650 (980 Ti)
2017: 471mm2 $700 (1080 Ti)
2018: 754mm2 $1200

If you only look at 2017 and 2018 you'd be tricked into thinking it's reasonable given the die size increase. Then you take a memory trip back to when AMD actually competed, and you realize the absurdity of thinking $1200 doesn't give them crazy margins. Especially when these margins increase every single year.

 

[LTC8K6]

Peter, you're overestimating the chip size effect. You're shocked it's only $1,200? You honestly think Nvidia is barely selling this at a profit? The same Nvidia that has been increasing margins every year?

The record margins (above Intel even when Intel competed against Bulldozer) were posted in this thread already. I know you read it.

Yes it is the biggest chip and it costs more than any other GeForce. But you forget about what true competition can do. Nvidia did not sell Fermi at a loss. Yet they offered a cutdown 520mm2 GPU for $289 MSRP. That's $324 today.

If a 520mm2 GPU could sell at $324 in 2018, then a 754mm can make profit far below $1200. Undoubtedly. You cover the R&D costs with workstation and Quadro cards, if necessary. But it is not neccsary, since we have 0 competition and thus Nvidia can gouge consumers as well as professionals.

You slipped and fell down Nvidia's gradual pricing slope.

2011: 520mm2 $290 (560 Ti Core 448)
2013: 561mm2 $500 (780 price cut)
2015: 601mm2 $650 (980 Ti)
2017: 471mm2 $700 (1080 Ti)
2018: 754mm2 $1200

If you only look at 2017 and 2018 you'd be tricked into thinking it's reasonable given the die size increase. Then you take a memory trip back to when AMD actually competed, and you realize the absurdity of thinking $1200 doesn't give them crazy margins. Especially when these margins increase every single year.

Cost to NV of producing a die in 2011 vs 2018?

 

[PeterScott]

Peter, you're overestimating the chip size effect. You're shocked it's only $1,200? You honestly think Nvidia is barely selling this at a profit? The same Nvidia that has been increasing margins every year?

The record margins (above Intel even when Intel competed against Bulldozer) were posted in this thread already. I know you read it.

Yes it is the biggest chip and it costs more than any other GeForce. But you forget about what true competition can do. Nvidia did not sell Fermi at a loss. Yet they offered a cutdown 520mm2 GPU for $289 MSRP. That's $324 today.

If a 520mm2 GPU could sell at $324 in 2018, then a 754mm can make profit far below $1200. Undoubtedly. You cover the R&D costs with workstation and Quadro cards, if necessary. But it is not neccsary, since we have 0 competition and thus Nvidia can gouge consumers as well as professionals.

You slipped and fell down Nvidia's gradual pricing slope.

2011: 520mm2 $290 (560 Ti Core 448)
2013: 561mm2 $500 (780 price cut)
2015: 601mm2 $650 (980 Ti)
2017: 471mm2 $700 (1080 Ti)
2018: 754mm2 $1200

If you only look at 2017 and 2018 you'd be tricked into thinking it's reasonable given the die size increase. Then you take a memory trip back to when AMD actually competed, and you realize the absurdity of thinking $1200 doesn't give them crazy margins. Especially when these margins increase every single year.

I guess they tricked Ryan Smith and Industry analysts as well.

If you follow silicon economics, you also know that cost/area of die has been INCREASING over time. So not only are big dies more expensive, they have been getting more expensive over time.

The size:cost relationship isn't linear either. A 754/471 60% size increase is going to get more than 60% cost increase.

754mm2 is enormous, ridiculously expensive die. It surpising it's in a consumer product at all, let alone selling for the same price as the Titan X with 471mm2 die.

 

[n0x1ous]

https://videocardz.com/newz/nvidia-publishes-geforce-rtx-2080-vs-geforce-gtx-1080-comparison

 

[LTC8K6]

https://videocardz.com/newz/nvidia-publishes-geforce-rtx-2080-vs-geforce-gtx-1080-comparison

Looks pretty good when you consider that's not even the ti...

 

[Head1985]

https://videocardz.com/newz/nvidia-publishes-geforce-rtx-2080-vs-geforce-gtx-1080-comparison

So 20-30% IPC gain.Because 2080 have only 15% more SP and same clocks but performance is 40-50% higher.
So 2080 is 10-20% faster than 1080TI
2070 will be 5-10% slower than 1080TI

 

[Despoiler]

Maybe some levity vs the competition and current Nvidia products. This makes me think Nvidia's claims are much like how they did(overdid) tesselation to create an advantage. They could do it a smart way, but nope it's got to be the hardest most intense way to create a gap.

https://forum.beyond3d.com/posts/2040568/

There's been a lot of chatter on Twitter from devs about Nvidia's 10Gray/s in the past few days. With some of them wondering how/when/where was this achieved given that there are a shit ton of ways to "Ray Trace" and WTH are those RT Cores all about exactly. Sebbi just blessed us with the following:

https://twitter.com/SebAaltonen/status/1032283494670577664

Claybook ray-traces at 4.88 Gigarays/s on AMD Vega 64. Primary RT pass. Shadow rays are slightly slower. 1 GB volumetric scene. 4K runs at 60 fps. Runs even faster on my Titan X. And with temporal upsampling even mid tier cards render 4K at almost native quality.

