AMD Vega 10, Vega 20, and Vega 11 GPUs Detailed

[btarunr]

AMD CTO, speaking at an investors event organized by Deutsche Bank, recently announced that the company's next-generation "Vega" GPUs, its first high-end parts in close to two years, will be launched in the first half of 2017. AMD is said to have made significant performance/Watt refinements with Vega, over its current "Polaris" architecture. VideoCardz posted probable specs of three parts based on the architecture.

AMD will begin the "Vega" architecture lineup with the Vega 10, an upper-performance segment part designed to disrupt NVIDIA's high-end lineup, with a performance positioning somewhere between the GP104 and GP102. This chip is expected to be endowed with 4,096 stream processors, with up to 24 TFLOP/s 16-bit (half-precision) floating point performance. It will feature 8-16 GB of HBM2 memory with up to 512 GB/s memory bandwidth. AMD is looking at typical board power (TBP) ratings around 225W.



Next up, is "Vega 20." This is one part we've never heard of today, and it's likely scheduled for much later. "Vega 20" is a die-shrink of Vega 10 to the 7 nm GF9 process being developed by GlobalFoundries. It will feature 4,096 stream processors, too, but likely at higher clocks, up to 32 GB of HBM2 memory running full-cylinders at 1 TB/s, PCI-Express gen 4.0 bus support, and a typical board power of 150W.

The "Vega 11" part is a mid-range chip designed to replace "Polaris 10" from the product-stack, and offer slightly higher performance at vastly better performance/Watt. AMD is expecting to roll out the "Navi" architecture some time in 2019, and so AMD will hold out for the next two years with "Vega." There's even talk of a dual-GPU "Vega" product featuring a pair of Vega 10 ASICs.

View at TechPowerUp Main Site

 

[AlienIsGOD]

already announcing a RX 480/470 replacement? lame Vega 11 should be faster hopefully

 

[the54thvoid]

Did AMD say they're positioning between GP104 and GP102? If so, that's bad news. Given the time frame of 2nd half 2017, not first quarter, it's almost definitely a Summer release for 2017. That gives Nvidia a huge leeway for current pricing and Volta development.
Also, of tremendous importance is that this is an investors conference so they need to say all the absolute best things.
I have a bad feeling about Vega. Even if it's better than GTX1080, AMD are saying it won't beat Titan X?
Sad face.

 

[FordGT90Concept]

If that picture is accurate, it's going to have a lot more than 4096 stream processors. That's how many the 28nm Fury X had and these chips appear to have the same die space (perhaps even more because the HBM chips appear smaller).

Is Global Foundries really that close to 7nm? How can GloFo be making such rapid improvements when Intel appears to be stuck? If this is accurate (I'd gander that Vega 20 is smoke and mirrors), GloFo could pass up Intel in process tech and that's quite unfathomable.

AMD not having an answer to Pascal for another three quarters is dire news. Vega 10 is no doubt beyond Pascal's reach but by the time it launches, it will have to contend with Volta.

If AMD's goal is to Nano (huge chip, low power) their entire product line up, that eats directly into AMD's profit margins. This news post is making me...

 

[psyph3r]

already announcing a RX 480/470 replacement? lame Vega 11 should be faster hopefully

Not at all. Vega is the next part of the market. 480 is budget low end. Vega is enthusiast high end.

 

[$ReaPeR$]

interesting developments on the AMD side of things. it seems that all available resources are thrown towards the release of zen, this will not bode well for the GPU market.

 

[hardcore_gamer]

If this is true, an affordable 4K card will continue to be just a dream for the next couple of years.

 

[qubit]

I'd love to see this trading blows with Volta, but I'm not holding my breath.

AMD really has to prove itself to us now.

 

[hardcore_gamer]

I'd love to see this trading blows with Volta, but I'm not holding my breath.

AMD really has to prove itself to us now.

Looking at the specs, Vega 10 will have similar performance of a GTX 1080.

Nvidia can easily charge $800 for their next mid range GV 104.

 

[Vayra86]

If that picture is accurate, it's going to have a lot more than 4096 stream processors. That's how many the 28nm Fury X had and these chips appear to have the same die space (perhaps even more because the HBM chips appear smaller).

