Nvidia Ampere & Turing (next GeForce) rumor thread

[Qwertilot]

They did, as I remember, somehow keep it very quiet in the run up to Pascal releasing.

 

[TheF34RChannel]

Shouldn't we be seeing all sorts of legit looking rumors/leaks at this point if cards are being released late summer or even early fall? I'm starting to think nothing launches until at least late fall.

Yeah it's too quiet for the rumours to be true.

It's one week till computex. Either we get there some rumours going or Turing is coming later. Beside Toms no one had heard of a July launch yet, i don't think we'll see a hardlaunch in July. Maybe a architecture presentation with release date, which will follow in august.

Tom's article sounds like speculation to me rather than them actually having heard something from insiders.

 

[alcoholbob]

There were a couple of Titan launches with even less leaks than the 1100 series...

 

[sontin]

nVidia will talk about their "NVIDIA’s Next Generation Mainstream GPU" at Hotchips: https://www.hotchips.org/program/

So the release will be in the next 2 1/2 months.

 

[TheF34RChannel]

nVidia will talk about their "NVIDIA’s Next Generation Mainstream GPU" at Hotchips: https://www.hotchips.org/program/

So the release will be in the next 2 1/2 months.

Not really something that we know:

- some guy will talk about them a bit (=not a launch)
- Nvidia always launches on their own public events (hot chips is closed)
- the CEO himself always announces cards.

 

[EXCellR8]

Jensen stated that the next GeForce products are still a ways a way.

https://videocardz.com/newz/jensen-huang-next-geforce-is-a-long-time-from-now

What Jensen told everyone was that they will be invited to the launch of new GeForce, but they should not expect it very soon. This confirms what we have been hearing from the industry for the past few months, NVIDIA has not given any details to their partners yet, so there is nothing to talk about yet.

 

[TheF34RChannel]

Jensen stated that the next GeForce products are still a ways a way.

https://videocardz.com/newz/jensen-huang-next-geforce-is-a-long-time-from-now

All vague talk to divert from the topic.

 

[Headfoot]

They could still do a 'pre-conference' announce right before Hot Chips with Jensen and all their normal party throwing hulaballoo, and have some other guy at Hot Chips proper. Not totally unprecedented for them

 

[Qwertilot]

One thing for sure(?) - with AMD doing a 7nm compute chip for the second half this year, NV won't be very far behind.
(If at all.).

That won't reflect consumer stuff of course.

 

[sze5003]

Something that showed up in my news feed. Similar story, they are still not coming out anytime soon, nothing specific though.

https://gizmodo.com/nvidia-ceo-says-new-gpus-wont-be-available-for-a-long-t-1826569249

 

[Bouowmx]

Is the next GeForce being on TSMC 7 nm going to be a real possibility now?

 

[jpiniero]

Is the next GeForce being on TSMC 7 nm going to be a real possibility now?

No. The next Tesla will be though. It is possible that Turing is just 1170/1180 however, just to have something new out there for BTS/Christmas shopping season.

 

[alcoholbob]

No. The next Tesla will be though. It is possible that Turing is just 1170/1180 however, just to have something new out there for BTS/Christmas shopping season.

I'm sure it'll be a while though. 12nm is basically a super refined process that originally started off as 20nm planar with finfets. Trying to build a 800mm2 die on a cutting edge process like 7nm (which iirc is denser than Intel's 10nm) will be likely too sky high right now price wise even for the Tesla line.

I'm sure Vega 7nm will come out in the 4 digits dollar wise as a frontier edition card and it's die size is only somewhere between rx480 and Vega 56/64 size wise.

 

[Qwertilot]

Yes, but I do think the massively competitive, expanding nature of that market kind of forces moonshot attempts.
(V100 pretty much was one after all)

Not very price sensitive either.

 

[reb0rn]

I almost not see logic making 2070/2080 in 12nm when Vega 7nm is around the corner, better have nothing and wait for 7nm to mature!

 

[jpiniero]

I almost not see logic making 2070/2080 in 12nm when Vega 7nm is around the corner, better have nothing and wait for 7nm to mature!

Vega 20 is a HPC focused card though. If you want a 'gaming card' you would need to wait until Navi.

 

[Pinstripe]

Nvidia's 12nm will likely be easily competitive with AMD's 7nm products.

Just like Maxwell 28nm had about the same performance/TDP ratio as Polaris 14nm. Lol.

