AMD Polaris architecture – GCN 4.0

[drunkenmaster]

Is this your way of admitting HBM1 was not up to the job ?

Do you ever get bored of trolling?

Nothing I said even hints at HBM not being up to the job. Is Maxwell not up to the job? If it is, then why is Nvidia making Pascal? Why did Nvidia make Maxwell when they had Kepler, and on and on and on.

Technology gets replaced with newer better technology... welcome to how the silicon industry works, is it your first day reading about the silicon industry, is this a new concept for you?

HBM is a monumental leap in technical capability and performance over gddr5. That doesn't mean it there was ever the slightest chance in hell it would be the last ever memory technology and saying something can be improved isn't remotely the same as saying something isn't good to begin with. HBM1 is more than enough for Fiji and for Polaris, HBM2 is more than enough for Vega, it's probably more than enough for the next generation as well but that doesn't mean there might not be something that can provide the same bandwidth for a lower power usage by then which will be superior.

You also don't get to HBM3, without going through HBM1 and HBM2, again another little way that the silicon industry works. It's amazing how much knowledge you have over everyone else because you've taken a heatsink off a Fiji, it trumps basic knowledge of the industry, physics, logic.

 

[bru]

Remember people the first iteration of most new memory techs are not necessarily as good as the last iteration of the outgoing memory tech.

HBM the concept is absolutely fantastic, even if the first generation is only a bit meh.

 

[drunkenmaster]

The first generation gave smaller memory controller size, 50W reduction in the amount of power required for 512GB/s of bandwidth, gave a bandwidth level completely unmatched by any previous card and Fury X is faster than 980ti/Titan X in most high bandwidth situations. As DX12 is coming in Fiji is just stretching it's legs further. Fiji's problem is not memory related, but the problem of making use of so many shaders. Async compute helps to utilise those shaders much more effectively and when it does the overall Fury X's performance is fantastic and only possible due to HBM.

If it instead used a GDDR5 controller, it would be a 512bit/8Ghz or so memory bus. It would use an extra 50W, it would have less shaders and would be lower performance. It enabled The Nano to be made which most people seem to love, performance in that form factor never achieved before, a performance per watt level that is amazing. Nvidia used up an entire architectural improvement with Maxwell and AMD matched that performance per watt simply by using HBM.

It's anything but meh.

 

[FoxEye]

If AMD does not even have a soft release with reviews in July they have essentially missed the back to school period.

Nope. If they had said "back to school 2016" then late August would qualify. Because any time before school starts would be "back to school", thus if they released Sept 1st they'd still have released in the back to school window.

But they rather mystically said "before back to school 2016", which could mean any time from the announcement until September 9th or so. Incredibly vague information.

But to say they have to release in July or they've missed back to school 2016 is just wrong, whichever way you look at it. Nobody is going back to school in July.

 

[bru]

If it instead used a GDDR5 controller, it would be a 512bit/8Ghz or so memory bus. It would use an extra 50W, it would have less shaders and would be lower performance. It enabled The Nano to be made which most people seem to love, performance in that form factor never achieved before, a performance per watt level that is amazing. Nvidia used up an entire architectural improvement with Maxwell and AMD matched that performance per watt simply by using HBM.

It's anything but meh.

Taking the part in red first. I completely agree, AMD's GCN1.2 ( Tonga and Fiji) architecture is great and performs brilliantly at any resolution, but the HBM equipped Fiji has shown that at lower resolutions it doesn't have consistent performance scaling.

So back to the non red part.
If Fiji had of been a GDDR5 equipped card, yes, it would not have performed as well as the HBM equipped Fiji does, but it would have had consistent performance scaling, properly with resolution unlike the HBM equipped Fiji.

 

[Orangey]

Back to school means the period people buy new school supplies in and also when new products appear to fill this demand.

Stop with the autistic overanalysis of charts and verbiage trying to divine The Truth. It isn't there.

 

[Kaapstad]

Do you ever get bored of trolling?

Nothing I said even hints at HBM not being up to the job. Is Maxwell not up to the job? If it is, then why is Nvidia making Pascal? Why did Nvidia make Maxwell when they had Kepler, and on and on and on.

Technology gets replaced with newer better technology... welcome to how the silicon industry works, is it your first day reading about the silicon industry, is this a new concept for you?

HBM is a monumental leap in technical capability and performance over gddr5. That doesn't mean it there was ever the slightest chance in hell it would be the last ever memory technology and saying something can be improved isn't remotely the same as saying something isn't good to begin with. HBM1 is more than enough for Fiji and for Polaris, HBM2 is more than enough for Vega, it's probably more than enough for the next generation as well but that doesn't mean there might not be something that can provide the same bandwidth for a lower power usage by then which will be superior.

