AMD 7nm gpus confirmed for 2018

[Daegun]

I'm a little surprised nothing has been posted about this (unless I'm just blind and missed it). 

 

http://ir.amd.com/news-releases/news-release-details/expanding-our-high-performance-leadership-focused-7nm

 

Quote

AMD’s next major milestone is the introduction of our upcoming 7nm product portfolio, including the initial products with our second generation “Zen 2” CPU core and our new “Navi” GPU architecture. We have already taped out multiple 7nm products at TSMC, including our first 7nm GPU planned to launch later this year and our first 7nm server CPU that we plan to launch in 2019.

 

AMD themselves confirm that they've already taped out 7nm products with at least one 7nm GPU launching later this year. Whether or not it's a consumer gpu is still yet to be announced but this is exciting either way. 

 

Personally I don't think this is the best news for nvidia. Not that it's likely to harm them much or take market share, but amd is announcing they're leapfrogging them in node at a time when people are outraged about the prices of the 2000 series.

[GoldenLag]

so more or less confirmation about what we allready knew. sadly nothing especially new is brought to the table

 

one thing im wondering is what AMD is going to do with 14nm and 12nm at GloFo; APUs? Console? Magic?

[yolosnail]

Wasn't this announced a while back?

[Daegun]

4 minutes ago, yolosnail said:

Wasn't this announced a while back?

I looked and all I could find was articles in the last day or so and ones about Navi being on 7nm, nothing about it launching this year. 

[yolosnail]

Just now, Daegun said:

I looked and all I could find was articles in the last day or so and ones about Navi being on 7nm, nothing about it launching this year. 

I'm sure I've read it before but having looked back I can't see it. Unless it was just commented in another topic

[GoldenLag]

7 minutes ago, Daegun said:

I looked and all I could find was articles in the last day or so and ones about Navi being on 7nm, nothing about it launching this year. 

i believe we have been expecting this since march after a roadmap update, at the launch of threadripper it was "confirmed" after they showed off a die i believe. between march and then it was a loose rumour, but got more and more solid through the summer as small details came to light

[Trixanity]

12 minutes ago, Daegun said:

I looked and all I could find was articles in the last day or so and ones about Navi being on 7nm, nothing about it launching this year. 

The 7nm GPU in question is a Vega GPU for machine learning which has been known for many months now (at least since computex officially).

Navi is still slated for 2019. No word on when but probably 2H unless they can somehow pick up the pace.

[cj09beira]

34 minutes ago, GoldenLag said:

so more or less confirmation about what we allready knew. sadly nothing especially new is brought to the table

 

one thing im wondering is what AMD is going to do with 14nm and 12nm at GloFo; APUs? Console? Magic?

they could move to 12nm Fdx

[B1scu1T]

Yeah this is about Vega Instinct if Im not mistaken

[Daegun]

21 minutes ago, GoldenLag said:

i believe we have been expecting this since march after a roadmap update, at the launch of threadripper it was "confirmed" after they showed off a die i believe. between march and then it was a loose rumour, but got more and more solid through the summer as small details came to light

 

16 minutes ago, Trixanity said:

The 7nm GPU in question is a Vega GPU for machine learning which has been known for many months now (at least since computex officially).

Navi is still slated for 2019. No word on when but probably 2H unless they can somehow pick up the pace.

My bad, I didn't realize that. Guess I jumped the gun a little bit here. 

[Coaxialgamer]

It isn't NAVI before any of you get your hopes up . It's their 7nm VEGA GPU for AI and compute .

This has been known for a while

[cj09beira]

Just now, Coaxialgamer said:

It isn't NAVI before any of you get your hopes up . It's their 7nm VEGA GPU for AI and compute .

This has been known for a while

there's also vega 12, you know that gpu we know so little about, (possibly 20 cus, and not much else) no clue if its 7nm though

[Coaxialgamer]

3 minutes ago, cj09beira said:

there's also vega 12, you know that gpu we know so little about, (possibly 20 cus, and not much else) no clue if its 7nm though

Such a small chip would likely use existing 12/14nm , assuming it's a dGPU .

for all we know it could be an igpu for a raven ridge successor.

[GoldenLag]

9 minutes ago, Daegun said:

 

My bad, I didn't realize that. Guess I jumped the gun a little bit here. 

Its still news, just not very exciting news.

 

AMD leaking is louder than their actual statements and releases, so this could habe very well gone un-noticed

[dizmo]

AMD doesn't seem to be that close to Nvidia, so I don't see it as bad news for them at all.

I'd love for them to prove me wrong with new cards.

