Support NeoGAF

[General Lee]

300mm^2 is hardly a top end though and the problem for AMD here will be that NV will be able to scale the GP104 below to any level basically if they need to. Sure, their margins will be lower but they'll block AMD from gaining market share this way. I'm kinda wondering if that's why we'll get three SKUs on GP104 in June.

I don't know why people are back to GDDR5X isn't available when Micron stated more than a month ago that they've started shipping the chips to customers and we even know what chips these are. The supply may be limited but it is available.

It's top end for this period. But yeah, I fully expect Nvidia to always bring out a bigger die no matter what, and I don't think AMD's answer to that is to try make an even bigger one. They'll have to succeed with a superior architecture.

As for GDDR5X Micron stated a tentative mass production date vaguely for "summer", and I don't see that as any kind of assurance that a graphics card like a possible GTX 1080 could be produced in mass quantities before next autumn earliest. It could be that Nvidia is only releasing a 1070 with 8 GHz GDDR5 chips and leaving 1080 for later. Of course they also might not call it a 1070 to begin with.

 

[Bboy AJ]

Is Polaris going to be better than Pascal? That chart on the previous page makes it sure look so.

 

[Kezen]

Is Polaris going to be better than Pascal? That chart on the previous page makes it sure look so.

Based on what ?

 

[Bolivar687]

Man, I sold two Hawaii's hoping to get something big this summer. If it doesn't pan out, I'll probably just grab an open box or second hand Fury X to hold me over until Vega. Would be nice if they at least had a Fiji refresh though.

 

[dgrdsv]

But Summer 2016 clearly means June. Then there will be some offset until board manufacturers integrate all the components. It could definitely be autumn until GDDR5X is seen in retail products.

Update: Leaker also says that GTX 1080 will feature GDDR5X memory, while GTX 1070 will stick to GDDR5 standard, both using 256-bit memory bus. Cards based on GP104 GPU are to be equipped with three DisplayPorts, HDMI and DVI.

.

As I've said it makes zero sense to not use GDDR5X in the upcoming generation straight away.

It's top end for this period. But yeah, I fully expect Nvidia to always bring out a bigger die no matter what, and I don't think AMD's answer to that is to try make an even bigger one. They'll have to succeed with a superior architecture.

They don't have a superior architecture. 8B transistors GM200 is beating a 8.9B transistors Fiji. 5.2B transistors GM204 is more or less keeping up with a 6.2B transistors Hawaii. Even a 3B transistors GM206 is able to keep up with 5B transistors Tonga. If (admittedly that's a big "if") this comparative picture won't change with Polaris vs Pascal AMD will actually need bigger dies to compete with NV.

For now I'm not expecting a 232mm^2 P10 to be close to 333mm^2 GP104 in performance as judging from this generation AMD will need a bigger die to compete with it.

 

[chaosblade]

Man, I sold two Hawaii's hoping to get something big this summer. If it doesn't pan out, I'll probably just grab an open box or second hand Fury X to hold me over until Vega. Would be nice if they at least had a Fiji refresh though.

You're probably looking for Vega then. The big chips aren't coming until later.

 

[Onemic]

Is there an estimated ETA on Polaris? Around the same time as Pascal or later down the line?

 

[Inspectah_Deck]

Is there an estimated ETA on Polaris? Around the same time as Pascal or later down the line?

Same time.

 

[Onemic]

Same time.

Oh, nice. So launch right before or during Computex?

Guess I'm gonna have to wait and see some benchmarks for both before making a decision now.

 

[TSM]

Same time.

I really hope so. Bring on the aggressive price undercutting.

 

[Bolivar687]

Oh, nice. So launch right before or during Computex?

Guess I'm gonna have to wait and see some benchmarks for both before making a decision now.

The Nvidia event is before Computex, maybe mid May. AMD might be as late as E3.

 

[Nachtmaer]

Oh, nice. So launch right before or during Computex?

Guess I'm gonna have to wait and see some benchmarks for both before making a decision now.

If you're looking at the mid range, like P10 will be, you might have to wait longer than that for nVidia's counterpart. GP106 is rumored to launch later than those.

 

[Onemic]

If you're looking at the mid range, like P10 will be, you might have to wait longer than that for nVidia's counterpart. GP106 is rumored to launch later than those.

Nope I want dat high end for VR. Im either gonna be getting 2 pascal 1080 cards for SLI or 2 of the highest end polaris cards for crossfire.

 

[Trogdor1123]

Could someone explain to me why Polaris is being viewed as such an important thing? Is it just competition reasons?

 

[potatohead]

Could someone explain to me why Polaris is being viewed as such an important thing? Is it just competition reasons?

Lower power consumption and high performance per watt means it's ideal for low-mid end tier desktop GPUs, laptop GPUs, and console GPUs.

