AMD GDC2016 Thread

[Silverforce11]

Folks who are expecting big dies moving forward on these new nodes, some really relevant info for you to digest:

https://youtu.be/4qJj1ViyyPY?t=3m32s

This leads to Navi's scalability. Likely be smaller chips on an interposer.

Simply put, new and smaller nodes do not yield well. Intel has had major issues with their current node and delays to 10nm.

AMD is aware of this and is planning for it, I wonder if NV will keep on focusing to make large monolithic dies.

This also relates to performance claims of GP104, how big do you expect it to be? Could next-gen FF really have shifted down, where ~232mm2 is mid-range and ~440mm2 is high-end, rather than aiming for 550-600mm2 dies...

Edit: Article @ http://www.pcper.com/news/Graphics-...past-CrossFire-smaller-GPU-dies-HBM2-and-more

 

[Head1985]

Its looks like new 3870 like age.He talks about performance/dolar many times.Maybe polaris 10 with performance like 390x/GTX980 at 250-300USD?

Also i expecting GP104 will be 300-350mm.

 

[raghu78]

Polaris 10 and 11 seem to be GDDR5 based. The key with Polaris is to target the high volume segment of the market and thats why we see AMD focus on GPUs which are in the 100-250 sq mm range. The other reason is GDDR5 is the only mainstream graphics memory technology available in very high volume and at low cost. Neither HBM nor upcoming technologies such as HBM2 or GDDR5X are ready for mainstream GPUs which range from USD 150 - USD 350. These new memory technologies are not very high volume and the cost is still not suitable for mainstream. Raja said they will introduce HBM2 when its cost makes sense for a wider range of mainstream products. According to anandtech AMD has confirmed Vega in 2017 with atleast 2 GPUs - Vega 10 and Vega 11. I think we will see those occupy the higher performance segments and will be larger GPUs than Polaris 10 and 11. My guess is

Polaris 11 - 110 sq mm, 1024 sp, 128 bit GDDR5
Polaris 10 - 232 sq mm, 2560-3072 sp, 256 bit GDDR5

 

[Azix]

Polaris 10 and 11 seem to be GDDR5 based. The key with Polaris is to target the high volume segment of the market and thats why we see AMD focus on GPUs which are in the 100-250 sq mm range. The other reason is GDDR5 is the only mainstream graphics memory technology available in very high volume and at low cost. Neither HBM nor upcoming technologies such as HBM2 or GDDR5X are ready for mainstream GPUs which range from USD 150 - USD 350. These new memory technologies are not very high volume and the cost is still not suitable for mainstream. Raja said they will introduce HBM2 when its cost makes sense for a wider range of mainstream products. According to anandtech AMD has confirmed Vega in 2017 with atleast 2 GPUs - Vega 10 and Vega 11. I think we will see those occupy the higher performance segments and will be larger GPUs than Polaris 10 and 11. My guess is

Polaris 11 - 110 sq mm, 1024 sp, 128 it GDDR5
Polaris 10 - 232 sq mm, 2560-3072 sp, 256 bit GDDR5

Both hynix and samsung are going to be mass manufacturing them in time for the GPUs to launch. Lot of chips to be sitting down waiting for next year. lower end gddr5 sure, but somethings going to have hbm. Vega might be fully hbm lineup

 

[iiiankiii]

Folks who are expecting big dies moving forward on these new nodes, some really relevant info for you to digest:

https://youtu.be/4qJj1ViyyPY?t=3m32s

This leads to Navi's scalability. Likely be smaller chips on an interposer.

Simply put, new and smaller nodes do not yield well. Intel has had major issues with their current node and delays to 10nm.

AMD is aware of this and is planning for it, I wonder if NV will keep on focusing to make large monolithic dies.

This also relates to performance claims of GP104, how big do you expect it to be? Could next-gen FF really have shifted down, where ~232mm2 is mid-range and ~440mm2 is high-end, rather than aiming for 550-600mm2 dies...

Edit: Article @ http://www.pcper.com/news/Graphics-...past-CrossFire-smaller-GPU-dies-HBM2-and-more

It sounds like he's saying that their goal is to bring VR ready GPUs to the mainstream. To me, that means ~R9 390 performance at a lower price point. If the entry level for VR is the GTX 970 / R9 390, I am expecting the pricing on Polaris 10 to be $249-$299 for it to make sense. That should effectively lower the entry barrier while offering insanely low TDP.

