AMD Polaris architecture – GCN 4.0

[CAT-THE-FIFTH]

Don't know because AMD are doing their whole "wall of silence" thing.

Because if there's a sure-fire way to get people excited about your product, it's to... say nothing (?) and let your competition release first and still... say nothing (?) So all we have are rumours and speculation, and mostly those aren't positive, making us quite pessimistic about AMD's new cards. AMD watch all this and try to keep our hopes alive by... saying nothing. So now we all mostly are just waiting to buy up those 1070s whilst AMD keep silent and absolutely on pain of death don't do anything that could be remotely considered good PR.

AMD logic!

The problem it means either a Fury X moment,so are delaying to try and make the card more competitive, or an HD4870 one where they are trying to see what Nvidia are doing and give no chance for Nvidia to react. ATI basically made zero noise about the HD4870 until they launched it two weeks after the GTX280. Then you had websites like Anandtech saying the following when the HD4870 launched:

For now, the Radeon HD 4870 and 4850 are both solid values and cards we would absolutely recommend to readers looking for hardware at the $200 and $300 price points. The fact of the matter is that by NVIDIA's standards, the 4870 should be priced at $400 and the 4850 should be around $250. You can either look at it as AMD giving you a bargain or NVIDIA charging too much, either way it's healthy competition in the graphics industry once again (after far too long of a hiatus).

To sum it up,basically the silence is a bad thing or a very good thing. If AMD can get close to a GTX1070 at a lower price it would be good,but we will soon find out how fast the GTX1070 and GTX1080 are,as for all we know the GTX1070 might be a bridge too far for Polaris 10 to match.

 

[pmc25]

Either way is a good chance Vega is coming late this year?

Very unlikely, in my view. It would be in dying days of '16, and even if NVIDIA squeeze out another full paper or semi-paper launch for an HBM2 product in December, it won't be worth AMD's while to follow suit. They're far better off hard launching at CES '17 (assuming adequate HBM2 supply), or 'showcasing' it at CES assuming poor HBM2 supply and launching some time later in Q1.

 

[AlamoX]

I don't get your logic. A cash injection doesn't make you surprised at the competition releasing a mid-end part, nor does it let you suddenly shift release schedules on a whim. The biggest stumbling blocks for Vegas and GP100 is HBM2 availability and 16nm yields, money doesn't help drastically here. the process will mature, engineering will improve yields and HBM2 is ready when it ready.

it wouldn't change process shedule, or part availability like HBM2, but engineering, drivers , logistic, etc, doesn't that matter at all ?
besides can a GPU vendor postpone the release of a GPU by a quarter, because memory availability, especialy when the chip itself is hybrid that can accommodate gddr also ? a quarter is still a significant drop in value of high tech products like a GPU.
and i could be wrong im just blabling, if my logic is far off, it's probably because im wrong , maybe last year they didnt wanna spend money on a gddr5x variant of vega, and now they can afford it ?

 

[Lokken86]

The 4870 was probably about as far behind the gtx280 as the fury x was to the 980ti but it was priced very well such made it a hit.

 

[pmc25]

To sum it up,basically the silence is a bad thing or a very good thing. If AMD can get close to a GTX1070 at a lower price it would be good,but we will soon find out how fast the GTX1070 and GTX1080 are,as for all we know the GTX1070 might be a bridge too far for Polaris 10 to match.

I don't think a full die P10 will have any difficulty matching or beating a 1070 in most games. I think the latter will struggle to get near a 980Ti much of the time ... and I think both GP104 cards will be really memory bandwidth gated. In titles that saturate memory bw quickly, they may perform very poorly.

 

[Gerard]

The problem it means either a Fury X moment,so are delaying to try and make the card more competitive,

Since when was fury x delayed to make it more competitive? Was more the hbm yields that held it back iirc.

 

[pmc25]

Since when was fury x delayed to make it more competitive? Was more the hbm yields that held it back iirc.

Interposer production ramping and supply more than anything.

 

[CAT-THE-FIFTH]

Since when was fury x delayed to make it more competitive? Was more the hbm yields that held it back iirc.

No,I am saying either Polaris 10 is a bit underwhelming like Fiji,so they are being silent about it,and trying to tweak it as much as possible,or either they know they are onto a winner,and trying not to give away anything to Nvidia.

 

[CAT-THE-FIFTH]

I don't think a full die P10 will have any difficulty matching or beating a 1070 in most games. I think the latter will struggle to get near a 980Ti much of the time ... and I think both GP104 cards will be really memory bandwidth gated. In titles that saturate memory bw quickly, they may perform very poorly.

It will be interesting to see how things panned out. Looking at the Nvidia slides for the GTX1080,it seems to be around 25% faster on average for a 37% increase in TFLOPS,so makes me wonder if memory bandwidth is a limiting factor. The GTX980TI seemed to scale quite well in performance when compared to the GTX980 when you looked at the TFLOPS rating.