At least that's what I've understood so far.

 

[Onemic]

Lower power consumption and high performance per watt means it's ideal for low-mid end tier desktop GPUs, laptop GPUs, and console GPUs.

At least that's what I've understood so far.

Oh, is that all it is? It wont be faster than the previous architecture?

 

[potatohead]

Oh, is that all it is? It wont be faster than the previous architecture?

I think that as well, but I wouldn't know why

 

[Onemic]

I think that as well, but I wouldn't know why

Well I guess my only choice is Pascal then.

 

[dgrdsv]

Could someone explain to me why Polaris is being viewed as such an important thing? Is it just competition reasons?

It's a new generation of AMD GPUs built on a new production process. It's viewed as important as any such new generation really.

 

[Bolivar687]

Could someone explain to me why Polaris is being viewed as such an important thing? Is it just competition reasons?

Its one of the biggest node shrinks we've ever seen. AMD has been billing it as a massive leap in terms of efficiency, affordability, and performance. Although the CEO's comments about mainstream desktops and some of the SKU leaks have put the last part in question.

 

[potatohead]

Well I guess my only choice is Pascal then.

The Polaris 11 is targeting the low-middle tier, but AMD will have other stuff competing at the high-end. I think that's Polaris 10? Can't even remember already

 

[Bolivar687]

The Polaris 11 is targeting the low-middle tier, but AMD will have other stuff competing at the high-end. I think that's Polaris 10? Can't even remember already

It might all come down to your definition of "mainstream" and "high end." Here's what an underclocked early sample of the cut down P10 looked like:

http://wccftech.com/amd-polaris-10-gpu-specs-leaked/

The SiSoft entry also includes the clock speed of the card, which at 800Mhz is surprisingly low. So we’re likely looking at an early engineering sample here. Especially considering the fact that 14nm FinFET clocks much higher than 28nm. So I’d expect frequencies upwards of ~1150Mhz on the production chips.

With that in mind, the GPU actually did well in the benchmark and came ahead of the GTX Titan Black. Which features a full GK110 chip and coincidentally is faster than a 2816 GCN core Hawaii XT GPU. Also funnily enough this is very close to the performance we’ve seen from the leaked GTX 1080 and GTX 1070 benchmarks.

Despite the low clock speed & the lower number of GCN cores this particular configuration of Polaris 10 still performed ahead of where you’d expect an R9 290X to perform. This in turn indicates that 4th generation GCN cores perform measurably better per clock than their predecessors.

 

[General Lee]

One thing to consider when looking at AMD's assumed 232 mm^2 die against Nvidias ~300 mm^2 GP104 is that Glofo 14 nm should be more dense than TSMC's design, and also AMD's transistor density has always been higher, so with these combined they might end up being very close in terms of transistor counts. Historically speaking AMD has usually had a clear advantage in die sizes relative to performance, but that was diminished after Kepler was introduced and Nvidia abandoned the hot clock architecture. Before that Nvidia always used their ~500 mm^2 chip to just marginally beat a significantly smaller AMD die.

 

[dgrdsv]

One thing to consider when looking at AMD's assumed 232 mm^2 die against Nvidias ~300 mm^2 GP104 is that Glofo 14 nm should be more dense than TSMC's design, and also AMD's transistor density has always been higher, so with these combined they might end up being very close in terms of transistor counts. Historically speaking AMD has usually had a clear advantage in die sizes relative to performance, but that was diminished after Kepler was introduced and Nvidia abandoned the hot clock architecture. Before that Nvidia always used their ~500 mm^2 chip to just marginally beat a significantly smaller AMD die.

Depending on the estimation you use the difference between GloFo's 14nm and TSMC's 16nm is rather negligible. A 232mm^2 14LPE chip would have a size of 238-267mm^2 on TSMC's 16nm process - which would still be smaller than what GP104 have.

A more dense transistor packing can be a factor but with a significantly smaller die size they're unlikely to be able to fit lots more of them which is needed to perform on the same level if the same comparative situation of Maxwell vs GCN will continue in Pascal vs Polaris. More dense packing also negatively influence the frequencies usually.

Judging from die sizes we have right now I wouldn't expect P10 to perform on GP104 level. There are still a lot of speculation in all of this though so who knows.

 

[MuchoMalo]

Oh, is that all it is? It wont be faster than the previous architecture?

The common assumption is that it'll be much slower than GP104, but I think that it makes sense to wait for actual benchmarks before making that assumption. People may be expecting more out of GP104 than what we'll actually see. It seems like people are expecting it to be much faster than GM200, but that's far from a forgone conclusion.

