AMD: Navi Speculation, Rumours and Discussion [2017-2018]

Status
Not open for further replies.
You do realize that the entire "article" is 3rd hand layman's conjecture based on the generous reading of GlobalFoundaries fevered dreams presentation for their stillborn 7nm process, right? The last thing that needed was your ill-informed twist on top.

You do realize we are talking about math..? And using it in the scaling of different lith & uarch..?


If that websites bothers you, you could just google up another webpage that was reporting the same thing. But u like spoon-fed, right...?

https://www.tomshardware.com/news/amd-7nm-gpu-vega-gaming,37228.html
https://hothardware.com/news/amd-7nm-finfet-vega-20-testing
https://www.anandtech.com/show/12910/amd-demos-7nm-vega-radeon-instinct-shipping-2018
https://www.pcgamer.com/amd-has-a-7nm-vega-instinct-graphics-card-running-in-its-lab/
https://www.techradar.com/news/amd-details-next-generation-7nm-radeon-instinct-graphics

etc...

I think you get the ide by now. I hope...
 
  • Like
Reactions: ECH
Yeah, in the land of dreams maybe. The transition of AMD from 28nm to 14nm only yielded 30% more performance and a ton of power consumption increases. You expect 14nm to 7nm transition to give you a 100% uplift over Vega with less power just because it's running on 7nm? What is this 7nm magic matetial? The dreams achiever?

Ironically this 7nm only appears to work in AMD's favor too! When NVIDIA does it they will not be achieving anywhere near the results of Vega 7nm, because 7nm is the dream land of Vega only.

?

You are displaying that you do not understand TSMC 7nm claims, nor what other respected people in semiconductor industry has said about AMD going from 14nm Global to 7nm TSMC. And some has mentioned as high as 100% uplift in games. Which suggest, that AMD is using very conservative uplift figures on their own 7nm Radeon Instinct slides. That performance uplift of 35% is based solely on node uplift for compute. (That is not counting clock uplift, or any refined architecture for gaming, etc.)

Not to mention, and as noted (within many of those links) that 7nm Vega with GDDR6 for gaming, would have about 20% higher clocks, too.


I don't think you have explored fab processes and scaling as much as you would like.
Not sure why you are calling 7nm magic material. Is it because Nvidia is locked out of TSMC's 7nm for until Aug 2019... so it seems mythical to you?
 
  • Like
Reactions: ECH
Which suggest, that AMD is using very conservative uplift figures on their own 7nm Radeon Instinct slides.
Oh god! Not this "AMD is sandbagging" again.
nor what other respected people in semiconductor industry has said about AMD going from 14nm Global to 7nm TSMC. And some has mentioned as high as 100% uplift in games.
Instead of pulling numbers out of thin air, try to exercise some logic to them first, and some data. For a 100% uplift, and without any architectural changes, Vega needs to run upward of 200% increase in clocks (running @3.5GHz!), or they would need to increase shader count to be over the 10k mark. And I believe BOTH of these things will be possible with the new 7nm Vega magic material. 7nm FTW.
Not sure why you are calling 7nm magic material. Is it because Nvidia is locked out of TSMC's 7nm for until Aug 2019... so it seems mythical to you?
NVIDIA is locked out of 7nm FOREVER. It's Vega's dream land only. I though we've established that already.
You are displaying that you do not understand TSMC 7nm claims, nor what other respected people in semiconductor industry has said about AMD going from 14nm Global to 7nm TSMC.
And you are displaying an incredible lack of any grounded reality. All the links you shared only mention gaming chips in the form of Navi. No Vega for gaming whatsoever. Only in your mind.
 
View attachment 2668
Isn’t TSMC 7nm pretty much similar to Intel 10nm and should we be thinking of it more like that ?
Of course. Who didn’t?
It bears remembering that the table above was some planned parameters. Though the final process is unlikely to differ much, it may not be identical to what was originally assumed (or so the rumors say).
 
