Vega GPU pictured at CES 2017

WYP

News Guru
An AMD Vega GPU has been pictured at CES 2017, confirming that the GPU has two HBM memory stacks and a die size of over 500mm²



Read more on AMD's Vega 10 GPU.
 
Last edited:
That thing is huge! Not the actual die, but the interposer.
I am now 100% confident I was right with 16GB Vega GPU as the top end model
 
I find it really interesting, how both GPU companies are keeping there cards (pun intended) close to there chest. I mean Nvidia haven't spoke about or shown a 1080ti (we all know it's coming), and AMD have been keeping Vega very hush hush too
 
I find it really interesting, how both GPU companies are keeping there cards (pun intended) close to there chest. I mean Nvidia haven't spoke about or shown a 1080ti (we all know it's coming), and AMD have been keeping Vega very hush hush too

Got to wonder if the 1080ti is depending on how good Vega is and Nvidia are waiting to see when it's released what they need to do to have a top GPU
 
Got to wonder what happens if Vega did beat the titan how would Nvidia respond could they make a titanxp ti

Ok b4 the people jump on me it's a wild out there idea and there's more chance you walking down the street and finding the winning lottery ticket 3 weeks in a row
 
Huge die, huge cost, huge chance of failure. All = $.

It was the interposer causing all sorts of issues with Fury.
 
Huge die, huge cost, huge chance of failure. All = $.

It was the interposer causing all sorts of issues with Fury.

No it was the cost of HBM.

Say the same thing about Nvidia
"Huge die, huge cost, huge chance of failure. All = $."
And this is smaller than Nvidias top chips.. Just sayin'
 
Huge die, huge cost, huge chance of failure. All = $.

It was the interposer causing all sorts of issues with Fury.

1. This die isn't as big as the Fury X
2. There are only two HBM2 dies, which is half of what the fury had, simplifying the design and reducing complex component placements on the interposer.
3. AMD have designed stuff on an interposer before and they can do it again.

AMD is literally the only company who are experienced in this type of design. Things are still early days and their current samples are likely no at their final clock speeds etc.

It is far too early to say how things will go with Vega but it is certainly not something that needs to be written off as a failure from the start.

Personally, I am hopeful about Vega, as it seems AMD are finally moving in the right direction architecture wise, focusing on more performance per GPU core and better resource management rather than throwing more GPU cores at the problem.
 
1. This die isn't as big as the Fury X
2. There are only two HBM2 dies, which is half of what the fury had, simplifying the design and reducing complex component placements on the interposer.
3. AMD have designed stuff on an interposer before and they can do it again.

AMD is literally the only company who are experienced in this type of design. Things are still early days and their current samples are likely no at their final clock speeds etc.

It is far too early to say how things will go with Vega but it is certainly not something that needs to be written off as a failure from the start.

Personally, I am hopeful about Vega, as it seems AMD are finally moving in the right direction architecture wise, focusing on more performance per GPU core and better resource management rather than throwing more GPU cores at the problem.


In addition, they designed this architecture to work smarter instead of work harder. It's doing more with what it has, rather than throwing more cores at it's tasks.
I read a very good article(imo at least) on what information is available about Vega. here
Everything here indicates working smarter and the reason why HBM2 is being used and not just being a simple placeholder for GDDR5, becuase that would be useless. GDDR5 is already more than enough, but for things outside of gaming, it is not. As explained in the link.

This makes me pretty excited. I am for sure getting Vega based off everything coming out.
 
Compared to something like the 1070 it's massive. Remember these are cut from circular wafers with bad areas so they are going to cost far more than something like the 1070 to manufacture.

Mark, I was referring to the Fury shortage because they were having trouble making the interposers and it was causing a lot of failures.

Yes it may be smaller than a Titan XP but I bet AMD can only dream of charging £1200 for a GPU. And I bet Nvidia are saving failed dies for their next card too. What can they do with failed Vega dies?

I know what you are getting at, and part of me finds it exciting but I just think it's too much too soon for very little reward in FPS terms. I think they would have been better putting Async on the back burner and concentrating on raw power and clocks like Nvidia did after Fermi.

The 7970 was AMD's Fermi and they have not learned from it it seems.

Oh and FWIW I am a huge fan of AMD. However, when they mess up I find it annoying. More so than when Nvidia mess up because they always get away with it thanks to their mindshare.

When AMD mess up it goes far deeper than upset AMD fans.
 
That thing is huge! Not the actual die, but the interposer.

It is not as big as the Fury X one.

I was given a Fury X one and it is quite a bit bigger.

I have very big hands and fingers but can still get all four finger tips on side of the package.
 
It is not as big as the Fury X one.

I was given a Fury X one and it is quite a bit bigger.

I have very big hands and fingers but can still get all four finger tips on side of the package.

From the naked eye it looks the same size. At least from what i saw on my old Fury X. But it's been so long, I'd take your word for it

Compared to something like the 1070 it's massive. Remember these are cut from circular wafers with bad areas so they are going to cost far more than something like the 1070 to manufacture.

You do realize ts roughly the same size right? It only seems bigger because they are including the HBM2 chips in the measurements

You're over thinking the cost. This whole thing isn't made at once. The die is made, the HBM2 is made on a different line, the interposer as well. They just add them together at the end. It's not mega expensive like you think. More than GDDR5X yes, but HBM2 has been in production for more than 6 months at full volume, it's mature now which means cost is down. This isn't there first time making it either, they now have 2 places to make it, instead of just one previously
 
Last edited:
From the naked eye it looks the same size. At least from what i saw on my old Fury X. But it's been so long, I'd take your word for it

Here is a pic of mine being benched.:D:)

CdHOxFv.jpg
 
Back
Top