You also don't get to HBM3, without going through HBM1 and HBM2, again another little way that the silicon industry works. It's amazing how much knowledge you have over everyone else because you've taken a heatsink off a Fiji, it trumps basic knowledge of the industry, physics, logic.

I only need to remember one single fact about HBM1 and that is - the performance is bad @1080p but improves if you overclock it on LN2.

Obviously no one is going to game using LN2 but it proves the basic concept that HBM1 clockspeed is way too low and throttles performance.

HBM1 is flawed end of.

 

[P4Clock]

HBM1 is flawed end of.

they made a leap, this is something that as not happened for a long time in the tech world..

 

[Scougar]

The first iteration of most new tech's is never that good, and often the previous tech performs better in some areas. Really no surprise.

I am darn glad that companies are trying out new memory techniques. Keep pushing the envelope, keep competition alive. Innovate or die.

 

[muziqaz]

Amazing how Kaap comes up with conclusion that HBM failed because he interpreted 1080p game benchmarks wrong
Are we now completely ignoring such thing as driver overhead at lower resolutions?

 

[CAT-THE-FIFTH]

From Wikipedia:

https://en.wikipedia.org/wiki/Back_to_school_(marketing)



From the second link from the US marketing site:



Another link:

http://adage.com/article/news/july-i-back-school-ads/242979/



Another one:

http://www.marketingcharts.com/traditional/back-to-school-stats-2015-57232/



http://www.pfsweb.com/blog/back-to-school-ecommerce-a-new-u-s-digital-holiday/





That is over a third of purchases made in July alone with something like 65% of people doing research during that time.


The first day of summer in the US this year is the 20th of June:

http://www.apples4theteacher.com/holidays/summer/first-day-of-summer.html

No AMD show during July will probably mean less interest in Polaris as they will probably only see Maxwell based systems in laptops and desktops.



Hence,if what the AMD CEO is saying is correct,for Polaris to be ready for the back to school season and be available in the middle of the year that would be either late June or sometime during July.

August/September is not the middle of the year.

If it is any later it would technically be not the middle of the year and would not in time for the back to school season as July is still consider an important month for sales too. It would be technically delayed by what the CEO said.

Whether it is a limited launch is another question but both the HD7970 and Fury X pretty much launched first and it took weeks for stock to actually being available,since it was all on pre-order IIRC.

I expect reviews will drop sometime during late June or sometime during July and then the cards will be on pre-order or have limited quantities in computer hardware retailers.

That is most likely what we will see IMHO.

Parents research and make back-to-school purchases at different times – July is all but the unofficial month for researching back-to-school products. A recent study by Google and Ipsos Research concluded that 65% of U.S. internet users started their back-to-school research in the month of July. Many of these researchers, however, waited until August to make purchases – 48% of U.S. internet users made back-to-school purchases in August, only 35% made purchases in July.

Nope. If they had said "back to school 2016" then late August would qualify. Because any time before school starts would be "back to school", thus if they released Sept 1st they'd still have released in the back to school window.

But they rather mystically said "before back to school 2016", which could mean any time from the announcement until September 9th or so. Incredibly vague information.

But to say they have to release in July or they've missed back to school 2016 is just wrong, whichever way you look at it. Nobody is going back to school in July.

What is with the internet and people not even reading sources - what you think back to school is irrelevant. What is relevant is what retailers consider is back to school. Whichever way you read it.

I like how some of you think your own personal views are better than Google and IPSOS MORI. Our government uses the latter company for market research.

Its a marketing term covering most of July and August - between both those analytics companies,80% of back to school purchases are between July and August. 65% of back to school research is done during July. No release or reviews in July and AMD has essentially has missed the period of any reasonable sales.

Plus:

The target is mid-2016, particularly in time for the major back-to-school season in the united states. Polaris should be available not just in the form of desktop solutions, but also notebook graphics chips.

September is not mid 2016. End of August is pushing mid 2016.

At this point either some of you cannot read or it is just like a Trump level way of doing stuff.

If you have a problem with what I say complain to Google and IPSOS MORI about their metrics and data analysis.

If AMD thinks more or less September(which late August) is mid 2016 and a great release date for Polaris,then they have screwed themselves.

According to Google and IPS0 MORI who I would trust more than two posters on the internet,they just lost 80% of potential back to school sales to Nvidia with Maxwell.

If Nvidia even get one Pascal GPU out in cards by then they will be screwed.

Plus AMD has already delayed the cards already:

http://semiaccurate.com/2016/03/22/why-did-amd-slip-polaris-and-vega/

AMD is slipping their GPU roadmap by a few quarters, but why and which ones are the interesting part. More important to the semiconductor space at large though is which one of the likely causes is behind the slip.