If you want true a high-end card, those come next year from both companies. Both are releasing their midrange stuff this year. I think most people here are off and Polaris 10 and GP104 will perform about the same as Fiji and GM200 respectively, but at under 150W.

 

[Mr. Pointy]

Anyone know where the Baffin specs are? I saw an article about it being a 16/20CU part, but I don't know where it is.

 

[Bolivar687]

Anyone know where the Baffin specs are? I saw an article about it being a 16/20CU part, but I don't know where it is.

16CUs, 128-bit memory interface, 1024 of the next gen GCN cores. The drivers that leaked a while back showed six Baffin SKUs - the leaked specs are for the bottom, most cut down one.

http://wccftech.com/amd-polaris-11-gpu-specifications-leaked-compubench/

Also, apparently the leaked 232mm^2 die wasn't a leak for either Polaris 10 or 11 - it was just some resume bullet point for a guy who left in July 2015.

http://wccftech.com/amd-polaris-gpu-die-size-232-mm-2/

 

[Deleted member 465307]

16CUs, 128-bit memory interface, 1024 of the next gen GCN cores. The drivers that leaked a while back showed six Baffin SKUs - the leaked specs are for the bottom, most cut down one.

http://wccftech.com/amd-polaris-11-gpu-specifications-leaked-compubench/

Also, apparently the leaked 232mm^2 die wasn't a leak for either Polaris 10 or 11 - it was just some resume bullet point for a guy who left in July 2015.

http://wccftech.com/amd-polaris-gpu-die-size-232-mm-2/

Would all of the Polaris 10 and 11 SKUs be revealed at the same time or would AMD wait until a few months to show off new SKUs?

 

[Bolivar687]

Would all of the Polaris 10 and 11 SKUs be revealed at the same time or would AMD wait until a few months to show off new SKUs?

Hard to tell, they've said this launch will be different from what they've done in the past. But it sounds like they're trying to hit every price point at once, maybe with a few of the lower Baffin SKUs trickling out through laptops as the back to school season kicks off.

 

[Frankfurter]

16CUs, 128-bit memory interface, 1024 of the next gen GCN cores. The drivers that leaked a while back showed six Baffin SKUs - the leaked specs are for the bottom, most cut down one.

http://wccftech.com/amd-polaris-11-gpu-specifications-leaked-compubench/

Also, apparently the leaked 232mm^2 die wasn't a leak for either Polaris 10 or 11 - it was just some resume bullet point for a guy who left in July 2015.

http://wccftech.com/amd-polaris-gpu-die-size-232-mm-2/

It was from someone working at AMD and it's specifically talking about 14nm and a 232mm² die.

Edit: Development for Polaris 10 almost certainly started way before July 2015.

 

[SneakyStephan]

It might all come down to your definition of "mainstream" and "high end." Here's what an underclocked early sample of the cut down P10 looked like:

http://wccftech.com/amd-polaris-10-gpu-specs-leaked/

I guess mentioning the word titan ( titan black) sounds more click baity than just saying 'less than a gtx 970'

There is no way that gp104 will not be a lot faster than a gtx 970 btw

The wccftech clickbait you linked confirms what the other poster said, that (according to that rumor, which is hopefully false) polaris 10 is not in the same performance bracket than gp104, but rather in that of what will likely become gp106

 

[Nachtmaer]

It was from someone working at AMD and it's specifically talking about 14nm and a 232mm² die.

Edit: Development for Polaris 10 almost certainly started way before July 2015.

Yeah, it takes about 3-4 years to go from design to actually fabbing a chip. CPUs take even longer because of their complexity. It could be that this was around the time they got their first test chips back from the fab for validation and possibly tweaking. Iirc, it used to be around 6 months to go from tape-out to production (if everything goes well), but given how more problematic and expensive smaller nodes become, it might even take longer nowadays.

There's also the possibility this was just some test chip that never was meant to go into production. Projects get canned or overhauled all the time. Although, if those rumored specs (40 CUs) are to be believed, ~230mm² shouldn't be too far from the real deal. Take Pitcairn at 212mm², more or less double it, fab it on a node with about twice the density and you end up in that range.

 

[Bolivar687]

It was from someone working at AMD and it's specifically talking about 14nm and a 232mm² die.

Edit: Development for Polaris 10 almost certainly started way before July 2015.

The resume doesn't specify whether it was Polaris 11 (which has six SKUs), Polaris 10 (2) or a GPU that might never make it to market, which the article admits is another possibility.

