(GURU3D RUMOR) AMD Polaris 10 GPU To Offer Near 980 Ti Performance For 299 USD?

[DaveSimmons]

Something I'd also love to see from AMD is a performance and power draw competitor to the 750ti, for $99 or less.

Something you can recommend to people with a 300 watt Dell PSU, fit into a mini-ITX, add to your old system to make it more usable as a backup.

 

[JDG1980]

AMD would like to regain market share while still making a decent profit on each unit sold. A Pitcairn-sized die on 14LPP is ideally positioned to let them do that.

Back in 2012, AMD released the cut-down Pitcairn (7850) for $249, and it was widely considered one of the best perf/$ offerings at the time. The full Pitcairn SKU (7870) was $349. I expect that we'll see similar pricing for Polaris 10 cards. I know a lot of people talk up how expensive FinFET chips allegedly are, and while I don't doubt the wafers are more expensive than 28nm wafers now, we should instead be asking how the cost compares to that of 28nm wafers back in 2012 when that node was just starting up. I found some contemporary articles talking about how expensive 28nm wafers were, so the current teething pains are not unique to FinFET. If anything, the current FinFET nodes are more mature than 28nm was in early 2012, since the early adopter premiums were paid by Apple and Samsung. I believe that the bill of materials for a Polaris 10 card now will be no more than it was for a Pitcairn card back in 2012. Memory bus width is the same, TDP/power consumption should be the same or lower, and I think GDDR5 costs have gone down far enough to offset the increased capacity on newer cards.

Think about what AMD is currently selling at these price points. R9 390 is in the $300-$330 range, and that's a Hawaii chip (438mm^2) with an ultra-wide 512-bit memory bus (more PCB layers, more GDDR5X chips) and a massive TDP (beefier power stage and larger cooler required). Polaris 10 has got to be more profitable than that, while still providing a better deal to customers.

I expect the cut Polaris 10 ("R9 480") to provide performance similar to R9 390/390X for $249, and the full Polaris 10 with higher clocks at $349 ("R9 480X") to provide performance roughly on par with Fury X overall: maybe ~10% higher at 1080p, and ~10% less at 4K. Both cards will provide considerably better perf/$ than anything currently on the market.

 

[swilli89]

AMD would like to regain market share while still making a decent profit on each unit sold. A Pitcairn-sized die on 14LPP is ideally positioned to let them do that.

Back in 2012, AMD released the cut-down Pitcairn (7850) for $249, and it was widely considered one of the best perf/$ offerings at the time. The full Pitcairn SKU (7870) was $349. I expect that we'll see similar pricing for Polaris 10 cards. I know a lot of people talk up how expensive FinFET chips allegedly are, and while I don't doubt the wafers are more expensive than 28nm wafers now, we should instead be asking how the cost compares to that of 28nm wafers back in 2012 when that node was just starting up. I found some contemporary articles talking about how expensive 28nm wafers were, so the current teething pains are not unique to FinFET. If anything, the current FinFET nodes are more mature than 28nm was in early 2012, since the early adopter premiums were paid by Apple and Samsung. I believe that the bill of materials for a Polaris 10 card now will be no more than it was for a Pitcairn card back in 2012. Memory bus width is the same, TDP/power consumption should be the same or lower, and I think GDDR5 costs have gone down far enough to offset the increased capacity on newer cards.

Think about what AMD is currently selling at these price points. R9 390 is in the $300-$330 range, and that's a Hawaii chip (438mm^2) with an ultra-wide 512-bit memory bus (more PCB layers, more GDDR5X chips) and a massive TDP (beefier power stage and larger cooler required). Polaris 10 has got to be more profitable than that, while still providing a better deal to customers.

I expect the cut Polaris 10 ("R9 480") to provide performance similar to R9 390/390X for $249, and the full Polaris 10 with higher clocks at $349 ("R9 480X") to provide performance roughly on par with Fury X overall: maybe ~10% higher at 1080p, and ~10% less at 4K. Both cards will provide considerably better perf/$ than anything currently on the market.

I think you nailed it. The question is $299 or $349. I'm more amd more confident that a fully enabled Polaris 10 will match a Fury X.