You might recall when SemiAccurate exclusively brought you the name Vega 10, at the time we had it pegged as VG10, we also had some dates and specs. Our sources were quite specific about the launches, and we are fairly sure that the ‘Capsaicin’ slide deck represents between a minor and significant slip. Lets take a look at this in detail.

Apparently the cards were meant to launch late Q1 2016 or sometime during early Q2 2016,according to someone who has seen the full article.

Mid 2016 is already a delayed date.

Now with some of you saying its Q3 2016,thats means a second delay on the first delay.

 

[drunkenmaster]

The fairly simple fact is that having 2-3 times the bandwidth required for 1080p doesn't explain a potential lack of performance at 1080p in the slightest. However having 64 Rops, which is exactly the same as the 290x, very much explains(as well as driver overhead) the lack of huge maximums at 1080p which drag the averages down.

But then Kaap committed himself, he ragged on HBM for 6 months before Fury X arrived, he showed a continued complete lack of understanding about how it worked, what it would do and how it would help reduce power. He decided to hate it(almost certainly because AMD would have it long before Nvidia) before it ever appeared and continues to blame everything in the world on it. 1080p performance, famine in Africa, terrorism... all down to that **** HBM. See a benchmark result you don't like, don't look for answers, don't look at the architecture and see what the difference might be... it's definitely HBM's fault because AMD suck and anything AMD came up with must be bad.

 

[Mauller]

I only need to remember one single fact about HBM1 and that is - the performance is bad @1080p but improves if you overclock it on LN2.

Obviously no one is going to game using LN2 but it proves the basic concept that HBM1 clockspeed is way too low and throttles performance.

HBM1 is flawed end of.

And yet this piece of drivel you keep spouting has been debunked multiple times when driver overhead is no longer an issue. Or do you like to ignorantly ignore all dx12 benches when fijis dx12 performance jumps ahead?

Or even scratch that, even in dx11 Fiji runs better than ever and runs faster than a 980ti in farcry primal and a few other games now.

The problem is not HBM and never has been, it has been shader utilisation.

 

[CAT-THE-FIFTH]

http://semiaccurate.com/2016/03/22/why-did-amd-slip-polaris-and-vega/

AMD is slipping their GPU roadmap by a few quarters, but why and which ones are the interesting part. More important to the semiconductor space at large though is which one of the likely causes is behind the slip.

You might recall when SemiAccurate exclusively brought you the name Vega 10, at the time we had it pegged as VG10, we also had some dates and specs. Our sources were quite specific about the launches, and we are fairly sure that the ‘Capsaicin’ slide deck represents between a minor and significant slip. Lets take a look at this in detail.

I don't have access to the full story but what I gather from someone who has access over there the original Semiaccurate story claimed Polaris in late Q1/early Q2 but it was pushed back to Q3. So the Mid 2016 date IS the pushed back date.

Not sure how accurate that info is to be honest.

So it looks like Polaris is delayed already. If AMD think September is a good release date for Polaris,like some think it is,Nvidia will be jumping for joy as according to Google and IPSOS MORI,80% of back to school sales are in July and August.

They won't be in time for back to school,they would have missed most of it.

Looks like Gibbo saying late summer has now been confirmed by SA too,and considering he said summer for Nvidia too it means AMD will probably be releasing Polaris at around the same time as Pascal now. AMD have always beaten Nvidia to the punch to a new node for the last 7 years at least IIRC,so this will be terrible.

So AMD have been telling porkies when saying "mid 2016" - September 2016 is pushing "mid-2016" a lot.

So,AMD have frakked this up and hopefully its not true,as they need some extra sales.

Nearly EVERY CPU and GPU they have released in the last few years has been delayed.

Polaris being released now would have really made Nvidia sweat.

 

[Mauller]

Back to school sales tend to involve prebuilt machines. So even if the public release of dGPU were later the first round of parts can still be going to OEMs. But Q3 is still middle of the year so across the summer June till September period.

 

[CAT-THE-FIFTH]

Back to school sales tend to involve prebuilt machines. So even if the public release of dGPU were later the first round of parts can still be going to OEMs. But Q3 is still middle of the year so across the summer June till September period.

Good point - this I had not considered. So release for public during July and August when 80% of back to school sales happens,and a later release for us lot.

But that would still mean they need to have reviews out at least in July though??

Otherwise the public would only see reviews for the Nvidia cards and will have no clue how the AMD ones fare and then just buy Nvidia.

AMD needs to get everything perfect,even timing,to breakout of the 4 to 1 advantage Nvidia has over them.

That means a decent product,a decent price and releasing ahead of Nvidia with time to spare.