Saw an interesting video about Navi, this guy expects a 35-45% improvement over Vega 64 and talks about possible improvements to the HBCC. He does unfortunately bring up the possibility of an MCM setup which I think everyone accepts as not happening with Navi, though he goes on to say he thinks Vega 20 might employ this in an Instinct card seeing as David Wang spoke about it not being suitable for "traditional gaming graphics", but being "all in" for professional workloads.

I dunno, he makes some interesting points. Thought it might get discussion going again seeing as how quiet this thread has been.

 
Saw an interesting video about Navi, this guy expects a 35-45% improvement over Vega 64 and talks about possible improvements to the HBCC. He does unfortunately bring up the possibility of an MCM setup which I think everyone accepts as not happening with Navi, though he goes on to say he thinks Vega 20 might employ this in an Instinct card seeing as David Wang spoke about it not being suitable for "traditional gaming graphics", but being "all in" for professional workloads.

I dunno, he makes some interesting points. Thought it might get discussion going again seeing as how quiet this thread has been.

Except that we've already seen Vega 20 and it's single chip with 4 HBM stacks on the interposer
 
Except that we've already seen Vega 20 and it's single chip with 4 HBM stacks on the interposer

Doesn't mean they couldn't come up with cards using more than one Vega 20 chip. Maybe 2x Vega 20 on interposers on top of a single MCM.
We know Vega 20 is equipped with xGMI for multi-GPU interconnect. It remains to be seen if this will be used between cards or(/and?) 2 GPUs in the same card.
 
If 7nm is mature enough that Vega 20 is to be sampled this year, and Navi is mature enough that it's already running, I don't see why it should be released as late as Q3'19.
 
If 7nm is mature enough that Vega 20 is to be sampled this year, and Navi is mature enough that it's already running, I don't see why it should be released as late as Q3'19.

Perhaps TSMC is not set up yet for full scale 7nm manufacturing?
 
Well, it's ready enough to be manufacturing dozens of millions of A12s for Apple. Granted, that probably also means there's competition for capacity—and clearly, AMD won't win that competition against Apple—but it's pretty decent scale!
 
TSMC's 7nm has been in mass production since June, heck, they're already ramping up (in risk production right now) 7nm EUV

Even for "big" chips like this one ?

For AMD, I guess they just don't have the ressources for launching it before. They have Zen2 before that, no ? And I guess a stock of Vega to sell.
 
Vega 20 will be a high end chip selling at a price that can absord bad yields.

Navi being a mid range product means that it's price sensitive. It's probably a chip in the ~200mm range, likely almost 2X the size of cell phone chips, so the yields may be far worse at this time.

Also, AMD maybe using their 7nm allocation for the next Zen chips first.
 
Perhaps TSMC is not set up yet for full scale 7nm manufacturing?
TSMC to tape-out of 100 chips in 2019 on 7nm process
October 22, 2018
TSMC is stating that it is making good steps on the 7 nm production process. It says to have a tape-out of at least fifty new chip designs before the end of the year. 7 nm with euv will reach volume production later in 2020. More than 100 different chips are slated for fabrication in 2019. Apple will get a huge chunk of that however on the list as well are AMD, Nvidia, Huawei, Qualcomm and Xilinx.
https://www.guru3d.com/news-story/tsmc-to-tape-out-of-100-chips-in-2019-on-7nm-process.html
 
New If 7nm is mature enough that Vega 20 is to be sampled this year, and Navi is mature enough that it's already running, I don't see why it should be released as late as Q3'19.
One reason is that 7nm is still not mature enough at this current time, the GPUs at the labs are early versions. Also as others have mentioned, AMD will allocate most of it's 7nm capacity to fighting the battle they can win, the one against Intel, meaning Epyc chips then Ryzen 3000. An AMD official hinted at this part specifically in an interview.
 
Status
Not open for further replies.
Back
Top