Is Global Foundries really that close to 7nm? How can GloFo be making such rapid improvements when Intel appears to be stuck? If this is accurate (I'd gander that Vega 20 is smoke and mirrors), GloFo could pass up Intel in process tech and that's quite unfathomable.

AMD not having an answer to Pascal for another three quarters is dire news. Vega 10 is no doubt beyond Pascal's reach but by the time it launches, it will have to contend with Volta.

If AMD's goal is to Nano (huge chip, low power) their entire product line up, that eats directly into AMD's profit margins. This news post is making me...

Intel is already expected to lose their process advantage. Intel is taking steps back from being the x86 leader 'no matter what' and its clear for all to see as they have moved away from tick/tock and do three steps between a shrink with Kaby Lake showing that third step to be pretty much no improvement whatsoever. They already scaled down several business units, sold some acquisitions (McAfee for ex) and are definitely a little bit worried even though they will never admit they are. Look at Intel right now, it has its dominance, but only on parts of the market that are stagnant or out of the consumers' eyesight (server, HPC). They know they can't live on those markets alone and they've been making moves to counter that, but so far their GPUs are still nothing special and way too expensive for what they offer, their IoT stuff is being overwhelmed by ARM offerings, and those smaller systems like NUC... well, I doubt that'll ever be more than a niche.

I think the market may very well turn around in AMD's favor. AMD is positioned a LOT better at the moment and they are making efforts to get into the picture for new markets, or already are in it, for example: consoles (again, PS4Pro and Scorpio), APIs, and custom SOCs. AMD is also becoming much more 'lean' than Intel and the funny thing is that AMD is actually ahead of Intel with regards to streamlining the company.

However in terms of process nodes, we know that the '7nm' and FD-SOI processes are not a 'real' node shrink in the original sense of the word. Intel still currently has the smallest 'real' node even on 14nm.

About Vega 10, positioning between GP104 and GP102 seems like a very smart move because by Q2 2017 the GP104 and GP102 will have aged abit. They will probably push that GPU as the bang/buck high end offering and do another HD7950 with it, probably leaning heavily on overclockability to push it towards GP102. With current pricing, the GP104 and GP102 won't be selling like hotcakes anyway and the only way they will sell bigger numbers is through price drops. AMD only needs to position their GPU just a little bit better and they will offer the HBM2 product versus the GDDR5 product that no one was really waiting for, but at a similar price point and with the added advantage of being 9 months newer than its competitors (this matters, look at all those who sold a 980ti to get an almost similar and situationally even weaker 1070).

Look at how Fury X excels on certain games in newer APIs and you can see how HBM2 on an even wider GPU will absolutely be king of the hill, making positioning between 104 and 102 more of a worst case scenario than anything else. In addition, all that matters for AMD is that they push large amounts of products with a little bit of margin, not some Titan XP equivalent for the halo effect that will be out of the picture for 99% of the potential market.

To compete, you don't need the top end product, you just need a product people want to buy.

 

[Basard]

interesting developments on the AMD side of things. it seems that all available resources are thrown towards the release of zen, this will not bode well for the GPU market.

Bodes just fine for a lot of us though. There are plenty of GPUs to choose from, just not so many CPUs.

 

[AlienIsGOD]

The "Vega 11" part is a mid-range chip designed to replace "Polaris 10" from the product-stack, and offer slightly higher performance at vastly better performance/Watt. AMD is expecting to roll out the "Navi" architecture some time in 2019, and so AMD will hold out for the next two years with "Vega." There's even talk of a dual-GPU "Vega" product featuring a pair of Vega 10 ASICs.

Not at all. Vega is the next part of the market. 480 is budget low end. Vega is enthusiast high end.

Polaris 10 is RX 480/470, possibly read the article all the way through

 

[64K]

I was thinking the high end Vega 10 would come in close to as fast as the Pascal Titan X. It will be somewhat disappointing if it doesn't but we won't know until reviews from legitimate sites like this one tell the full story. Looks like a long wait for that though.

 

[Caring1]

What's with reducing the FP first from double to single, now half?
"with up to 24 TFLOP/s 16-bit (half-precision) floating point performance"
Obviously reducing the compute side increases gaming usability, as proven by Nvidia cards doing the same.

