AMD Radeon VII Announcement and Discussion

[vipa899]

Regarding it competing with RTX2080, nvidia made it easier this time around (not on purpose of course) directing resources for Ray Tracing instead of further pushing performance on traditional rendering.

Theres the 2080Ti too, its quit abit faster then the 2080. Its a monster in normal rasterized games, AMD competing with the 2080 is a good thing though.
AMD does a much better job on the vram department, 16gb HBM2. Thats what we need for 2020+. The RTX line hasnt really advanced over the 10 series line in the amount of memory they got.

Is this the GPU that many think could be in the PS5?

 

[swaaye]

It's as exciting as when Vega 64 mostly competed with 1080!!!!!! heh. At least they stay in the game.

 

[mrcorbo]

Is this the GPU that many think could be in the PS5?

Nope. That'll be derived from AMD's next GPU architecture, Navi.

 

[Digidi]

There was an email exchange with driver developers concerning why attempts to allocate memory for NGG were failing. It was stated that drivers were not ready to support NGG for GFX9. Vega 20 is a new chip, but it's still GFX9.
It was promised that support would come for a next generation GPU.
Vega 20 is potentially GFX9.something, but keeping the base number the same normally denotes a variation of the same generation.

For me this is the greatest question why? If the silicon or logic was broken they had enough time to test it with Vega 10 and change it in Vega 20. So the hole situation is strange.

 

[vipa899]

Nope. That'll be derived from AMD's next GPU architecture, Navi.

Navi is going to be a mid range gpu?

https://www.google.com/amp/s/www.pcgamesn.com/amd/amd-navi-gpu-release-date-performance?amp

Would think of a showing at CES but no navi there.

 

[mrcorbo]

Navi is going to be a mid range gpu?

https://www.google.com/amp/s/www.pcgamesn.com/amd/amd-navi-gpu-release-date-performance?amp

Would think of a showing at CES but no navi there.

Navi is an architecture. The initial GPUs built using it may be mid-range, but that doesn't necessarily mean the architecture can't also be used to build a high-end (or low-end part). That would depend on whether the architecture has limitations on how it can be configured that prevent it from scaling in one direction or the other.

 

[rcf]

I've just watched a video where someone remembered this:

"AMD RTG Boss Mike Rayfield Retires Amidst Chatter About ‘Disengaged Behavior'"

Rayfield was disengaged and very superficial. He would delete all his emails each week that he could not get around reading and would never work weekends or after 5 PM. He would pontificate in an abrassive manner and couldn’t remember any names [ of his peers].

Rayfield denied to Lisa Su that there was a talent problem and that the talent that had recently exited form AMD wasn’t actually an issue. He diend that there was a talent war going on.

He presented suggestions that werent really feasible such as ‘Radeon VII ‘ which was to be a Vega 20 based consumer facing part that cost $750 to build and would barely tie in with an [NVIDIA] GTX 1080 Ti.

 

[CarstenS]

I've just watched a video where someone remembered this:

"AMD RTG Boss Mike Rayfield Retires Amidst Chatter About ‘Disengaged Behavior'"

Well, maybe AMD has figured out how to make a Radeon VII for less than 750 US-$, now that they are asking 699 US-$ including a game bundle for it, thus making Rayfields supposed idea feasible?

 

[Picao84]

Theres the 2080Ti too, its quit abit faster then the 2080. Its a monster in normal rasterized games, AMD competing with the 2080 is a good thing though.
AMD does a much better job on the vram department, 16gb HBM2. Thats what we need for 2020+. The RTX line hasnt really advanced over the 10 series line in the amount of memory they got.

Is this the GPU that many think could be in the PS5?

No, that would be a Navi GPU.

 

[Picao84]

Well, maybe AMD has figured out how to make a Radeon VII for less than 750 US-$, now that they are asking 699 US-$ including a game bundle for it, thus making Rayfields supposed idea feasible?

Or maybe they ended up surprised by the "Meh" RTX series. They might have expected more from nvidia next gen at the time, making the Vega VII a dud. Or nvidia pushing the prices up helped them sell this part with a profit. Had the Rtx2080 been priced like 499, they could not have launched Vega VII at 699.

 

[Geeforcer]

Well, maybe AMD has figured out how to make a Radeon VII for less than 750 US-$, now that they are asking 699 US-$ including a game bundle for it, thus making Rayfields supposed idea feasible?

Or they were too far along with nothing really to fall back on other than extremely high probability of something that costs “$750 to build” Q4Y18 costing << $750 to build in, say, Q2Y19.

 

[Rootax]

It's a "surprise" announcement, but yeah, 1800 is peak, if it's like Vega, i will not run this high... I like the 16gb move. I feel good with my Vega FE. I can run it @ 1700mhz (watercooled), 1050 for the ram, and it doesn't look very bandwidth limited... If Vega 20 has really 128tmu, then I could believe the gain they talk about, but if not, I call BS.

 

[itsmydamnation]

Or they were too far along with nothing really to fall back on other than extremely high probability of something that costs “$750 to build” Q4Y18 costing << $750 to build in, say, Q2Y19.

or people just make stuff up....

Honnestly, out side of packaging costs/yields whats so fundamentally different for HBM vs GDDR6. A Dram bit is a Dram bit is a Dram bit. It's not like they yield differently. Of course there will be some yield loss on the die stacking but lets be real here, unless your claiming that after ~4 years of hbm products in the market no one can manufacture one without terrible packaging/integration yields.

 

[yuri]

There was an email exchange with driver developers concerning why attempts to allocate memory for NGG were failing. It was stated that drivers were not ready to support NGG for GFX9. Vega 20 is a new chip, but it's still GFX9.
It was promised that support would come for a next generation GPU.
Vega 20 is potentially GFX9.something, but keeping the base number the same normally denotes a variation of the same generation.

The Vega graphic IP belongs to the generation 9 family. The minor ISA/IP changes withing this family are reflected through the additional revision number.

• VEGA 10 GFX9.00
• RAVEN RIDGE GFX9.02
• VEGA 12 GFX9.04
  
• VEGA 20 GFX9.06
• RAVEN RIDGE 2 (PICASSO?) GFX9.09

Navi IP belongs to the generation 10 family. Unconfirmed subrevisions:

• NAVI10 LITE GFX10.00
• NAVI10 GFX10.10

 

[Frontino]

I don't know how old you all are, but I don't remember the last time a partially disabled 3 square cm die was released at 700 bucks.

 

[Picao84]

I don't know how old you all are, but I don't remember the last time a partially disabled 3 square cm die was released at 700 bucks.

I do, mid range Quadro parts

Now seriously, I think you are right and this is a first for the consumer market. Best high price / small die I remember in recent times was the GK104, under 300m2 and priced at 499?

 

[Frontino]

By best you mean highest? Because there have been way better price/die-size ratios in the past.
And may I remind you that GK104 on the GTX 680 was a fully enabled die?

 

[SpaceBeer]

7nm is the most expensive and most complex process so far. And increase is exponential, not liner. I.e. I think this die costs ~$200 or even more. Add complex packaging, 16GB HBM, expensive board (VRM, cooling)... and fact that maybe 0,5-1% gamers are willing to pay $500+ for AMD card, and it seems like ~$700 is more or less the lowest reasonable price.

 

[itsmydamnation]

So you think a 7nm wafer is 20k to 30k?

 

[SpaceBeer]

No. I think it's above 10k, but yields for ~330 mm^2 are less than 50%, probably less than 40%. So if they can get 60 functional chips and have to pay 12k for wafer, it is ~$200