 

[moonbogg]

I think their main focus is getting a capable GPU in the hands of many people for VR, just like the guy said. The price for such a card has to be good, very good actually. I think $300.00 is too much personally. That's too high for a main stream product ON TOP of the cost of a VR set up. That is getting very expensive for most people, especially for something that they have no real experience with and its a gamble. $250 or less I say otherwise its just too much.

 

[Blitzvogel]

Something I'd also love to see from AMD is a performance and power draw competitor to the 750ti, for $99 or less.

Something you can recommend to people with a 300 watt Dell PSU, fit into a mini-ITX, add to your old system to make it more usable as a backup.

That's what Polaris 11 is shaping up to be. Only problem is that with the PS4 Neo and perhaps Nintendo NX, the bar for console games could go up quite a bit, effecting PC ports as a result. The best case scenario I think for an external power-less Polaris card is the 2.5 to 3.0 TFLOP range - well below the 4 TFLOP graphics figure for the PS4 Neo.

 

[SteveGrabowski]

I would not be surprised if we see a 2304 sp GDDR5 based Polaris 10 match 980 Ti at USD 299

That would be the most epic troll job ever, considering they're talking about Polaris 10 like it's offering 390 level performance by focusing on VR and how they were demoing Polaris 11 against the GTX 950 back in January. Polaris 10 being like 3 tiers of performance above Polaris 11 would be strange.

 

[VirtualLarry]

Something I'd also love to see from AMD is a performance and power draw competitor to the 750ti, for $99 or less.

Something you can recommend to people with a 300 watt Dell PSU, fit into a mini-ITX, add to your old system to make it more usable as a backup.

But with competitive performance. A 750ti with today's games, is really barely even cutting it. (Sub-30FPS @ 1080P on Med. on most games.)

Edit: To add, I consider the 750ti to be barely better than a GT740 GDDR5 card. I mean, I know it IS better, but not by enough to matter. GTX950 is the minimum card I would consider these days for minimally-competent 1080P gaming on modern titles. Anything lower, you should run the game in low-res (below 1080P) mode.

 

[xpea]

But with competitive performance. A 750ti with today's games, is really barely even cutting it. (Sub-30FPS @ 1080P on Med. on most games.)

You are few quarters late. Nvidia already has the GTX950 taking his power only from the PCIe bus. Asus, EVGA, Gigabyte and MSI have this less than 75W puppy in their catalog:
http://www.anandtech.com/show/10250...-gtx-950-with-75w-power-consumption-to-lineup

Still not great for 1080p gaming but at 28nm mode, it's impressive. AMD is very very late in this segment and their window is very tight as last rumors say that GP106 is now expected to hit the shelves during this summer...

 

[book_ed]

Think about what AMD is currently selling at these price points. R9 390 is in the $300-$330 range, and that's a Hawaii chip (438mm^2) with an ultra-wide 512-bit memory bus (more PCB layers, more GDDR5X chips) and a massive TDP (beefier power stage and larger cooler required). Polaris 10 has got to be more profitable than that, while still providing a better deal to customers.

I expect the cut Polaris 10 ("R9 480") to provide performance similar to R9 390/390X for $249,

R290 was ~ $250 when R390 launched and the same chip just jumped in price to $330. Getting the ~ same performance of R390/x to $250 would mean jumping down to roughly the same price/performance Hawaii had before AMD decided to jack up the prices. As a consumer I would feel a bit ripped off.

That's what Polaris 11 is shaping up to be. Only problem is that with the PS4 Neo and perhaps Nintendo NX, the bar for console games could go up quite a bit, effecting PC ports as a result. The best case scenario I think for an external power-less Polaris card is the 2.5 to 3.0 TFLOP range - well below the 4 TFLOP graphics figure for the PS4 Neo.

Even though more powerful consoles will be launched, Xbox One is still the weakest link. You can't ignore that and not code for it.

 

[Blitzvogel]

Even though more powerful consoles will be launched, Xbox One is still the weakest link. You can't ignore that and not code for it.

True, but there are alot of sub-1080p games on it.

 

[kraatus77]

R290 was ~ $250 when R390 launched and the same chip just jumped in price to $330. Getting the ~ same performance of R390/x to $250 would mean jumping down to roughly the same price/performance Hawaii had before AMD decided to jack up the prices. As a consumer I would feel a bit ripped off.



Even though more powerful consoles will be launched, Xbox One is still the weakest link. You can't ignore that and not code for it.

