AMD GDC2016 Thread

[Qwertilot]

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.

No, why? This is nothing but good news in terms of medium term competition.

For starters it makes it clear that AMD have a sane, coherent strategy to produce an entire top to bottom line up of 14nm cards, and on a sane time scale.

A real relief. They've been doing all sorts of odd things since 20nm fell through and might have continued.

They absolutely need to do a total top to bottom stack refresh, so they release the bottomish bits of it first, sell them at a bit of premium for the first ~12 months. The power efficiency should allow that.

Then once the rest of the line up rolls out these cards can drop down to form their low/mid end for the 2-3 years until we get the next die shrink. At worst that's enough competition to keep NV plausibly honest.

In all honesty, they should probably get a bit of market share back. Stuff like simply having a plausibly competitive notebook GPU can't help but be useful.

 

[StereoPixel]

Curious about the HBM2 inclusion in Vega on the roadmap and what that means for Polaris? Though he didn’t say it outright, it appears that Polaris will be using HBM1, leaving me to wonder about the memory capacity limitations inherent in that. Has AMD found a way to get past the 4GB barrier? We are trying to figure that out for sure.
Why is Polaris going to use HBM1? Raja pointed towards the extreme cost and expense of building the HBM ecosystem prepping the pipeline for the new memory technology as the culprit and AMD obviously wants to recoup some of that cost with another generation of GPU usage.

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

 

[exar333]

Kind of disappointing. Instead of getting HBM2 first, after getting to market first with HBM1, it almost seems the opposite.

I wonder if NV had a better strategy to avoid HBM1 and go immediately for HBM2?

 

[Elixer]

This is more standard PR style:

https://www.youtube.com/watch?v=Vq3A2OiRfco

https://www.youtube.com/watch?v=H0yrzjEReX8

And it's pretty dang hype-worthy.

Total War: Warhammer pushing a ton of Async Compute was something I wasn't aware of. Thought AMD wouldn't have any $ to sponsor many major games.

Yes, this is much better, and they should have went this way with the live stream event.
These have a very low viewer count as well, why wasn't this part of the press package or part of the live stream?

 

[Bacon1]

https://youtu.be/fqLea0QUW1k Roy Taylor talking Drivers and DX12.

 

[Snarf Snarf]

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

I'm at work and can't watch that video right now, does the fact that Polaris is on HBM actually come out of Raja's mouth or is this an assumption by the interviewer based on the lack of HBM2 in the line up?

 

[jpiniero]

I'm at work and can't watch that video right now, does the fact that Polaris is on HBM actually come out of Raja's mouth or is this an assumption by the interviewer based on the lack of HBM2 in the line up?

AT confirmed that Polaris is using GDDR5.

 

[iiiankiii]

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

HBM1 on Polaris? That's not how I reed it at all from watching the interview. He sounds like he's hinting NO HBM1 OR HBM2 on Polaris. He mention higher cost and difficulty. HBM1 is still pretty difficult to produce. I'm thinking GDDR5; not HBM1.

 

[Snarf Snarf]

Roger, thanks guys. Polaris having HBM1 was contradicting all known information that I had seen prior but I was ready to be impressed if the words came out of Raja's mouth.

 

[Silverforce11]

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.

Raja definitely mention Moore's Laws in perf/w, die sizes and perf/$ and how Polaris was designed to reach the masses.

I'm gonna take a guess at the prices.

Polaris 10:
Top-Tier: $349
2nd-Tier: $299

Polaris 11:
Top-Tier: $199
2nd-Tier: $169

Roy also presented and mention how they want to raise the bar on VR capable GPUs at a lower cost than the 290/970. While these prices would be good, they aren't that awesome a deal considering the current prices of 380/X and 390/X that Polaris 11 and 10 are replacing.

You get similar or better performance, 2.5x perf/w gains, at a slightly reduced price.

 

[Silverforce11]

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.

??

28nm is a very good yielding node. Milking 04 and 00 dies for high prices is a good business move that brought a lot of margins and profit. It was not done out of necessity.

14/16ff will be required as lots of industry signs that it is not good yielding and more suitable for small chips.

Moving forward, 10nm and below, will definite favor smaller chips. So the tech has to shift and hence the focus on multi-die GPU designs, probably with interposers or not, but the API needs to be ready for it when the time comes.

 

[Game_dev]

Kind of disappointing. Instead of getting HBM2 first, after getting to market first with HBM1, it almost seems the opposite.

I wonder if NV had a better strategy to avoid HBM1 and go immediately for HBM2?

There will be some who will try to put a positive spin on even the worst news. IF AMD is going to wait a year or more for a flagship part that is truly dire news for them.

 

[Arachnotronic]

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.

Gosh this is a good post. :thumbsup:

 

[casiofx]

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.

Did you forgot Titan cards? That's why nvidia being called greedy.

Understand that.

 

[Silverforce11]

Did you forgot Titan cards? That's why nvidia being called greedy.

Understand that.

He failed to mention NV did this milking during the mature 28nm node, first with Kepler then with Maxwell.

What Raja was talking about is exactly the opposite. Smaller FF nodes will yield poorly and favor smaller chips. And so smaller chips will be a necessary focus for GPUs, and he talks about design and challenges to make multi-chip possible moving forward.

Instead of selling a huge GPU die, their long term goal is to go with multiple smaller GPUs packaged together. ie. They are still going to sell you a high performance GPU, it's just made up with multiple smaller chips. Nowhere does milking comes into question.

 

[Qwertilot]

Raja definitely mention Moore's Laws in perf/w, die sizes and perf/$ and how Polaris was designed to reach the masses.

I'm gonna take a guess at the prices.

