AMD Polaris architecture – GCN 4.0

[FoxEye]

They talk about minimum spec needed for VR being a 290X, then he talks about "We can now produce GPUs capable of running VR at a lower cost, higher volume and lower power."

One can imply the other even thought he did not mention which Polaris he means. it is why i said just conjecture. but i doubt that AMD would make thier bigger polaris chip with just 290X performance. Which then hints towards it being small polaris.

He doesn't say that it's small Polaris tho...

Listen, if I'm a car maker, and I say "we can now make cars that are capable of 0-60 in 3 seconds", does that mean that their 8-seater people carrier varient is going to do 0-60 in 3 seconds? No way, Jose.

As you said, he talks about being capable of producing chips to VR spec. He doesn't say that small Polaris will be capable of VR. That is not the target market for small Polaris.

If you don't agree, fine. I will eat my underpants with a slice of Hovis and post here if you're correct.

 

[Orangey]

Well i stay on this side until there's evidence stating that Samsung will NOT be a manufacturer for AMD, just GloFo alone.

You just lost the argument there by not abiding by the rules of debate.

And they already said they're only using GF for Polaris. I doubt they will issue any more statements on the matter.

 

[drunkenmaster]

You just lost the argument there by not abiding by the rules of debate.

And they already said they're only using GF for Polaris. I doubt they will issue any more statements on the matter.

If the first 3 months production is at Samsung, then everything switches to GloFo, the statement that Polaris WILL be made at GloFo is still true.

One of the keys behind second sourcing and using the same process is that you can move fabs without having to tapeout a new version of the chip. Moving between Samsung/GloFo should be theoretically seamless. However GloFo have started production of 14nm, LPP was supposed to be qualified in Q3 and start production this quarter.

The reality is it doesn't really matter, people say Samsung because it's the same process and the actual fab shouldn't make any difference. It wouldn't make any difference is Apple switched 5k wafers to GloFo and AMD made 5k wafers at Samsung instead. It will probably work out that AMD produce at Glofo, it wouldn't surprise me however if they taped out and got test chips back from Samsung simply because they could do it earlier.

There just isn't a huge need to differentiate between them when they use the exact same process.

 

[Mauller]

He doesn't say that it's small Polaris tho...

As you said, he talks about being capable of producing chips to VR spec. He doesn't say that small Polaris will be capable of VR. That is not the target market for small Polaris.

If you don't agree, fine. I will eat my underpants with a slice of Hovis and post here if you're correct.

I know that he says polaris in general but there are only two polaris chips. And i doubt the larger polaris will have 290x performance.

The polaris efficiency demonstration was also running at 850Mhz. This can also mean that it has a lot of headroom if you consider that the new node could let it run at over 1ghz at stock. Also in terms of the efficiency demonstration we do not know the max performance of the part yet, just that it was capable of matching the 960 at stable 60fps with settings shown.

Add some marmite skid stains to those undies if you eat any

 

[FoxEye]

I know that he says polaris in general but there are only two polaris chips. And i doubt the larger polaris will have 290x performance.

The polaris efficiency demonstration was also running at 850Mhz. This can also mean that it has a lot of headroom if you consider that the new node could let it run at over 1ghz at stock. Also in terms of the efficiency demonstration we do not know the max performance of the part yet, just that it was capable of matching the 960 at stable 60fps with settings shown.

Add some marmite skid stains to those undies if you eat any

But he also doesn't say that either chip will have 290X performance. He says that - the 290X/970 - is the baseline perf needed for VR. He then says that they will be able to produce GPUs that are VR-ready.

That means meeting, or /exceeding/ 290X performance. And that's all it means.

Nowhere does he say that either small Polaris or middle Polaris will have the performance of a 290X. That's entirely your reading of the presentation.

Once again, he says "we will be able to produce chips for VR at lower cost, higher yields, and lower power consumption."

From this you have "read between the lines" and decided that one Polaris chip will have 290X performance. That is, I'm afraid, jumping to conclusions

 

[flopper]

But he also doesn't say that either chip will have 290X performance. He says that - the 290X/970 - is the baseline perf needed for VR. He then says that they will be able to produce GPUs that are VR-ready.

That means meeting, or /exceeding/ 290X performance. And that's all it means.

Nowhere does he say that either small Polaris or middle Polaris will have the performance of a 290X. That's entirely your reading of the presentation.

Once again, he says "we will be able to produce chips for VR at lower cost, higher yields, and lower power consumption."

From this you have "read between the lines" and decided that one Polaris chip will have 290X performance. That is, I'm afraid, jumping to conclusions

Cant be sensible and logical with people here.
Its over their heads.

Polaris brings us a new era of gaming and VR so the future is indeed brighter with AMD Polaris.

 

[AlamoX]

Polaris will have VR ready GPUs(390) under the bar of 349$ thats what the slide says, nothing more, although that might be a hint that polaris will cost around 50$ more for the same perf equivalent

 

[Mei]

Polaris will have VR ready GPUs(390) under the bar of 349$ thats what the slide says, nothing more, although that might be a hint that polaris will cost around 50$ more for the same perf equivalent

if you mean roys talk thats not what he was saying at all
his whole point at the end was cards at the $350 price point, also known as the 290/970 need to get cheaper for VR to make it to the masses (as thats the min spec)

maybe the reality is prices will go up but thats not what he was trying to say for sure lol

 

[Mauller]

Fox, If you continue the quote, he also states 'that means in the second half of this year and forwards, more people will be able to run those headsets.'