First , that 250$ price wasn't for all over the world but mostly US-ish and it was occasional price, second it was mainly retailers/oem who did that to clear inventory, not the AMD. and third, 390/x launched with better coolers, better clocks, more but not necessary vram and lower launch prices than 290/x. and finally the people who are complaining it's price weren't going to buy it anyway, and are mostly same people who praised og titans 's performance for 500$ extra because hurr-durr 20% bettah short lived performance over 7970.

In nvidia's case more performance = more price

but suddenly everyone expects Amd to stay cheap option and when they deliver " omg how dare they charge extra for more performance".

(hint - 7970 even though they only charged what nvidia's best card was selling for that time aka 580)

 

[airfathaaaaa]

First , that 250$ price wasn't for all over the world but mostly US-ish and it was occasional price, second it was mainly retailers/oem who did that to clear inventory, not the AMD. and third, 390/x launched with better coolers, better clocks, more but not necessary vram and lower launch prices than 290/x. and finally the people who are complaining it's price weren't going to buy it anyway, and are mostly same people who praised og titans 's performance for 500$ extra because hurr-durr 20% bettah short lived performance over 7970.

In nvidia's case more performance = more price

but suddenly everyone expects Amd to stay cheap option and when they deliver " omg how dare they charge extra for more performance".

(hint - 7970 even though they only charged what nvidia's best card was selling for that time aka 580)

7970 was ok.. the 7970ghz edition was a trully rip off

 

[mohit9206]

Another business analyst wanna be. You have NO idea about either company's corporate strategy for market position and it's insane to assume they won't charge x amount for y performance with ZERO knowledge of their cost structure or market strategy.

Leave the business analyst stuff alone and let these conversations be about being an enthusiast. Or maybe they should hire you as CMO to ensure they get huge margins and marketshare!

You're right am no business analyst but i know that first and foremost any company wants to earn a profit on every product sold. I doubt they would make much if any profit by selling 980ti level card for $300.
Also assuming they do still make a decent profit by selling it at $300 but they still won't sell it at $300. Pricing is always set to the maximum amount the market will bear. And its clear to me and surely to AMD as well that most people will be plenty happy with 980Ti level performance for $450 so why price it lower and reduce your profits when the market is happy to pay $450?
Well assuming market is unhappy with $450 and sales are low compared to the competition then there's always price cuts. But underpricing your product right from the start is bad for business.

 

[airfathaaaaa]

You're right am no business analyst but i know that first and foremost any company wants to earn a profit on every product sold. I doubt they would make much if any profit by selling 980ti level card for $300.
Also assuming they do still make a decent profit by selling it at $300 but they still won't sell it at $300. Pricing is always set to the maximum amount the market will bear. And its clear to me and surely to AMD as well that most people will be plenty happy with 980Ti level performance for $450 so why price it lower and reduce your profits when the market is happy to pay $450?
Well assuming market is unhappy with $450 and sales are low compared to the competition then there's always price cuts. But underpricing your product right from the start is bad for business.

well not if you put the "VR" in that business

i mean they could very well push their cards "mostly" on how they can manage vr perf and not so much about gaming perf along with other vr capable peripherals with amd "blessing"
so pricing so low with that in mind could very well work

 

[book_ed]

True, but there are alot of sub-1080p games on it.

Indeed, but you still require to have playable frame rates and don't take out too much IQ or else the first adopters would be rightfully p****d off. However, we may have better optimized high "value" effects, such as high tessellation would mean more objects that would benefit of this technique would actually use it, instead of just a few over tessellated. We'll see, but I don't imagine a different path in improving IQ compared to previous gen - PS3/xbox360 vs. pc.

First , that 250$ price wasn't for all over the world but mostly US-ish and it was occasional price, second it was mainly retailers/oem who did that to clear inventory, not the AMD. and third, 390/x launched with better coolers, better clocks, more but not necessary vram and lower launch prices than 290/x. and finally the people who are complaining it's price weren't going to buy it anyway, and are mostly same people who praised og titans 's performance for 500$ extra because hurr-durr 20% bettah short lived performance over 7970.

In nvidia's case more performance = more price

but suddenly everyone expects Amd to stay cheap option and when they deliver " omg how dare they charge extra for more performance".