[cj09beira]

1 minute ago, Coaxialgamer said:

Such a small chip would likely use existing 12/14nm , assuming it's a dGPU .

for all we know it could be an igpu for a raven ridge successor.

from the driver's code its a dgpu 

[GoldenLag]

15 minutes ago, cj09beira said:

they could move to 12nm Fdx

Hopefully they are actually moving to 7nm consumer or else im going to be dissapointed

[cj09beira]

Just now, GoldenLag said:

Hopefully they are actually moving to 7nm consumer or else im going to be dissapointed

for things like apus 12nm fdx might make alot of sense as its efficient and shouldn't be as expensive as 7nm

[GoldenLag]

Just now, cj09beira said:

for things like apus 12nm fdx might make alot of sense as its efficient and shouldn't be as expensive as 7nm

Very true. The 3200G will still be using Glofo. Hopefully its another 100$ value monster

[Coaxialgamer]

Just now, cj09beira said:

for things like apus 12nm fdx might make alot of sense as its efficient and shouldn't be as expensive as 7nm

12nm FDX is based on FD-SOI , not finfet .

Using it would require chips to be completely re-worked for the process. 14nm to 12nm was easy for AMD because Gloflo kept the toolchains and design rules the same, but this would be an entirely different chip compared to the others.

Plus , FDX is more aimed at the IOT and embedded market , there's no real garantee it could reach the performance targets that GF's finfet processes can.

[Taf the Ghost]

On 9/3/2018 at 2:37 PM, GoldenLag said:

so more or less confirmation about what we allready knew. sadly nothing especially new is brought to the table

 

one thing im wondering is what AMD is going to do with 14nm and 12nm at GloFo; APUs? Console? Magic?

The 3rd Gen APUs will be a GloFo 14/12nm. That's been known for a while. There also seems to be several other Semi-custom projects that are expected to crop up. AMD isn't done with 14nm designs for probably another 2-3 years, aside from plans to produce Polaris parts until 2024. (AMD moved their entire embedded platform development to the node, so it's going to be cropping up in all sorts of projects for a long, long time.)

 

On 9/3/2018 at 3:32 PM, GoldenLag said:

Very true. The 3200G will still be using Glofo. Hopefully its another 100$ value monster

"Picasso" is going to be an interesting one. It's Zen2 cores, as those are on 7nm, but that means, unless it's just a minor respin on 12nm improvements, there is are running design improvements from the Zen & Vega teams. Given GloFo's exit from 7nm, it looks like Ryzen 3rd Gen is probably H2 2019, so the next run of the G-series is going to be really interesting to see what's changed. The Zen+ improvements were already there with Raven Ridge, which means if there's anything new, it should be a prelude to improvements with 7nm.

 

However, and this is probably more interesting to me than anyone else, AMD is so deep into the semi-custom game that they clearly still have design teams working on improvements. So we might see some interesting things with Zen+ v2 or something. Nothing big will change, but it'll be interesting to see what iterative improvements can be found.

[Arika]

48 minutes ago, dizmo said:

AMD doesn't seem to be that close to Nvidia, so I don't see it as bad news for them at all.

I'd love for them to prove me wrong with new cards.

They have a lot of catching up to do in GPU performance. But I fear they're dumping more money into their CPU division to stay afloat and their GPUs are just kind of "there" but not their focus

[MMKing]

 

Here it is for all those interested. Some Japanese site calculated the die to be approximately 358-360mm^2 based on the size ratio to the HBM2 chips sitting next to it. Assuming the memory is clocked at least as high as the Vega 64 HBM2 memory is (1.89Gb/s) , we are looking at a bandwidth of 967GB/s minimum.

 

In terms of bandwidth, AMD has a clear lead. But can't really match Nvidias 48GB Quadro RTX 8000, which can further be doubled to 96GB using two cards with NVlink. In terms of performance, there are a few more considerations to take into account. Nvidias card is using a TU102GL core, which has a die size of 754mm^2 on the TSMC 12nm process. The 12nm full correct name is ''12 FinFET compact'', which is based on the 16nm FinFET process. The 12nm process shares Gate Length size (34nm) and Contacted Gate Pitch (90nm) but has a smaller (tighter? Narrower?) Minimum Metal Pitch.

 

Calling out performance numbers at this point will be difficult, but both products will be fairly competitive with one another disregarding the memory size and memory bandwidth differences. Further more, considering how the RTX Quadro was marketed, it seems Nvidia's primary focus is to grab market shares from Intel by convincing the enterprise market to use it's RTX GPUs rather than Intel's CPUs. Which makes sense, since you can't really grow as a company by taking shares from AMD.

 

I seriously doubt we will be seeing any 7nm parts from AMD this year, and at the earliest in the 2nd quarter of 2019 with the release of Zen 2. TSMC is very much occupied with manufacturing the Apple A12 CPU for the Apple Iphone, so the only way AMD can convince TSMC to produce 7nm GPUs is if AMD is charging enterprise prices for them.
 

 

Bellow is the original launch announcement from June, but it's good to get a more recent confirmation of past launch dates.

Spoiler

 

 

[B1scu1T]

Jeez dude.... mark as NSFW before posting nudes

[Guest]

2 hours ago, GoldenLag said:

so more or less confirmation about what we allready knew. sadly nothing especially new is brought to the table

 

one thing im wondering is what AMD is going to do with 14nm and 12nm at GloFo; APUs? Console? Magic?

sell it to Intel