AMD Vega 10, Vega 11, Vega 12 and Vega 20 Rumors and Discussion

[Anarchist4000]

Supposedly there are two Vega chips, but I do agree that three would be necessary. Perhaps things will be better with Navi.

Then there's two Vega chips for cards ranging between $250 and $550. Maybe one RX Vega and a Fury Vega?

Two chips, but multiple potential configurations: 10, 11, (11+11), (10+11), (10+10)? Expectation that 11x2 could fit within a single package and be faster than 10. Leaving open the possibility of a traditional dual product having 4 dice. APUs should cover low to mid tier products. Given financial constraints, skipping low/mid discrete parts that would be replaced by comparable APUs makes sense.

Unlikely, if you recall the Infinity Fabric talks about validating a new design in 3 hours, it's possible they made more designs. Fast enough turnaround time they could have spun a who new product.

 

[fehu]

The Vega 11+11 will be called Vega 110

 

[ImSpartacus]

Yep, but AMD needs a top to bottom solution, not just the upper most performance tier, at 12TF and then the 5,8TF Polaris.
They should try asap to fill this gap with Vega, not with Polaris rehashes.

That's simply not consistent with the reality that AMD must operate under.

AMD doesn't have the budget to pull it off.

After their initial 28nm GCN release in 2012, AMD had released one chip every year. Hawaii in 2013, Tonga in 2014 and Fiji in 2015.

They got wise in 2016 and dedicated resources to two smaller OEM-friendly chips with Polaris 10 and 11, but that was also a process change, so they knew they had to do two in one year.

So far, 2017 looks like Vega 10. We know basically nothing about Vega 11 and Raja only mentioned "RX Vega" in the recent presentation. That sounded like one gpu to me.

Get used to rebrands.

 

[AlNom]

(or replace X with something else)

N

 

[Michellstar]

That's simply not consistent with the reality that AMD must operate under.

AMD doesn't have the budget to pull it off.

After their initial 28nm GCN release in 2012, AMD had released one chip every year. Hawaii in 2013, Tonga in 2014 and Fiji in 2015.

They got wise in 2016 and dedicated resources to two smaller OEM-friendly chips with Polaris 10 and 11, but that was also a process change, so they knew they had to do two in one year.

So far, 2017 looks like Vega 10. We know basically nothing about Vega 11 and Raja only mentioned "RX Vega" in the recent presentation. That sounded like one gpu to me.

Get used to rebrands.

That´s the sad thing for them, because apart from people getting the greatest and fastest, the majority just settles for midrange parts, and you can´t get the latest IP with AMD, and Vega apparently has all the goodies.

 

[Deleted member 13524]

"Vega" in the name is in same position as Fury was, so most likely "RX Nano", "RX Vega" and "RX Vega X" (or replace X with something else)

But Fury didn't use that name as codename. That would be Fiji.

You're saying AMD is going to release cards named "Vega"? Transitioning the codename directly to the final name in the market would be unprecedented. Plus, it would once again break AMD's newly-changed naming scheme.

 

[jacozz]

My guess is that we will see Vega 10 in May-June, and Vega 11 in August-September. Similar to last years Polaris launch.

 

[Kaotik]

But Fury didn't use that name as codename. That would be Fiji.

You're saying AMD is going to release cards named "Vega"? Transitioning the codename directly to the final name in the market would be unprecedented. Plus, it would once again break AMD's newly-changed naming scheme.

Of course I'm saying that, because that's what AMD already said in Capsaicin & Cream, and that's the official branding with logos and all. "Radeon RX Vega". Only difference to Fiji/Fury would be the fact that Vega happened to be codename too.

edit:

Skip to 1:34:40 or so

and straight quote from Raja:
"It's going to be called Vega, it's not a 590, it's not a 490, the gaming GPUs going to be called Radeon RX Vega"

 

[ImSpartacus]

But Fury didn't use that name as codename. That would be Fiji.

You're saying AMD is going to release cards named "Vega"? Transitioning the codename directly to the final name in the market would be unprecedented. Plus, it would once again break AMD's newly-changed naming scheme.

I definitely see where you're coming from, but Koduri seemed pretty clear about "Radeon RX Vega" around 1:36:00 of the video below. The cat is out of the bag.

"It's going to be called Vega. It's not a 590. It's not a 490. The gaming gpu is going to be called Radeon RX Vega."

 

[Deleted member 13524]

Oooh ok.