(hint - 7970 even though they only charged what nvidia's best card was selling for that time aka 580)

I gave it as reference price. I don't get the launch price where I live as well (so that potentially $250 launch price can be $300 or more) or some major deals, it's just to have a base we can talk - you can also factor in some deals bigger retailers will have for some models (if I remember correctly, an acquaintance of mine bought a custom R290 for around $250 in USA), so lets keep things simple. In my country, the price difference between a R290 custom and R390 was about 50 euros (give or take, depending on the shop, model, etc.). As long as you have a good cooler on that R290, you can potentially come quite close if not equal to R390, minus the extra 4GB vRAM which I'd say it's pointless for the majority of the gamers.

Anyway, their business, their choice on how to position the products, but if I was on the market for buying in that price range, I'd look new and used as well. A sh R290 custom was a far better deal than a brand new GTX970 for roughly the same performance - resolution depending, of course, just as custom R290 was cheaper and practically the same as a more expensive R390 even compared as "new" and no used.

 

[3DVagabond]

One problem here. Taking Samsung's claim of <0.2 defects/cm^2 at face value.

Using a die of 15mm x 15.5mm with a 300mm wafer and a defect density of 0.19/cm^2 gives 155 good full die out of 238 max. This does not leave a lot for cut down models which should have higher sales volumes.

By the way, if a wafer costs $4000, we have fab costs as $26/die assuming we throw all the defective ones away. With harvesting, I'll bet sub $20/die.

We could have low power high perf/W full dies with DDR5 and high power, high performance, lower perf/W full die models with GDDR5X. This would be exactly as the 4850 and 4870 line.

So cut version and several full versions.

AMD has cut down perfectly good dies in previous gens.

 

[Vesku]

You're right am no business analyst but i know that first and foremost any company wants to earn a profit on every product sold. I doubt they would make much if any profit by selling 980ti level card for $300.
Also assuming they do still make a decent profit by selling it at $300 but they still won't sell it at $300. Pricing is always set to the maximum amount the market will bear. And its clear to me and surely to AMD as well that most people will be plenty happy with 980Ti level performance for $450 so why price it lower and reduce your profits when the market is happy to pay $450?
Well assuming market is unhappy with $450 and sales are low compared to the competition then there's always price cuts. But underpricing your product right from the start is bad for business.

Given their marketshare position it makes a lot of sense to offer a better than expected price/performance ratio. Heck it made sense to Nvidia to do so with the 970 even without such a lopsided sales ratio versus their competitor.

Complementing AMD's exclusive position in consoles with a high install base of VR capable gaming cards will greatly benefit them in terms of developer mindshare going forward. This should act as a counter force to Nvidia initiatives like GameWorks, VRWorks, and being preferred GPU of the Unreal Engine team. We are already seeing what the initial console wins have started to do for AMD in the PC space with DX12.

If Polaris 10 is really sub-250mm2 it could very well be in SKUs under the $300 mark at launch. Given the expectations of a node shrink GPU generation if there is a Polaris SKU that is not really much different from 390(X) performance then it kind of has to slot in somewhere in the $200s or risk being a flop.

I really wouldn't be surprised to see a $299-350 Polaris 10 with near 980Ti performance as the top SKU. Even at 232mm2 that should put the transistor count similar to full Hawaii yet it will have design improvements and should have a decent bump in clocks in the top end performance oriented SKU.

 

[3DVagabond]

You're right am no business analyst but i know that first and foremost any company wants to earn a profit on every product sold. I doubt they would make much if any profit by selling 980ti level card for $300.
Also assuming they do still make a decent profit by selling it at $300 but they still won't sell it at $300. Pricing is always set to the maximum amount the market will bear. And its clear to me and surely to AMD as well that most people will be plenty happy with 980Ti level performance for $450 so why price it lower and reduce your profits when the market is happy to pay $450?
Well assuming market is unhappy with $450 and sales are low compared to the competition then there's always price cuts. But underpricing your product right from the start is bad for business.

It's supply and demand. We know they've had difficulty meeting their WSA @ GloFo. It's possible the number of wafers they have to fill means they have huge supply and need to price them lower because of huge supply.

 

[xpea]

Complementing AMD's exclusive position in consoles with a high install base of VR capable gaming cards will greatly benefit them in terms of developer mindshare going forward.