To that extend, I'm predicting
$299 for the "R9 480" ~ R9 390X
$399 for the "R9 480x" ~ R9 390X + 10-15%

 

[Stuka87]

It sounds like he's saying that their goal is to bring VR ready GPUs to the mainstream. To me, that means ~R9 390 performance at a lower price point. If the entry level for VR is the GTX 970 / R9 390, I am expecting the pricing on Polaris 10 to be $249-$299 for it to make sense. That should effectively lower the entry barrier while offering insanely low TDP.

To that extend, I'm predicting
$299 for the "R9 480" ~ R9 390X
$399 for the "R9 480x" ~ R9 390X + 10-15%

I think you mean for Polaris 11 to be in the lower price range. Polaris 10 = big GPU, Poarlis 11 = smaller. And it will be priced lower than that.

 

[AtenRa]

On that note: here's an interview with Raja Kouri on PcPer.

It sounds like he's saying that their goal is to bring VR ready GPUs to the mainstream. To me, that means ~R9 390 performance at a lower price point. If the entry level for VR is the GTX 970 / R9 390, I am expecting the pricing on Polaris 10 to be $249-$299 for it to make sense. That should effectively lower the entry barrier while offering insanely low TDP.

To that extend, I'm predicting
$299 for the "R9 480" ~ R9 390X
$399 for the "R9 480x" ~ R9 390X + 10-15%

And how exactly $399 is mainstream ???

From what Raja said, i would say Polaris Graphics Cards will span in the $100 to $250 market.

 

[iiiankiii]

And how exactly $399 is mainstream ???

From what Raja said, i would say Polaris Graphics Cards will span in the $100 to $250 market.

Yeah, you're right. I think I'm $50 off for the "R9 480". $250 sound right on the money for a mainstream "VR ready" GPU. However, I believe there will be a disproportionately higher cost "X" version. $349-$399 wouldn't shock me. If you look back at the history of these launch, the "X" version was always way more expensive than the non X version.

 

[Head1985]

It will be same as 3870/4870 era.Cheap cards on small SKU.Its not bad have GTX980 performance for 250-300USD.But i think nv with GP104 will destroy it.
Polaris 10 will be comparable with 1060(GP106) and not with 1070/1080(gp104)

 

[raghu78]

And how exactly $399 is mainstream ???

From what Raja said, i would say Polaris Graphics Cards will span in the $100 to $250 market.

Roy's statement at VRLA (17:50 - 18:20) on Polaris GPUs running faster than R9 290X, consuming lower power, costing lesser and available in higher volume combined with Raja's statements on making VR reach a wider audience mean I think the Polaris 10 will have multiple SKUs in the USD 249- USD 349 price range. My guess is the slowest and cheapest Polaris 10 SKU will be faster than R9 290X.

https://www.youtube.com/watch?v=p010lp5uLQA
http://www.pcper.com/news/Graphics-...past-CrossFire-smaller-GPU-dies-HBM2-and-more

My thoughts go back to HD 7870 which came out at USD 349 and the HD 7850 at USD 249. The HD 7870 was quite close to GTX 580. I think thats possible again with the top Polaris 10 SKU getting close to 980 Ti/Titan X.

 

[Head1985]

My thoughts go back to HD 7870 which came out at USD 349 and the HD 7850 at USD 249. The HD 7870 was quite close to GTX 580. I think thats possible again with the top Polaris 10 SKU getting close to 980 Ti/Titan X.

Pretty much zero chance that 232mm2 SKU will be close to 980TI/TITANX.
Best case scenario is GTX980 performance and way cheaper.

 

[Mondozei]

Simply put, new and smaller nodes do not yield well. Intel has had major issues with their current node and delays to 10nm.

AMD is aware of this and is planning for it, I wonder if NV will keep on focusing to make large monolithic dies.

Why should this surprise you or anyone else? The slowdown of Moore's law has been blindingly obvious for the last 5 years.

Yet you and RS and others on this forum kept bashing NV for "greed" for their X04/X00 strategy, now we see AMD doing the exact same thing with their later release of Vega in 2017.

It would be helpful if people stopped living in the past and started to finally absorb the reality. Now that AMD is doing it, it becomes clear that for many forum members, the reality is suddenly visible. One would hope that looking at the GPU world didn't require one's favourite GPU vendor to do a move in order to see it, but alas, this is where many are.