 

[Xzibit]

Did AMD say they're positioning between GP104 and GP102? If so, that's bad news. Given the time frame of 2nd half 2017, not first quarter, it's almost definitely a Summer release for 2017. That gives Nvidia a huge leeway for current pricing and Volta development.
Also, of tremendous importance is that this is an investors conference so they need to say all the absolute best things.
I have a bad feeling about Vega. Even if it's better than GTX1080, AMD are saying it won't beat Titan X?
Sad face.

The source for the article

Vega 10 will be released in first quarter of 2017, it has 64 Compute Units and 24TF 16-bit computing power. Vega 10 is based on14nm GFX9 architecture. It comes with 16GB of HBM2 memory with a bandwidth of 512 Gbps. The TBP is currently expected at around 225W. Meanwhile, dual Vega 10 will be released in second quarter of 2017 and TBP should be around 300W.

 

[Kaotik]

AMD hasn't announced "Polaris 10 replacement" or anything similar, it's all just VideoCardz speculation of how the chips will be. Earlier rumors actually suggest that Vega 11 is bigger than Vega 10, and as we know from what Raja said earlier, the numbering used since Polaris only tells you which chip was designed first, not their performance.

If that picture is accurate, it's going to have a lot more than 4096 stream processors. That's how many the 28nm Fury X had and these chips appear to have the same die space (perhaps even more because the HBM chips appear smaller).

The chip in the image is Fiji used in Fury X. The HBM1 chips on it are actually smaller than HBM2-chips used by GP100 (and in future Vega)

 

[dyonoctis]

Ow. So Vega will actually feature mid-range gpu with hbm. I wasn't expecting that, and didn't expect a replacement of Polaris 10 so fast. Captain tom was almost right.

 

[FordGT90Concept]

...their IoT stuff is being overwhelmed by ARM offerings, and those smaller systems like NUC... well, I doubt that'll ever be more than a niche.

As more and more people move from cable and satellite TV to IPTV/internet streaming, those products are in great demand and that demand is growing. The thing is, the profit margins on them are so small, Intel would rather use their fabs to build Core I# and Xeon processors that they can sell for a hefty mark up. NUCs basically get the left overs (older fabs) because Intel really doesn't care. The fact of the matter is that market will never have huge profit margins--it's always destined to be a volume seller. This is the same reason why Intel doesn't care about smartphones.

I think the market may very well turn around in AMD's favor. AMD is positioned a LOT better at the moment and they are making efforts to get into the picture for new markets, or already are in it, for example: consoles (again, PS4Pro and Scorpio), APIs, and custom SOCs. AMD is also becoming much more 'lean' than Intel and the funny thing is that AMD is actually ahead of Intel with regards to streamlining the company.

Consoles (they are custom SOCs so addressing both here...) are in the same boat as NUCs, tablets, and smartphones: volume products with tight profit margins. AMD may dominate the console market but NVIDIA dominates the desktop market. Games are developed on Windows which overwhelmingly run on NVIDIA hardware. Developers may be quite familiar with the ins and outs of GCN because of optimization for consoles but they're optimized on NVIDIA first because that's what they're coding on.

AMD is open-sourcing virtually all of its APIs. That's great for Linux but that isn't going to translate to profits for AMD. Well it could because AMD is more appealing now on Linux but realize we're talking about a minority of minority of systems here.

What's with reducing the FP first from double to single, now half?
"with up to 24 TFLOP/s 16-bit (half-precision) floating point performance"
Obviously reducing the compute side increases gaming usability, as proven by Nvidia cards doing the same.

It's a DirectX 12 thing. Some things, especially textures, don't require 32-bits of precision. At 16-bit, it should be able to process two calculations for the price of one.

Developers aren't using 16-bit because hardware support is iffy. Five years from now, 16-bits to handle textures will likely become common place.

 

[the54thvoid]

So Vega is 1st quarter (thanks @Xzibit). That's far better. Launch of Zen and Vega should light a fire in the stagnant brush of PC tech.