Sorry ?
Estimated shipment of VR headset for 2016 is only 2.5 million units
(source: http://www2.deloitte.com/global/en/...dia-virtual-reality-billion-dollar-niche.html )
This VR blurb is not where the money will be made in 2016. As usual, AMD is too much looking forward and doesn't think enough of the present. Then, when the function really matters, they are behind...

 

[3DVagabond]

Sorry ?
Estimated shipment of VR headset for 2016 is only 2.5 million units
(source: http://www2.deloitte.com/global/en/...dia-virtual-reality-billion-dollar-niche.html )
This VR blurb is not where the money will be made in 2016. As usual, AMD is too much looking forward and doesn't think enough of the present. Then, when the function really matters, they are behind...

Yeah. We're seeing that now with DX12. /s

 

[xpea]

Yeah. We're seeing that now with DX12. /s

yeah it worked so well :thumbsup::thumbsup::thumbsup:

 

[USER8000]

yeah it worked so well :thumbsup::thumbsup::thumbsup:

It seems to be working now - the last two quarters has seen AMD desktop marketshare increase! :thumbsup:

Maybe instead of making lame excuses,can you please tell me why my Maxwell card is sucking so hard compared to the equivalent AMD one in those DX12 games and even a number of recent Nvidia sponsored,DX11 ones??

I am still waiting for the Async driver Nvidia promised for us.

Been with Nvidia for like the last three years - thought the deteriorating performance on my GTX660TI was probably a one off due to its weird memory controller.

The replacement card seems to be doing worse,quicker.

 

[xpea]

It seems to be working now - the last two quarters has seen AMD desktop marketshare increase! :thumbsup:

you proved exactly my point. this "looking forward" vision doesn't work.
After being laminated during one full year, they finally start to recover a bit of market share ? it's all bad because they lost lot of customers and when the inflection point is coming, booom next gen is here...

 

[guskline]

I have to giggle at all of the "analysis" I've been reading in this thread. I'll give my 2 cents opinion because I'm lucky if it's worth more than 2 cents
First what do you expect AMD to compare their new high Polaris to? A GTX 970? That's old (BEST Selling card but priced too low for what AMD wants to do). A FuryX? Why ravage your own brand when the successor, Vega, probably isn't ready yet (Vega won't be cheap I'm expecting $700-850 maybe more).

Why GTX 980 TI? Because it was the top end Nvidia card, other than the TitanX, that Nvidia has in the stable. Nvidia hurt AMD's release of Fury X price wise with the GTX 980 TI and for a variety of reasons, sales wise the GTX 980 TI clocked the Fury X sales.

AMD hopes it pulls the focus away from Nvidia's soon to be released Pascal cards AND gives AMD market ability to raise the price of it's highest end Polaris over $500 or near it (perhaps $499 my prediction).

Think about it. The GTX980TI is @$600 now but will probably fall in price when the Pascals drop. We can argue until the cows come home as to whether or not the Fury X is better or worse than the GTX 980 TI but seriously which of those two video cards sold more? Moreover, the GTX 970 was a sales phenomenon, despite the 3.5 vs 4 G Vram debacle.

I have not doubt the Polaris line is solid, but AMD needs a reason to justify a price increase. Telling a consumer that your mid range chip nearly equals, or perhaps slightly surpasses the competitor's high end chip that now sells at @$600 and the AMD expensive version is @$100-$150 cheaper looks, at least to me to be the strategy to price the high end Polaris at@$500 (probably $499).

Just my 2 or is that now devalued to 1 cent analysis.

 

[USER8000]

you proved exactly my point. this "looking forward" vision doesn't work.
After being laminated during one full year, they finally start to recover a bit of market share ? it's all bad because they lost lot of customers and when the inflection point is coming, booom next gen is here...

Well,after my experiences with Pascal(and my previous Kepler card),AMD would really need to have a totally rubbish card for me to go Nvidia a third time in a row now. Kind of seeing the same from mates having the GTX960 and GTX970 too.

If the current AMD cards hold their own in more and more modern games,and the GTX960 and GTX970 don't,it is going to push more people towards AMD.

You have to consider,something similar happened with the Nvidia FX series. Nvidia sold more of them than the forward looking 9000 series.

Yet,once the X800 hit,despite being technically inferior to the 6000 series,ATI shipped MORE cards than Nvidia. The bad taste from the FX helped ATI sell more cards.