 

[Mondozei]

Roy's statement at VRLA (17:50 - 18:20) on Polaris GPUs running faster than R9 290X, consuming lower power, costing lesser and available in higher volume combined with Raja's statements on making VR reach a wider audience mean I think the Polaris 10 will have multiple SKUs in the USD 249- USD 349 price range. My guess is the slowest and cheapest Polaris 10 SKU will be faster than R9 290X.

https://www.youtube.com/watch?v=p010lp5uLQA
http://www.pcper.com/news/Graphics-...past-CrossFire-smaller-GPU-dies-HBM2-and-more

My thoughts go back to HD 7870 which came out at USD 349 and the HD 7850 at USD 249. The HD 7870 was quite close to GTX 580. I think thats possible again with the top Polaris 10 SKU getting close to 980 Ti/Titan X.

Those guesses sound reasonable, except the last part, which is probably a tad optimistic. Polaris will be a mainstream play through and through.

The reality is that a Titan X-level performance is way more than necessary for the vast majority of gamers and it would severely undercut Vega 10/Greenland in 2017, which AMD must hope to sell to more people than just for VR(which always needs more power). GPU companies tweak their releases all the time, cut their SKUs in all kinds of ways. Having a top Polaris GPU at about 349 performing like the Fury non-X would be powerful and in line with historical trends. It could even be priced lower to perhaps 299. Vega will likely come out fast in 2017 by looking at their chart, look for Q1, so no rush just yet to undercut the upcoming release that fast.

 

[AtenRa]

Sorry but $300 is not mainstream.

And we alreay have R9 390 and GTX 970 at $300 price as a bare minimum for VR. If they want to make VR mainstream they have to go at or bellow $200, that is what mainstream starts to have a meaning.

 

[raghu78]

Those guesses sound reasonable, except the last part, which is probably a tad optimistic. Polaris will be a mainstream play through and through.

The reality is that a Titan X-level performance is way more than necessary for the vast majority of gamers and it would severely undercut Vega 10/Greenland in 2017, which AMD must hope to sell to more people than just for VR(which always needs more power). GPU companies tweak their releases all the time, cut their SKUs in all kinds of ways. Having a top Polaris GPU at about 349 performing like the Fury non-X would be powerful and in line with historical trends. It could even be priced lower to perhaps 299. Vega will likely come out fast in 2017 by looking at their chart, look for Q1, so no rush just yet to undercut the upcoming release that fast.

Was the GTX 660 (GK106) mainstream. you bet it was. Did it get close to GTX 580 (GF110). you bet it did. btw did GTX 660 undercut GTX 680. No it did not.

I am not saying it will happen. I am saying its possible. AMD's Polaris talks about 2.5x perf/watt wrt 28nm. Fiji was 275w. 275/2.5 = 110w. Incidentally 596 sq mm / 2.5 = 238 sq mm. Not far away from the rumoured 232 sq mm die size we have been hearing about. Anyway its just 3 months before we get to know how good Polaris will be.

http://www.anandtech.com/show/10145/amd-unveils-gpu-architecture-roadmap-after-polaris-comes-vega

"Vega is currently scheduled to come relatively quickly after Polaris. Depending on how literal you interpret this chart, the far left edge of the Vega box does fall into 2016, though obviously AMD intends to leave themselves some wiggle room here and not tie themselves down to specific dates. The fact that Vega comes this soon after Polaris is interesting; it seems hard to believe that it’s a direct successor to Polaris – I can’t see AMD replacing Polaris parts in less than a year – so this points to Vega being more of a cousin, and is where AMD’s naming system isn’t especially helpful in deciphering anything further.

With Polaris confirmed to use GDDR5, Vega is notable for being the first AMD architecture to use HBM2, and the first parts in general to use HBM tech since Fiji. I’m presuming these are higher-end GPUs to complement the Polaris GPUs (the smaller of which we know to be a low-power laptop design), which is where HBM would be more cost-effective, at least at current prices.

Meanwhile AMD has also confirmed the number of GPUs in the Vega stack and their names. We’ll be seeing a Vega 10 and a Vega 11. This follows Polaris GPU naming – which has finally been confirmed – with Polaris 10 and Polaris 11. I have also been told that Polaris 11 is the smaller of the Polaris GPUs, so at this point it’s reasonable to assume the same for Vega."


I agree with anandtech's reasoning that Vega 10 and 11 complement Polaris 10 and 11 instead of replacing Polaris GPUs. Vega will address higher performance segments with larger GPUs and HBM2.