 

[medi01]

Is Global Foundries really that close to 7nm? How can GloFo be making such rapid improvements when Intel appears to be stuck?

My version:

1) By "skipping 10nm to save time" (lol)
2) "Collaboration with IBM" magic (lol)
“We are well positioned to deliver a differentiated 7nm FinFET technology by tapping our years of experience manufacturing high-performance chips, the talent and know-how of our former IBM Microelectronics colleagues and the world-class R&D pipeline from our research alliance. No other foundry can match this legacy of manufacturing high-performance chips.”
3) Cause VCDZ made it up (welp)
However:
“The technology is expected to be ready for customer product design starts in the second half of 2017, with ramp to risk production in early 2018”.
http://techfrag.com/2016/09/16/7nm-finfet-chip-production/

Developers may be quite familiar with the ins and outs of GCN because of optimization for consoles but they're optimized on NVIDIA first because that's what they're coding on.

Other explanations can exist.
Consoles are rather weaksauce so it makes sense/you are forced to optimized for them.
As for PC, oh well, dudes can buy another 1.5$k card, who fecks.

So Vega is 1st quarter (thanks @ Xzibit). That's far better. Launch of Zen and Vega should light a fire in the stagnant brush of PC tech.

Amen.

And we need AMD to get back into game.

 

[qubit]

Did AMD say they're positioning between GP104 and GP102? If so, that's bad news. Given the time frame of 2nd half 2017, not first quarter, it's almost definitely a Summer release for 2017. That gives Nvidia a huge leeway for current pricing and Volta development.
Also, of tremendous importance is that this is an investors conference so they need to say all the absolute best things.
I have a bad feeling about Vega. Even if it's better than GTX1080, AMD are saying it won't beat Titan X?
Sad face.

Looks like they're once again gonna be forced into playing the "value card" against NVIDIA. At least NVIDIA are still releasing decent cards, even if they're competing with themselves mostly and we have to pay more for less.

We can only hope that AMD pull a rabbit out of the hat with DX12 performance, as we've seen one or two benchmarks show a significant performance increase. Somehow I suspect that NVIDIA is gonna neutralize that advantage though.

 

[Captain_Tom]

Usually we use single precision to calculate gaming compute right?

If so I am reading 512GB/s and 12 TFLOPs. That would put it at or slightly above the Titan.

 

[m1dg3t]

What? No wood screws?

 

[Captain_Tom]

Looks like they're once again gonna be forced into playing the "value card" against NVIDIA. At least NVIDIA are still releasing decent cards, even if they're competing with themselves mostly and we have to pay more for less.

We can only hope that AMD pull a rabbit out of the hat with DX12 performance, as we've seen one or two benchmarks show a significant performance increase. Somehow I suspect that NVIDIA is gonna neutralize that advantage though.

The only way Nvidia can neutralize that advantage is if AMD continues to fail to capture marketshare. Early reports show AMD capturing marketshare with polaris, but imo they will need to capture at least 40% for them to stop worrying about Nvidia's numbers advantage.


Otherwise note that this will have 12 TFLOPS vs 11 of the Titan X, and 512 - 1 TB/s of bandwidth beats the 480 of the Titan X. Even if it doesn't soundly beat the Titan X, it will easily trade blows if these specs are true.

P.S. Where is this 512 GB/s coming from? HBM2 comes in 720 GB/s and 1 TB/s flavors so I am calling BS on that spec.

 

[bug]

Did AMD say they're positioning between GP104 and GP102? If so, that's bad news. Given the time frame of 2nd half 2017, not first quarter, it's almost definitely a Summer release for 2017. That gives Nvidia a huge leeway for current pricing and Volta development.
Also, of tremendous importance is that this is an investors conference so they need to say all the absolute best things.
I have a bad feeling about Vega. Even if it's better than GTX1080, AMD are saying it won't beat Titan X?
Sad face.

Even worse, it's a 225W part. GP102 is 250W, but that's without HBM and Vega 10 is supposed to be "between GP104 and GP102". Meaning weaker than GP102.
AMD is skating to where the puck is, not where it will be

Also, first half of 2017 can easily turn into a July or back-to-school launch. Oh well, I wasn't planning on buying any of these anyway.