 

[alcoholbob]

Nvidia's 12nm will likely be easily competitive with AMD's 7nm products.

Just like Maxwell 28nm had about the same performance/TDP ratio as Polaris 14nm. Lol.

980 Ti and Vega 64 have virtually the same perf per watt, and this is with an extra ~15W 980 Ti is burning due to using GDDR instead of HBM.

If anything I suspect 7nm Vega will still be behind 12nm Nvidia in perf per watt, especially if you compare apples to apples (Volta with HBM vs Vega 20)

 

[amenx]

Update from Tom's German site:

https://www.tomshw.de/2018/06/12/nv...der-was-wir-ueber-turing-nicht-wissen-update/

 

[tviceman]

Looks like a late July / early August release. I'm sure it will be blower models first, followed by AIB custom cards 2 months later. I'm anticipating (guessing) that Pascal's FE prices ($699 and $449) will be the across-the-board MSRP prices, but I also think we might see a $50 price hike in the GTX 1170/1070 model at least until AMD can finally provide some sort of competition.

My best guesses.
$699 GTX 1180/2080 for 25-30% faster performance than GTX 1080 TI
$499 GTX 1170/2070 for 5-10% faster performance than GTX 1080 TI
$349 GTX 1160/2060 TI for GTX 1080 performance
$249 GTX 1160/2060 for GTX 1070 performance

If Nvidia rolls out the GT106 chip before AMD has any new cards on the market, Nvidia will have a ~215 mm2 chip as fast as AMD's 512 mm2 and using nearly 1/3 the power. Ouch.

 

[PeterScott]

Looks like a late July / early August release. I'm sure it will be blower models first, followed by AIB custom cards 2 months later. I'm anticipating (guessing) that Pascal's FE prices ($699 and $449) will be the across-the-board MSRP prices, but I also think we might see a $50 price hike in the GTX 1170/1070 model at least until AMD can finally provide some sort of competition.

My best guesses.
$699 GTX 1180/2080 for 25-30% faster performance than GTX 1080 TI
$499 GTX 1170/2070 for 5-10% faster performance than GTX 1080 TI
$349 GTX 1160/2060 TI for GTX 1080 performance
$249 GTX 1160/2060 for GTX 1070 performance

If Nvidia rolls out the GT106 chip before AMD has any new cards on the market, Nvidia will have a ~215 mm2 chip as fast as AMD's 512 mm2 and using nearly 1/3 the power. Ouch.

I think a lot of people are going to be disappointed on pricing/performance.

NVidia, usually has almost equal price/performance to the old cards when they first come out, because the old Cards have dropped significantly in price since introduced, so the new cards are better Price/Performance than the old cards introductory pricing, but not so much compared to current discounted pricing.

But now, old card prices haven't dropped since introduction at all. So I expect across the boards generation pricing will go up this release, so NVidia can capture more revenue/profit at each price point.

 

[Midwayman]

I think a lot of people are going to be disappointed on pricing/performance.

NVidia, usually has almost equal price/performance to the old cards when they first come out, because the old Cards have dropped significantly in price since introduced, so the new cards are better Price/Performance than the old cards introductory pricing, but not so much compared to current discounted pricing.

But now, old card prices haven't dropped since introduction at all. So I expect across the boards generation pricing will go up this release, so NVidia can capture more revenue/profit at each price point.

You're probably right, but man. Fing miners.

 

[jpiniero]

Keep in mind if this is really a 12 nm product, the die size isn't going to be much smaller than GP102.

I still think it's at least $799 for the faster and $549-$599 for the slower. This FE pricing.

 

[alcoholbob]

Keep in mind if this is really a 12 nm product, the die size isn't going to be much smaller than GP102.

I still think it's at least $799 for the faster and $549-$599 for the slower. This FE pricing.

I think most of the gains this gen are going to be async (what we already see with Titan V) and memory bandwidth (20GBps GDDR6 is potentially >80% more memory bandwidth than G5X).

 

[sze5003]

This can't be right. This article is speculating a $999 price for the 1180, not Ti model. Then $1500 for the same model with more vram??

They are basically isolating new and old customers with that price. Pretty sure this isn't correct, it can't be. If it is, what the hell are they going to charge for a Ti model, 2k?

https://www.notebookcheck.net/Nvidi...tput-new-price-details-unveiled.308962.0.html