My guess is

Polaris 11 - 110 sq mm GPU (128 bit GDDR5)
Polaris 10 - 230 sq mm GPU (256 bit GDDR5)
Vega 11 - 350 sq mm GPU ( HBM2)
Vega 10 - 450 sq mm GPU (HBM2 and Firepro flagship)

 

[el etro]

To that extend, I'm predicting
$299 for the "R9 480" ~ R9 390X
$399 for the "R9 480x" ~ R9 390X + 10-15%

Is what i think too.

 

[Game_dev]

So no high end chip from AMD this year? No one will wait for Vega.

 

[S.H.O.D.A.N.]

Yet you and RS and others on this forum kept bashing NV for "greed" for their X04/X00 strategy

No one is bashing NVidia for releasing mid range chips first. It's the premium prices for mid range products that get people going.

So no high end chip from AMD this year? No one will wait for Vega.

If Polaris 10 can push 60fps on 1440 in Hitman, at 390x price point, then yeah, no one's going to be waiting for Vega. I'm sure AMD will be devastated because of that.

 

[S.H.O.D.A.N.]

double post

 

[Azix]

great, wont feel like upgrading till 2017. GDDR5 can find a corner. Not buying another GPU using that.

 

[Snarf Snarf]

I think everyone was being a little too optimistic, hoping to get a true flagship this early on a new node. As much as I want a single card for 144Hz 1440p AAA gaming my wallet appreciates that I'll have another 8-9 months with my 980ti. All the 290/390/980ti owners here can ride this out and wait to see how the big boys perform in DX12 when there are more titles out.

That being said I think Polaris 10 is exactly what AMD needed, 1440p performance at $250 would put a serious dent in that 80/20 market share assuming they get on shelves before GP104 launches.

*edit* I can't spell

 

[Mopetar]

No one is bashing NVidia for releasing mid range chips first.

I think to some degree it's because unless Polaris has some substantial architectural improvements and NV only releases something that amounts to a die shrink that the mid-range NV part is going to have the performance advantage.

Even if you're an AMD fan you'd want them to release something more powerful.

 

[Bacon1]

Also think about the push to multi-gpu.

Lower end cards are better price/perf, plus its easier entry for people.

If you want more perf? Just add a new card. With DX12 / Vulkan offering much better multi-gpu support, and Stardock coming out with some DX12 abstraction layer or something (don't think details have released) but they want to make it dead simple to get multi-gpu support in DX12 games for everyone.

That means you can have 2 power efficient cards running at 90% + scaling instead of a single power hungry monster card.

I mean look right now, you can get CFX 390 for $550 (275 x 2) or a Fury X for $600. (Both XFX @ newegg right now).

Multi gpu push makes a ton of sense, both for consumers and them once game support is there, which DX12 is making much better.

Less products to create = less R&D = better prices, higher yields.

The amount of people who buy the top end cards is very small.

Buying 2-3 mid range instead makes more sense once they scale 90%+ in games.

Also cross-vendor multi-gpu support which means they have to keep competing even after selling their first batches, because you could add their competitor card for more perf instead of buying a second from them.

 

[MrTeal]

I think to some degree it's because unless Polaris has some substantial architectural improvements and NV only releases something that amounts to a die shrink that the mid-range NV part is going to have the performance advantage.

Even if you're an AMD fan you'd want them to release something more powerful.

Depends on what nVidia releases. So far we have essentially one verified fact about Pascal which is that the die used in the Drive PX2 is slated to have <= 4TFLOP/s at whatever clocks it runs at. That's less the the 980 and about 2/3rds of a 390X. The Drive PX2 also has a TDP of 250W, so one GPU is likely <100W.

Everything else is speculation and rumour. If AMD releases 110mm² and 232mm² dies this year, I sure hope nVidia beats them in performance, but I'll be waiting to see on launch day.

Maybe launch day + a week or two, just to make sure 8GB of VRAM isn't really 7GB + 1GB equivalent of wood screws.

 

[3DVagabond]

There is no way GP 104 is slower than TITANX.
If Polaris 10 is only 232mm2 then GP104 will be way faster.At 300-320mm2 it will crush TITANX very easy.
Polaris 10(if its only 232mm2) alias 7870 will be competing VS 390x/GTX980 and later vs 1060.

Vega most likely is only 1 tier above polaris10 so it will compete vs GP104 and not 500mm2 BIG pascal.

We're not sure of Polaris 10's size. Although the 232mm^2 does have evidence to suport it. We have no idea about Vega. None!