NVIDIA GeForce 3070/3080/3080 Ti (Ampere): RTX Has No Perf Hit & x80 Ti Card 50% Faster in 4K! (Update 6 ~ Specs / Overview / Details)

[RhysMusique]

Why don't they just call it the RTX 2753 Ti Super X, RTX 2758 Super non-X, and the GTX 1013GT Ti? Their naming scheme seems too complicated being 3070 or whatever? kappakappakappa

[Real_PhillBert]

And I remember when 4GB on a GPU was a lot. Shit, my GTX570 has 1250MBs of VRAM.

[RhysMusique]

1 minute ago, Real_PhillBert said:

And I remember when 4GB on a GPU was a lot. Shit, my GTX570 has 1250MBs of VRAM.

I remember the GTX 970 coming out (doesn't seem 5 minutes ago?!) and there was the whole controversy of it having 3.5GB + 500MB but they had quoted 4GB

[Real_PhillBert]

1 minute ago, RhysMusique said:

I remember the GTX 970 coming out (doesn't seem 5 minutes ago?!) and there was the whole controversy of it having 3.5GB + 500MB but they had quoted 4GB

Yup. That was five years ago now!? Jebus, I'm getting old.

[pas008]

3 hours ago, Taf the Ghost said:

A final "tapeout" is the final submitted silicon for implementation. They'll now get the first test parts back and work through them. Normally 4-5 months to get the first parts back, if everything is working well enough. Then you have to go through the testing process. How long the Tapeout to Launch takes isn't really predictable. If they have to "respin" a couple of times, it adds months.  We also don't know if Nvidia has all of the dies they're going to launch with complete, as some will come later in the process. 

 

In theory we could see a late Spring 2020 release for Nvidia, but, as I mentioned, Samsung is still in the Volume Ramp phase on their node. We also have no idea what a HPC GPU will look like on their node. Those clocks are going to matter a lot.

last yr they said mass production here

 

https://www.anandtech.com/show/13496/samsung-starts-mass-production-of-chips-using-its-7nm-euv-process-tech

 

and considering other tapeouts to release a yr seems normal but I get what you saying

[pas008]

@Taf the Ghost

 

part of me think samsung undercut tsmc for nvidias and licensed with amd is because they might possibly have their own gpu plans

 

lease one ip while making another for low price seems kinda off

 

maybe i'm just reading between too many lines

[Roswell]

On 10/30/2019 at 3:27 PM, Jurrunio said:

let's see if the flagship will break the $2000 price level for the reference model at launch

GPU pricing is so far out of hand it's not even funny. 7 years ago flagships cost $499. Flagship prices are up something stupid like 400% now since the 500 series.

 

Hell, you could get 2 flagships STUCK TOGETHER for $699! (GTX 590)

[tankyx]

Just now, Vitamanic said:

GPU pricing is so far out of hand it's not even funny. 7 years ago flagships cost $499. Flagship prices are up something stupid like 400% now since the 500 series.

 

Hell, you could get 2 flagships STUCK TOGETHER for $699! (GTX 590)

We've seen flagships going for $1000 (inflation adjusted) before. But yeah, it was a fluke before, and now it seems to be a standard

[Jurrunio]

10 minutes ago, Vitamanic said:

Hell, you could get 2 flagships STUCK TOGETHER for $699! (GTX 590)

to be fair old flagships like that can blow up with stock firmware and has poor factory thermals

[emosun]

10 hours ago, Xilefian said:

 

if gta 5 can run on dx9 hardware then idk if the versions really justify whole new numbers yet

I still can't believe a dx9 console with 512mb of uma can run that game. Those people deserve a medal

[The Blue Cryptid]

On 10/30/2019 at 2:31 PM, emosun said:

people are so excited about raytracing and I still don't get why dx10 is better than dx9

There aren't even that many games around with Ray tracing. And sure there "updating" old games with it (or so they say), but I personally feel emmersion from high graphic games with out RT just fine. Despite this.... Imma still probably upgrade my cpu and gpu to this. Unless this launch makes the 2060 cheaper

[Taf the Ghost]

2 hours ago, pas008 said:

@Taf the Ghost

 

part of me think samsung undercut tsmc for nvidias and licensed with amd is because they might possibly have their own gpu plans

 

lease one ip while making another for low price seems kinda off

 

maybe i'm just reading between too many lines

Samsung is up to something. They licensed a huge chunk of AMD IP and paying are for direct support from AMD. RDNA is pretty cool, but Samsung is definitely up to a lot more.

 

The move from Nvidia still seems strange. Nvidia may not want to be on the same nodes as AMD, though that was true in the past before the complete move to GloFo by AMD's GPU division. Most likely answer is that Samsung wanted another high-volume producer for their node, to cover costs, which lowers Nvidia's costs. Or there really is something going on with AMD's patents on the newer nodes and licensing issues.

[dgsddfgdfhgs]

whatever 

wait for the super cards

[pas008]

2 hours ago, Taf the Ghost said:

Samsung is up to something. They licensed a huge chunk of AMD IP and paying are for direct support from AMD. RDNA is pretty cool, but Samsung is definitely up to a lot more.

 

The move from Nvidia still seems strange. Nvidia may not want to be on the same nodes as AMD, though that was true in the past before the complete move to GloFo by AMD's GPU division. Most likely answer is that Samsung wanted another high-volume producer for their node, to cover costs, which lowers Nvidia's costs. Or there really is something going on with AMD's patents on the newer nodes and licensing issues.