But the quote is about Polaris and 14nm. With the part I am talking about hinting at performance and price segment of one of the products. Which is where I go back to it more than likely being the dekstop version of small Polaris rather than medium. Since medium would more than likely exceed/match an overclocked 980ti.

 

[FoxEye]

Fox, If you continue the quote, he also states 'that means in the second half of this year and forwards, more people will be able to run those headsets.'

But the quote is about Polaris and 14nm. With the part I am talking about hinting at performance and price segment of one of the products. Which is where I go back to it more than likely being the dekstop version of small Polaris rather than medium. Since medium would more than likely exceed/match an overclocked 980ti.

You're expecting much too much from small Polaris. The jump between the 380X and the 290X is ~40%.

They won't be able to reduce power consumption by 50% and give a 40% jump in performance at the same time.

That I'm afraid is just impossible.

Small Polaris focus is "console class" gaming in "thin and light" notebooks at low power.

Middle Polaris will be ~980ti perf, maybe a bit higher for the full fat version.

 

[Mauller]

They have said that with improvements in the architecture they have around 2x the performance for equivalent number of previous gen cores. Polaris is a big leap in many regards. So small could have more performance than expected.

Consider half the number of cores as Hawaii but on a new node with fin-fett. They did say 30% power saving from architecture with 70% from process. But the architecture had a performance boost as well. So they must have removed some major bottleneck for a doubling in performance per core.

But if we take them talking about medium Polaris with it having 980ti level performance then they are going to he very cheap in comparison. Essentially flooding the market at a far lower price for the performance.

Not that I am complaining, 290x - 980ti performance at a mainstream price point would be amazing.

 

[FoxEye]

What do you call mainstream pricing? I'm still concerned that expectations are unrealistic here. Off the charts, even.

If AMD don't have totally revolutionary performance here, they need to manage expectations better. Can see a lot of people being upset on launch. Not because Polaris won't be good, but because it's not going to give people 980ti perf for £300.

 

[AlamoX]

What do you call mainstream pricing? I'm still concerned that expectations are unrealistic here. Off the charts, even.

If AMD don't have totally revolutionary performance here, they need to manage expectations better. Can see a lot of people being upset on launch. Not because Polaris won't be good, but because it's not going to give people 980ti perf for £300.

they already did, AMD said 14nm bring more performance/watt, and they are Focusing on efficiency, it's an indirect way to say, we wont be pushing performance up we will be pushing power consumption down

 

[Steampunk]

they already did, AMD said 14nm bring more performance/watt, and they are Focusing on efficiency, it's an indirect way to say, we wont be pushing performance up we will be pushing power consumption down

How many times does it need to be said? More efficient transistors = more transistors in the power budget = more performance. High power efficiency does not equal low power. Even in the low end, low power product, you make a smaller die and use less transistors.

 

[Orangey]

If the first 3 months production is at Samsung, then everything switches to GloFo, the statement that Polaris WILL be made at GloFo is still true.

One of the keys behind second sourcing and using the same process is that you can move fabs without having to tapeout a new version of the chip. Moving between Samsung/GloFo should be theoretically seamless. However GloFo have started production of 14nm, LPP was supposed to be qualified in Q3 and start production this quarter.

The reality is it doesn't really matter, people say Samsung because it's the same process and the actual fab shouldn't make any difference. It wouldn't make any difference is Apple switched 5k wafers to GloFo and AMD made 5k wafers at Samsung instead. It will probably work out that AMD produce at Glofo, it wouldn't surprise me however if they taped out and got test chips back from Samsung simply because they could do it earlier.

There just isn't a huge need to differentiate between them when they use the exact same process.

But the ill-informed think Samsung (being more prestigious) wafers will be magically "better", imbuing Polaris with the power to beat Pascal. They won't admit it now that I've pointed it out, but I enjoy stomping on false dreams nonetheless.

 

[Mauller]

But the ill-informed think Samsung (being more prestigious) wafers will be magically "better", imbuing Polaris with the power to beat Pascal. They won't admit it now that I've pointed it out, but I enjoy stomping on false dreams nonetheless.

As Pascal will be adding FP64 support back again then it will not be as big a jump over maxwell as maxwell was to kepler in terms of gaming performance.

The only reason maxwell had the performance. overclocking and power savings over keplers was due to the drop in fp64 support. Which is why maxwell is only in low end quadro's while the k6000 is still a kepler part.

 

[Mei]

What do you call mainstream pricing? I'm still concerned that expectations are unrealistic here. Off the charts, even.

If AMD don't have totally revolutionary performance here, they need to manage expectations better. Can see a lot of people being upset on launch. Not because Polaris won't be good, but because it's not going to give people 980ti perf for £300.

lol yeh unless all AMD's cards magically disappear including the new furyx2 i cant see them not being priced accordingly, atleast for the first few months
i guess it depends how aggressive AMD wants to be
can they afford to pricecut the current range by lots??

i should think they will be more selling them on new features

 

[FoxEye]

And to the best of our knowledge, since they haven't said anything to the contrary since, they still want to shake their "budget brand" reputation. They want to be a "premium brand", and that does not fill me with the confidence to predict massive price/perf jumps.

If you want 980ti perf, you're still going to have to dig deep into your wallet to pay for it.

 

[CAT-THE-FIFTH]

It looks like Polaris is targetting price points under $350:

http://wccftech.com/amd-polaris-architecture-vr-minimum-spec/

 

[Gregster]

It looks like Polaris is targetting price points under $350:

http://wccftech.com/amd-polaris-architecture-vr-minimum-spec/

Yer, looks like they are releasing a 290X/970 GPU that is at the $349 price point and as that is an AMD slide, I would put some stock into that. It translates to £300, which is an ok price.