 

[D.P.]

it wouldn't change process shedule, or part availability like HBM2, but engineering, drivers , logistic, etc, doesn't that matter at all ?
besides can a GPU vendor postpone the release of a GPU by a quarter, because memory availability, especialy when the chip itself is hybrid that can accommodate gddr also ? a quarter is still a significant drop in value of high tech products like a GPU.
and i could be wrong im just blabling, if my logic is far off, it's probably because im wrong , maybe last year they didnt wanna spend money on a gddr5x variant of vega, and now they can afford it ?

I would be extremely surprised if vega supported GDDR5(x), large parts of the chip design would be based around HBM2whcih means it would not be very optimal at all with the different characteristics of GDDR5.

Its not so much postpoining but getting yields, supply, drivers and firmware up to speed. they may be able to pull back a little to have lower availability at launch if everything keeps to schedule, and with a little pushing but its fairly limited.

Anyway, I don;t think lack of finances meant that vega wasn't going to be released until Q4/2106-Q1/2017 but things like HBM2 availability. Lack of finances might have meant they decided to skip on a mid-sized chip but I'm sure they would've other plans, perhaps sell Polaris so cheap you can xfire them for less than 1080 prices.

 

[Jono8]

Well I am glad AMD are going to reveal/announce Polaris by the end of May. Want to get something new in June and will be good to have the full picture on which is better.

 

[AlamoX]

I don't think a full die P10 will have any difficulty matching or beating a 1070 in most games. I think the latter will struggle to get near a 980Ti much of the time ... and I think both GP104 cards will be really memory bandwidth gated. In titles that saturate memory bw quickly, they may perform very poorly.

1080 has 320g/s bandwidth, thats almost like a TitanX, i think it will be fine at high rez, 1070 on the other hand, could scale badly in 4k

 

[pmc25]

No,I am saying either Polaris 10 is a bit underwhelming like Fiji,so they are being silent about it,and trying to tweak it as much as possible,or either they know they are onto a winner,and trying not to give away anything to Nvidia.

I doubt they're hugely concerned with what NVIDIA are doing. The latter rushed out a paper launch with what will be limited availability 3-5 weeks from now and volume months later. That would suggest NVIDIA are the ones that are worried. AMD have still showed more confirmed working hardware with Polaris than NVIDIA have Pascal, at both CES and the Capsaicin event.

 

[pmc25]

1080 has 320g/s bandwidth, thats almost like a TitanX, i think it will be fine at high rez, 1070 on the other hand, could scale badly in 4k

They're claiming to be a long way ahead of a TX, and 1070 to match one. The reality's likely to be far more modest. It was thought that TX had just enough bandwidth to not be gated by it. The number of shaders and those clocks (even the lower specified ones, not the headline grabbing demo'd ones) are going to starve quickly.

I suspect this is the main reason their resolution gating software for VR was developed.

 

[Rroff]

I doubt they're hugely concerned with what NVIDIA are doing. The latter rushed out a paper launch with what will be limited availability 3-5 weeks from now and volume months later. That would suggest NVIDIA are the ones that are worried. AMD have still showed more confirmed working hardware with Polaris than NVIDIA have Pascal, at both CES and the Capsaicin event.

Never mind they've given a dozen cards to reviewers and working with people like id with cards but hey don't let facts get in the way of your narration.

 

[Jono8]

I doubt they're hugely concerned with what NVIDIA are doing. The latter rushed out a paper launch with what will be limited availability 3-5 weeks from now and volume months later. That would suggest NVIDIA are the ones that are worried. AMD have still showed more confirmed working hardware with Polaris than NVIDIA have Pascal, at both CES and the Capsaicin event.

Lol, what?

You do realise loads of people actually already have 1080Gtx's and are benchmarking and reviewing them as we speak right ...

http://www.guru3d.com/news-story/geforce-gtx-1080-photos.html

"So at last nights event there where a couple of setups showing the GeForce GTX 1080. The cards where available in large quantities alright."

 

[Rroff]

Plus as we've only seen one indication of the cards tapeout, etc. which was roundly denounced at the time - but here we are any way - we pretty much have to assume it is pretty much on that timetable and will be available in volume sooner rather than later.

 

[pmc25]

Maybe I should take back what I said about Vega not coming this year.

3Dcentre are usually very reliable ... I had rubbished all the GP102 and Vega rumours, but if they're running with it now, maybe they're true ...

http://www.3dcenter.org/news/amd-zieht-den-vega-launch-angeblich-auf-oktober-2016-vor

That would certainly be very interesting. Particularly if NVIDIA choose to completely forgo HBM2 in consumer cards until Volta (mid '18).

 

[Rroff]

Dunno nVidia have been doing some strange things with regard to interposers so could be they are struggling with it or could mean something completely different but might indicate HBM2 won't come on Pascal any time soon. I don't really buy the SP numbers though unless they are counting both FP32 and 64 in there which is a strange thing to do - nVidia don't usually scale them up like that - when claims are made of them shooting up generation on generation it usually tends to indicate fake info.

 

[queamin]

I think Nvidia have a steep learning curve with interposer's, more so than Amd.