Maybe they might do an Intel thing 

Intel had nvidia crosspatent which ran out then licensed amd

Now they releasing gpu lol

 

Maybe Intel sees that gpus are something bigger in the future besides just gaming

And maybe Samsung sees it too

Considering they can do much more shit now days with gpus and horsepower can be upped by adding another card in those areas

 

But I'm prolly reading in between the lines too much

Lol

[Taf the Ghost]

9 hours ago, pas008 said:

Maybe they might do an Intel thing 

Intel had nvidia crosspatent which ran out then licensed amd

Now they releasing gpu lol

 

Maybe Intel sees that gpus are something bigger in the future besides just gaming

And maybe Samsung sees it too

Considering they can do much more shit now days with gpus and horsepower can be upped by adding another card in those areas

 

But I'm prolly reading in between the lines too much

Lol

When Samsung licensed RNDA, it wasn't just for patents. They were pretty clear about that.

 

https://www.amd.com/en/press-releases/2019-06-03-amd-and-samsung-announce-strategic-partnership-ultra-low-power-high

 

They're going to be using AMD designs in their SoCs and, more than likely, working with AMD on designs as well. This is really Samsung keeping up with Apple, as Apple already has a fully developed CPU and GPU infrastructure & IP. (Apple has been poaching people for the last several years, especially from Intel.)  But, still, Samsung is up to something bigger if they're buying designs from AMD.

 

That said, the Nvidia move to Samsung's nodes is strange. Nvidia has been with TSMC since the 90s, so the lack of discussion around the move is strange.

[realpetertdm]

I expect the 3080 to be as overpriced expensive as the 2080 & Ti. There is no reason for Nvidia to cut prices when they domimate the high end market. The 3070 might get price cuts depending on what AMD releases.

 

 

[Harry Voyager]

The pricing of the 3080 TI will depend entirely on the performance of the AMD and Intel competition at that performance tier. nVidia has been able to charge what they are because the GTX 1080 TI and RTX 2080 TI had essentially no competition and were/are the absolute best in class. 

 

If Big Navi rofl stomps the RTX 2080 TI at $650 then, unless the RTX 3080 TI roflstomps the Big Navi, they'll have to bring the price back down to earth. If. 

 

As I understand it, the Samsung selection was simply capacity driven. TSMC is making a fton of chips right now and are massively expanding capacity but still can't keep up with demand for Navi, Zen and all of the other mobile chips. So it's a business calculation that either TSMC is going to have more gum than they can chew this generation, or they're going to have a bubble burst, and neither are spots you want to be jumping in to unless you've got no choice. 

[Rakanoth]

I want cheaper cards.

[BiG StroOnZ]

Small, but significant update to this story (Geforce RTX 3000 Ampere arriving to Data centers in March and to Consumers June 2020):

 

Quote

NVIDIA is slowly, but steadily preparing the launch of their Ampere generation GPUs; allegedly starting March 2020, the first models will be out for data-centers. Geforce 3000-series for desktop PCs would then appear in June. This is reported by the Chinese website HKEPC, citing Chris Caso, an analyst with Raymond James.

 

This means around Computex, in June 2020, the first consumer variants with Ampere chips would see the light of day. The launch would involve the high-end cards first and slower models thereafter. Ampere is said to be manufactured at Samsung on 7nm - using its Extreme Ultraviolet (EUV) method 7nm+. In comparison to Turing, the clock rates could be increased by 200 to 300 MHz. 7nm makes transistors smaller, and that allows for more transistors to be placed on a chip. It is expected that Ampere will offer significantly more Raytracing cores than Turing did. 

 

Source 2: https://www.guru3d.com/news_story/geforce_rtx_3000_ampere_data_center_marchconsumers_june_2020.html

Source 3: https://www.techradar.com/news/nvidia-rtx-3080-graphics-card-could-be-powering-gaming-pcs-in-june-2020

[BiG StroOnZ]

16 minutes ago, VegetableStu said:

wait, datacenters? is it for Geforce Now? also what's to happen to the Titan line? o_o

 

According to the translated article, it seems what they mean, is that perhaps, Tesla parts/products are coming out first (for Data Center). Also, this shouldn't have anything to do with the Titan line.

[tankyx]

30 minutes ago, VegetableStu said:

wait, datacenters? is it for Geforce Now? also what's to happen to the Titan line? o_o

Maybe the SKUs produced are going for the Data centers first. Also, cloud gaming/computing companies Shadow (for sure) and Stadia (Most likely) are going to use RTX3xxx gpus instead of Titans and Quadros for drivers reasons.

[Mira Yurizaki]

2 hours ago, tankyx said:

Maybe the SKUs produced are going for the Data centers first. Also, cloud gaming/computing companies Shadow (for sure) and Stadia (Most likely) are going to use RTX3xxx gpus instead of Titans and Quadros for drivers reasons.

There's almost no difference between consumer and workstation/datacenter GPUs aside from the drivers and possibly better binning on NVIDIA's end for reliability. So it's very likely going to be a Quadro.

[Drak3]

6 minutes ago, Mira Yurizaki said:

There's almost no difference between consumer and workstation/datacenter GPUs aside from the drivers and possibly better binning on NVIDIA's end for reliability. So it's very likely going to be a Quadro.

Don’t Quadros also use ECC?

[Mira Yurizaki]

6 minutes ago, Drak3 said:

Don’t Quadros also use ECC?

That too but you can toggle it off. If it's an application like Stadia, they likely won't have it on because there's no point.

[tankyx]

24 minutes ago, Mira Yurizaki said:

There's almost no difference between consumer and workstation/datacenter GPUs aside from the drivers and possibly better binning on NVIDIA's end for reliability. So it's very likely going to be a Quadro.

The probem is more on drivers support