So I guess they're apparently distancing themselves from the Fury brand. What a shame

 

[ImSpartacus]

Oooh ok.

So I guess they're apparently distancing themselves from the Fury brand. What a shame

The real shame is that he spoke in the "singular" (and, of course, technically only provided one name).

I've got a sneaking suspicion that there's only one Vega 10-based card coming up I the near future.

So no Vega 11 (yet) and no cut down "Vega 10 Pro" card (yet).

Seems weird, but that's the vibe that I'm getting. It might mean that Vega 10 simply isn't profitable at any price below $700 (assuming amd mimics the 1080 Ti's price as a ceiling).

 

[Kaotik]

The real shame is that he spoke in the "singular" (and, of course, technically only provided one name).

I've got a sneaking suspicion that there's only one Vega 10-based card coming up I the near future.

So no Vega 11 (yet) and no cut down "Vega 10 Pro" card (yet).

Seems weird, but that's the vibe that I'm getting. It might mean that Vega 10 simply isn't profitable at any price below $700 (assuming amd mimics the 1080 Ti's price as a ceiling).

I don't know - sure 14nm is a lot more expensive than 28nm was, but it has less memory chips too, and Fiji was sold on profit at $400 (to my understanding it was still profitable there, don't have any actual details, though)

 

[Malo]

Maybe they just decided that

1. There's enough hype build-up around the Vega name, just like with Zen and
2. It's a cool enough name for the marketing department to use

So stay with Vega.

 

[fehu]

And the apu based on Vega and Zen will be the Vegan

 

[Putas]

Polaris improved the fixed function geometry hardware significantly. Unfortunately it was a middle tier product, bottlenecked by raw performance, so the improvements were mostly seen in synthetic benchmarks.

Can you remind me of such benchmarks?

 

[xEx]

Also the HBBC is more important in the mid and low tier GPU where ram is less so I really don't understand why AMD will only want it in the top end part with very little volume so devs wont have much motivation to optimize for the new tech.

 

[3dilettante]

Also the HBBC is more important in the mid and low tier GPU where ram is less so I really don't understand why AMD will only want it in the top end part with very little volume so devs wont have much motivation to optimize for the new tech.

The HBCC's handling of heterogeneous memory seems to be focused on Vega's HPC, big data, and professional markets. This is where there are large and potentially distributed storage pools and maybe even peer to peer communication with other hosts, GPUs, or accelerators.
Reading into the local host's memory sort of falls out of that, and the HBCC is overkill for that purpose--assuming it's all that optimal for accomplishing this result in client hardware.
Per AMD's HPC proposal, this could be done transparently to software, but it takes a big capacity and efficiency hit. Doing it well requires more work and might not be justified for games in general, much less games expected to run on cheap GPUs.

 

[no-X]

Can you remind me of such benchmarks?

https://www.computerbase.de/2016-06/radeon-rx-480-test/11/

 

[ImSpartacus]

I don't know - sure 14nm is a lot more expensive than 28nm was, but it has less memory chips too, and Fiji was sold on profit at $400 (to my understanding it was still profitable there, don't have any actual details, though)

You're probably right about the profitability. I would like to think that there will be two Vega 10-based cards (For AMD's sake if nothing else).

Maybe they just decided that

1. There's enough hype build-up around the Vega name, just like with Zen and
2. It's a cool enough name for the marketing department to use

So stay with Vega.

The problem is next year.

Nvidia ha a strong "Titan" brand that shows up every year. AMD apparently just trashed their "Fury" brand, so will "Vega" be the replacement?

Phrased a different way, will we see Vega-branded Navi-based GPUs in the consumer market?

 

[xEx]

The HBCC's handling of heterogeneous memory seems to be focused on Vega's HPC, big data, and professional markets. This is where there are large and potentially distributed storage pools and maybe even peer to peer communication with other hosts, GPUs, or accelerators.
Reading into the local host's memory sort of falls out of that, and the HBCC is overkill for that purpose--assuming it's all that optimal for accomplishing this result in client hardware.
Per AMD's HPC proposal, this could be done transparently to software, but it takes a big capacity and efficiency hit. Doing it well requires more work and might not be justified for games in general, much less games expected to run on cheap GPUs.

Yes but at the same time AMD showed us how a 2GB with HBC have the performance than 4GB so in mid to low tier where you have less RAM a HBC have bigger impact. Also anyone find funny how AMD show us how 2GB are enough while Nvidia release an 11(!!!!) GB gaming card?