Cut guesses, which were probably very reasonable. Don't forget though, that these cards seem to have a projected lifespan of around 3 years. Premium to start with as 'new', the top AMD cards and so efficient.

They'll then drop quite a bit in price over time, maybe quite fast once Vega etc rolls out.

Come next Christmas and the VR headsets should be rather more mature and these things will be very affordable. The same with NV's equivalents of course.

And none of this has ever really been about milking from anyone. Its just sane behaviour!

 

[USER8000]

LOL,if AMD is using a 232MM2 die and selling it like at £200 to £250 it is no different as the HD7870 which had a similar sized die and sold for about the same price.The HD6870 was a similar size die and sold for around £200.

Considering that AMD is actively saying they are targetting a lower price point with Polaris 10 and 11 over the previous generation,some people are having pretty distorted thinking to make this sound like AMD are trying to do a Titan,Titan Black,Titan Z and Titan X level of milking.

It is most likely Vega 10 and Vega 11 will be hitting the £300 to £500 price points the HD7950, HD7970,R9 290,R9 290X and Fury and Fury X had.

Plus people forget,ATI and AMD might just stick to sell single GPU cards under £300 until Vega 10 and 11 hit - they pretty much did this for the HD3000,HD4000 and most of the HD5000 and HD6000 series.

 

[maddie]

Raja definitely mention Moore's Laws in perf/w, die sizes and perf/$ and how Polaris was designed to reach the masses.

I'm gonna take a guess at the prices.

Polaris 10:
Top-Tier: $349
2nd-Tier: $299

Polaris 11:
Top-Tier: $199
2nd-Tier: $169

Roy also presented and mention how they want to raise the bar on VR capable GPUs at a lower cost than the 290/970. While these prices would be good, they aren't that awesome a deal considering the current prices of 380/X and 390/X that Polaris 11 and 10 are replacing.

You get similar or better performance, 2.5x perf/w gains, at a slightly reduced price.

Are you assuming that Polaris 11 can run VR at the equivalent level of a 390/970 card?

If not, then the prices you have for Polaris 10 is too high for AMD to substantially lower the entry cost barrier for VR, as stated by AMD, which is around $300 now.


Roy Taylor: http://www.gamecrate.com/interview-amds-roy-taylor-dawn-virtual-reality-age/12842

If you look at the minimum spec for either the Oculus or the HTC, and then you look at how many units of the minimum spec have been sold since their launch, so I'm talking about the Radeon 290 or GeForce GTX 970, according to Jon Peddie Research, the total install base of those parts or better is 7.5 million units. So we're going to have to make it possible to run good quality VR at a much lower price. And I'm confident with Polaris we're going to have a big impact to help that.

 

[Silverforce11]

Are you assuming that Polaris 11 can run VR at the equivalent level of a 390/970 card?

If not, then the prices you have for Polaris 10 is too high for AMD to substantially lower the entry cost barrier for VR, as stated by AMD, which is around $300 now.

Polaris 10 from what little info they release, should perform higher than 390X. 390X can't sustain 60 fps in Hitman at 1440p, even in DX12.

So my guess are Polaris 10 #1 $349 giving ~Fury performance, while Polaris 10 #2 $299, giving ~$390X performance.

So they would raise the bar, $299, 290/970 -> 390X performance. Not by much though.

I don't expect them to be priced lower, it would be too generous and they do need $$ badly.

Polaris 11, no way will it match 290/390 levels. ~120mm2 die is just too small for that kind of performance levels.

ps. Ofc I could be very wrong, and AMD suddenly becomes a charity and price 390X beating performance for $249 or something and 390X class performance for $199.. if they do, I would immediate get 2x Polaris 10 for a CF setup.

Edit: For rough performance estimate, you can calculate it on die size: 232mm2 * 2.4x density = equivalent of ~556mm2 GCN on 28nm. Add uarch improvements, there is potential there for ~Fiji performance. So it should land above 390X easily. For Polaris 11, 120mm2 * 2.4x density = equivalent of ~288mm2 GCN on 28nm. Well below Hawaii, which is ~438mm2. With uarch changes, it can approach Tonga, 380X performance. No more than that.

 

[USER8000]

If Polaris 10 equals a Fury,then Polaris 11 which looks to be half the size in die area,is probably going to be around R9 380 level then,but with PCI-E power connector or even bus powered.

However the R9 380 is around £150 to £160,and thee R9 380X around £180 to £200 so it will be interesting to see how they price Polaris 10 and 11.

 

[itsmydamnation]

There will be some who will try to put a positive spin on even the worst news. IF AMD is going to wait a year or more for a flagship part that is truly dire news for them.

its the 16th of the 3rd now, vega's box is more in 2016 then 2017, explain how AMD are waiting a year or more. Now i dont have 12 fingers so i might be getting a little bit confused....

but even last day of 16 is only 9 months away.......

 

[Game_dev]

its the 16th of the 3rd now, vega's box is more in 2016 then 2017, explain how AMD are waiting a year or more. Now i dont have 12 fingers so i might be getting a little bit confused....

but even last day of 16 is only 9 months away.......

Polaris is due in the fall, would Vega arrive a few months later? No.

 

[itsmydamnation]

Polaris is due in the fall, would Vega arrive a few months later? No.

i dont speak american , so i have no idea when the fall is. Polaris is due mid year and its box is dead bang in the middle of the year on the marketing slide.

remembering meeting back to school requires AMD to ship parts to vendors and laptop manufactures months eailer then when we see the product on the shelf.

 

[Hitman928]

Fall = Autumn

Just fyi.

 

[itsmydamnation]

so fall is right now :sneaky:

ps. i'm in Australia it is currently autumn.