 

[Mtom]

http://semiaccurate.com/2016/03/22/why-did-amd-slip-polaris-and-vega/





So it looks like Polaris is delayed already. If AMD think September is a good release date for Polaris,like some think it is,Nvidia will be jumping for joy as according to Google and IPSOS MORI,80% of back to school sales are in July and August.

They won't be in time for back to school,they would have missed most of it.

Looks like Gibbo saying late summer has now been confirmed by SA too,and considering he said summer for Nvidia too it means AMD will probably be releasing Polaris at around the same time as Pascal now. AMD have always beaten Nvidia to the punch to a new node for the last 7 years at least IIRC,so this will be terrible.

So AMD have been telling porkies when saying "mid 2016" - September 2016 is pushing "mid-2016" a lot.

So,AMD have frakked this up and hopefully its not true,as they need some extra sales.

Nearly EVERY CPU and GPU they have released in the last few years has been delayed.

Polaris being released now would have really made Nvidia sweat.

What is this stupid nonsense?
SA cites that the Capsaicin roadmap shows delays compared to what they knew. Capsaicin roadmap shows polaris at mid 2016 just as AMD said.

Oh i see, so they predicted Q1 release, and because AMD will release Polaris mid 2016 (when they told they will) now its a delay and slip, because they don't do as SA said so. "Insert facepalm pic here"

 

[CAT-THE-FIFTH]

What is this stupid nonsense?
SA cites that the Capsaicin roadmap shows delays compared to what they knew. Capsaicin roadmap shows polaris at mid 2016 just as AMD said.

According to the full article,Polaris was meant to be late Q1 or early Q2. It has been delayed a quarter already. The Capsaicin roadmap is the new one with the delayed releases.

They have just massively cut their head start over Nvidia if what Gibbo says is true,ie,late summer now for Polaris and "summer" for Pascal.

I just hope what Mauller says is true - all the prebuilts get released more in early summer.

AMD releasing on a new node at the same time as Nvidia has not happened for a very long time and it would be disaster for them.

Edit!!

Late summer as Gibbo says would make it delayed over a quarter then.

So that probably confirms good old GF are being used not Samsung.

 

[Steampunk]

http://semiaccurate.com/2016/03/22/why-did-amd-slip-polaris-and-vega/

So it looks like Polaris is delayed already. If AMD think September is a good release date for Polaris,like some think it is,Nvidia will be jumping for joy as according to Google and IPSOS MORI,80% of back to school sales are in July and August.

They won't be in time for back to school,they would have missed most of it.

Looks like Gibbo saying late summer has now been confirmed by SA too,and considering he said summer for Nvidia too it means AMD will probably be releasing Polaris at around the same time as Pascal now. AMD have always beaten Nvidia to the punch to a new node for the last 7 years at least IIRC,so this will be terrible.

So AMD have been telling porkies when saying "mid 2016" - September 2016 is pushing "mid-2016" a lot.

So,AMD have frakked this up and hopefully its not true,as they need some extra sales.

Nearly EVERY CPU and GPU they have released in the last few years has been delayed.

Polaris being released now would have really made Nvidia sweat.

I think you may be over-reacting here. It says "mid-2016 is the pushed back date". So it's gone from "late Q1/early Q2 to Q3" means it's gone from March/April to July/August/September. Given AMD have stated "back to school", it seems likely to be early Q3 rather than late Q3.

Three months is only 16 weeks, and may be a way to make sure Nvidia can't react too much with clockspeed changes. If AMD is sure they have a winner and Pascal had to come out at the same time with no chance for Nvidia to react, AMD may be betting on a decent win.

 

[CAT-THE-FIFTH]

I think you may be over-reacting here. It says "mid-2016 is the pushed back date". So it's gone from "late Q1/early Q2 to Q3" means it's gone from March/April to July/August/September. Given AMD have stated "back to school", it seems likely to be early Q3 rather than late Q3.

Three months is only 16 weeks, and may be a way to make sure Nvidia can't react too much with clockspeed changes. If AMD is sure they have a winner and Pascal had to come out at the same time with no chance for Nvidia to react, AMD may be betting on a decent win.

I really hope you are right!! AMD needs a homerun with Polaris and a few months of Maxwell looking meh would help them.

Unless they have some inside information saying Pascal is delayed or Polaris is much better,I would not be cutting it close with Nvidia. I still remember when Nvidia sold more FX cards than ATI with the 9000 series,despite the former being crap!

The reason why the HD5000 series did so well is since they had a six month headstart on Nvidia,had decent price,decent performance and decent power consumption. But they could not still outsell Nvidia but did gain back marketsahre,and once Fermi released Nvidia gained back sales.