 

[sze5003]

He didn't lie to you in the end. He didn't buy one he bought two....His way of saving face.

Nah he didn't order anything his wife is a doctor and would kill him, dispose of him and I would never hear from him. I on the there hand am not married so yea I could order 2 but I don't blow money like that and I'd never want to do sli.

 

[LTC8K6]

So 20-30% IPC gain.Because 2080 have only 15% more SP and same clocks but performance is 40-50% higher.
So 2080 is 10-20% faster than 1080TI
2070 will be 5-10% slower than 1080TI

2070 might keep up with the 1080ti.

 

[jpiniero]

So 20-30% IPC gain.Because 2080 have only 15% more SP and same clocks but performance is 40-50% higher.
So 2080 is 10-20% faster than 1080TI
2070 will be 5% slower.

More likely clocks than IPC. Clocks are hard to determine with GPU Boost 3/4 involved.

 

[LTC8K6]

More likely clocks than IPC. Clocks are hard to determine with GPU Boost 3/4 involved.

I doubt clocks are that much different. GPU Boost 4 might give a little advantage.

But either way, faster is faster.

 

[crisium]

You have no thoughts on profit margins increasing every year?

It seems clear to me that the increases in price per area more than make up for increased die size costs.

I'll present the data in another way. edit: price per area

2011: $0.55
2013: $0.89
2015: $1.08 edit: fixed
2017: $1.48 record margins this year and this is smaller than all of the other chips edit: fixed
2018: $1.59

Yes it's not a linear relationship, but that's why I presented all the years in my earlier post. 2018 only looks reasonable compared to 2017. Do you disagree? 601mm2 $650 profit, and 3 years later 754mm2 $1200 less profit?

Do you believe NV profits for Q3 and Q4 2018 will decrease?

 

[Head1985]

2070 might keep up with the 1080ti with the mem bw advantage.

The gap between 2070 and 2080 will be around 20%.2080 still have 28% more SP.At best in some games 2070 will be at 1080Ti level, but slower average.

 

[LTC8K6]

The gap between 2070 and 2080 will be around 20%.2080 still have 28% more SP.At best in some games 2070 will be at 1080Ti level, but slower average.

There's actually no mem BW advantage, I was wrong about that.

Still we likely have the 2070 right at the 1080ti doorstep.

 

[Malogeek]

The gap between 2070 and 2080 will be around 20%.2080 still have 28% more SP.At best in some games 2070 will be at 1080Ti level, but slower average.

By the time 2070 is released, pricing will be around the same for both cards, the 1080ti maybe a bit cheaper. Which means I'd rather have a 2070 with the extra functionality, especially since I don't game at 4k.

The big question is whether the cut down RT cores on the 2070 are good enough to actually enable RT in games.

 

[sze5003]

Apparently nda ends in 15 minutes?

https://videocardz.com/newz/nvidia-publishes-geforce-rtx-2080-vs-geforce-gtx-1080-comparison

 

[crisium]

Apparently nda ends in 15 minutes?

https://videocardz.com/newz/nvidia-publishes-geforce-rtx-2080-vs-geforce-gtx-1080-comparison

So they are claiming 50% increase without DLSS and 100% with DLSS from 1080 to 2080?

Ok things got more interesting. A lot more.

 

[zinfamous]

2070 might keep up with the 1080ti.

One would hope, being that 1080 ti is cheaper.

 

[Malogeek]

Apparently nda ends in 15 minutes?

Where are you getting that from?

 

[Head1985]

So they are claiming 50% increase without DLSS and 100% with DLSS from 1080 to 2080?

Ok things got more interesting. A lot more.

Those IPC gains are crazy it must be even more gain than maxwell vs kepler or it boost to 2.4Ghz

 

[alcoholbob]

More likely clocks than IPC. Clocks are hard to determine with GPU Boost 3/4 involved.

15% more cores, but 40% more memory bandwidth. So you are going to naturally get about ~25% more performance. But I believe this will start topping out with the higher SKUs because the memory bandwidth on a per-core basis starts shrinking.

We also get about ~10% improved IPC in async/vulkan/DX12 carried over from Volta.

I don't think you actually need any new IPC increases from volta to turing to get these kind of numbers, especially when you choose specific titles to bench against (Titan V was 65% faster than 1080 Ti in certain titles, clockspeed normalized). I'm sure in older DX11 titles you won't see the same perf lift.

 

[sze5003]

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

Where are you getting that from?