I guess mentioning the word titan ( titan black) sounds more click baity than just saying 'less than a gtx 970'

There is no way that gp104 will not be a lot faster than a gtx 970 btw

The wccftech clickbait you linked confirms what the other poster said, that (according to that rumor, which is hopefully false) polaris 10 is not in the same performance bracket than gp104, but rather in that of what will likely become gp106

That SiSoft score is also within the range of GTX 980s on that site. I'm not sure why you keep calling it clickbait but I'm guessing you also read that this is an early sample underclocked at 800mhz. The newer leak for the same Polaris 10 SKU is now at 1050mhz, which is 30% faster, and it's expected that 14nm allows for much higher clocks than that. Likewise, you surely also read there are at least two SKUs for Polaris 10 and the leaked samples we keep hearing about is the lower, cut down version. Obviously nothing is set in stone, this is all rumors and speculation, almost all of which is trending the other way than what a lot of posters here are trying to portray.

 

[dgrdsv]

AMD Polaris 10 GPU To Offer Near 980 Ti Performance For 299 USD?
Granted we've heard this before from them but it would cool if this will turn out to be true.

 

[Kezen]

AMD Polaris 10 GPU To Offer Near 980 Ti Performance For 299 USD?
Granted we've heard this before from them but it would cool if this will turn out to be true.

It would certainly entice Nvidia to price their cards more competitively.

 

[kami_sama]

AMD Polaris 10 GPU To Offer Near 980 Ti Performance For 299 USD?
Granted we've heard this before from them but it would cool if this will turn out to be true.

If this is true (I doubt it is, though), it will be the card for VR.

 

[Renekton]

AMD Polaris 10 GPU To Offer Near 980 Ti Performance For 299 USD?
Granted we've heard this before from them but it would cool if this will turn out to be true.

IINM that rumor started from WCCFTech, so very unlikely to be true.

 

[icecold1983]

AMD Polaris 10 GPU To Offer Near 980 Ti Performance For 299 USD?
Granted we've heard this before from them but it would cool if this will turn out to be true.

not buying it

 

[Nikodemos]

AMD Polaris 10 GPU To Offer Near 980 Ti Performance For 299 USD?
Granted we've heard this before from them but it would cool if this will turn out to be true.

As a resident rabid AMD fan

Spoiler

atic

, I'll rate this an "exceedingly unlikely"/10.

 

[ThoseDeafMutes]

Pray for it to be true, plan for it to be false.

 

[GHG]

AMD Polaris 10 GPU To Offer Near 980 Ti Performance For 299 USD?
Granted we've heard this before from them but it would cool if this will turn out to be true.

Selling my 980ti.

Brb.

 

[hooijdonk17]

All I want is Polaris 11..

 

[mrklaw]

I think its time I started to dismiss this entire year as a low-mid range warm over while both companies get to grips with 14nm production before the proper chips next year. That has the benefit of helping me not be too disappointed if true, and being nicely surprised if they exceed expectations.

AMD are talking way too much about efficiency and performance per watt to think they have a big chip coming, and even on the Nvidia side people are hoping for maybe something a bit more powerful than 980Ti for a bit less money.

 

[dgrdsv]

As a resident rabid AMD fan

Spoiler

atic

, I'll rate this an "exceedingly unlikely"/10.

Weeeeeell... If P10 is Hawaii on 14nm with 256 bit GDDR5X then technically it may somewhat reach Fury's performance if it'll be clocked high and the architectural optimizations will give it some 10-20% on average. Fury is some +10-15% of performance to 390X, not a gap that can't be closed by a better architecture with higher clocks.

Reaching 980Ti is a different matter however, especially if we're assuming that it's factory OC 980Ti cards which it needs to reach.

What I don't buy is the $300 price though. If it'll beat Fiji and they'll price it on $300 then what will they sell between $300 and $1500?

 

[V_Arnold]

What I don't buy is the $300 price though. If it'll beat Fiji and they'll price it on $300 then what will they sell between $300 and $1500?

Can I say it? Should I say it?

NVIDIA CARDS!

Spoiler

Okay, that was predictable

.

Anyway, they can have dual-GPU setups for that pricerange or something that....that actually succeeds the 980ti?

 

[Bl@de]

AMD Polaris 10 GPU To Offer Near 980 Ti Performance For 299 USD?
Granted we've heard this before from them but it would cool if this will turn out to be true.

I don't believe it ... but I hope it's true. 980 ti performance for 299.- would be exactly what I'm looking for. Would switch to AMD again with a card like that (unless Nvidia has something very similar).

 

[Ozorov]

Pray for it to be true, plan for it to be false.

This

 

[Bolivar687]

After the way it was introduced at CES, the Hitman demonstration at GDC, the leaked benchmarks for the smaller Polaris 10, and everything RTG has said about the market position for the lineup, I'm more surprised people keep insisting it won't happen.

 

[Irobot82]

I own a 7950, so in the end, I WIN when I upgrade.

 

[demented waffle]

980TI performance for 300 would be sweet. That would actually get me to upgrade from my 290x this year. I'll believe it when I see it tho.