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

[linkgoron]

Polaris 10 can be a great chip if it offers good performance/power/price. It doesn't have to beat a 980 Ti to be worthwhile; an R9 390X replacement at a lower price & better power consumption would be a good product that, depending on what NVIDIA has in the market at the time, could be very much worthy of recommendation.

Yes, I should've phrased myself more carefully. The fact is that Polaris 10 isn't bringing anything new with regards to performance, and will probably be weaker than Fury X, and the common forum user (or at least this is correct for me) wants to be blown away by performance.

If you're a 970/980/290/390 (x)/Fury owner, upgrading to Polaris would probably be more of a side-grade. Pascal will probably be a much better upgrade path for more performance. I'm sure that 290x users are itching for an upgrade, and if you're looking for something significant from AMD you have to wait at least 5/6 more months for Vega.

Polaris will probably be a great upgrade path for 380x/960 (or lower), htpc (SFF?) and mobile, however it is less exciting (IMO) and that's why rumors have been slowly inflating Polaris performance numbers.

 

[moonbogg]

That sounds very elitist.

Thanks man. I happen to know that I am actually pretty elite based on the 980ti's in my sig. That's not me talking. Its the market data. They do make me feel pretty E-buff though.

Where did I mention AMD? And tbh, am I wrong? Someone who IS faithful to the nV brand would probably upgrade to Pascal wouldn't they? And the same applies to the AMD faithful.
My response was only to poke fun of the insinuated lack of driver support for older cards in the post that I responded to.
You should buy whatever card ticks all your boxes

Not unless they are a significant upgrade to what I already have. Also, not now anyways because I can't afford to waste money on mid range garbage. I am elite, but only when I can afford to be elite in FULL BLOWN FASHION. None of this mid range crap for me any more. If I can't buy the good stuff, then I won't buy anything. I'll wait, just like I'm waiting for my super elite Skylake-E.

Absolutely. Lowering the price of 980Ti performance to $300-$350 would be outstanding. Many folks who can't or won't leave that price bracket will be very happy.

I'd prefer to keep the riff raff in the sub $400 segment, so this works out nicely for me. If you aren't elite, then stay off my ELITE LAWN!


jk guys calm down love you all

 

[xthetenth]

Exactly this. So much this! Haha...

I've always wanted to take two people rabidly arguing over one gpu over the other because its "20%" faster, take them and put them in front of two beige boxes, not allowed to check settings, and tell me which one is "better".

If a card gets you within a legit 20-25% of another card, its the same gaming experience. I could never fathom paying hundreds of my hard earned dollars additional to go from 48 FPS to 56 FPS but that's just me. If the Polaris 10 gets us within striking distance of a 980 ti for $300 its a smash hit in my book.

That's a point and what finally got me to go in on freesync. It'd have to be awful for me to actually get noticeably slower performance.

You said that the "Nv faithful" (implication: "stupid, brand-loyal fanboys") are going to buy Pascal, as though Pascal isn't going to be a product worth buying on its own merits.

A -> B != B -> A

If all the brand loyal fanboys buy Polaris, that doesn't mean everyone buying Polaris is a brand loyal fanboy. There's room for people who are spendy and know it, and have built their budgeting and part picking around it.

 

[airfathaaaaa]

Yes, I should've phrased myself more carefully. The fact is that Polaris 10 isn't bringing anything new with regards to performance, and will probably be weaker than Fury X, and the common forum user (or at least this is correct for me) wants to be blown away by performance.

If you're a 970/980/290/390 (x)/Fury owner, upgrading to Polaris would probably be more of a side-grade. Pascal will probably be a much better upgrade path for more performance. I'm sure that 290x users are itching for an upgrade, and if you're looking for something significant from AMD you have to wait at least 5/6 more months for Vega.

Polaris will probably be a great upgrade path for 380x/960 (or lower), htpc (SFF?) and mobile, however it is less exciting (IMO) and that's why rumors have been slowly inflating Polaris performance numbers.

 

[FatherMurphy]

Yes, I should've phrased myself more carefully. The fact is that Polaris 10 isn't bringing anything new with regards to performance, and will probably be weaker than Fury X, and the common forum user (or at least this is correct for me) wants to be blown away by performance.

If you're a 970/980/290/390 (x)/Fury owner, upgrading to Polaris would probably be more of a side-grade. Pascal will probably be a much better upgrade path for more performance. I'm sure that 290x users are itching for an upgrade, and if you're looking for something significant from AMD you have to wait at least 5/6 more months for Vega.

Polaris will probably be a great upgrade path for 380x/960 (or lower), htpc (SFF?) and mobile, however it is less exciting (IMO) and that's why rumors have been slowly inflating Polaris performance numbers.

I think this is probably right and goes to the heart of the recent statements by Roy Taylor and Raj that AMD is targeting the 970/290 level of performance, or a bit over that, at a lower, mainstream price and lower power point, and that, to paraphrase Roy Taylor, "all we've heard about Pascal is that it is a high-end chip between $600 and $1000."

The signs point to these chips competing in entirely different performance brackets. If that is the case, then the compelling question is will Nvidia price their higher performing GP104 variants competitively on a $/performance metric. Or (more likely), will Nvidia price its higher performing chips with a premium because there will not be much, if any, competition (at least for the higher end variant of GP104 with the speculated GDDR5X memory).

 

[Orvogg]

Yes, I should've phrased myself more carefully. The fact is that Polaris 10 isn't bringing anything new with regards to performance, and will probably be weaker than Fury X, and the common forum user (or at least this is correct for me) wants to be blown away by performance.

If you're a 970/980/290/390 (x)/Fury owner, upgrading to Polaris would probably be more of a side-grade. Pascal will probably be a much better upgrade path for more performance. I'm sure that 290x users are itching for an upgrade, and if you're looking for something significant from AMD you have to wait at least 5/6 more months for Vega.

Polaris will probably be a great upgrade path for 380x/960 (or lower), htpc (SFF?) and mobile, however it is less exciting (IMO) and that's why rumors have been slowly inflating Polaris performance numbers.

Thats allright. I can't tell you what to do with your hard earned cash and I can certainly not speak for every 290 owner out there, but I'd say that chances are that if one is still using a 290 that is because there is simply nothing worth while to upgrade to. And if nV does as nV has done in earlier generations, the price for new hardware will still be such that there is nothing interesting to upgrade to.

But mark my words, this crowd might get blown away by Polaris anyways, only they play the long game, Thus Polaris may serve a big role in selling Vega

 

[thesmokingman]

Sooo glad I didn't dump my 290 lightnings. I can soldier on now till Vegans drop.

 

[moonbogg]

Sooo glad I didn't dump my 290 lightnings. I can soldier on now till Vegans drop.

If I was Vegan I would also drop in about 6 hours due to starvation.

Also, I find it funny that people place the 290 and 970 in the same performance category. What the hell happened to 290 being compared to 780? Wasn't it compared to the 780? Not a single person has mentioned ANY Kepler card during these discussions. Not even once I don't think. Will Maxwell be just as forgotten in 6 months?
You have to admit how incredible that is. Its like the entire Kepler family of GPU's are absolute trash and totally incapable of playing any modern games. That's how it seems because people don't even include them in discussions for the sake of useful comparison.

 

[xLegenday]

Is he serious? Is Roy Taylor really claiming NVIDIA won't have mid-range and low-end Pascal chips, too?

There have been spy-shots of GP106 and obviously there is GP104 coming too which isn't going to go into a $1k Titan card.

By the way, kinda funny coming from a rep of a company that just released a $1500 card for VR. LOL!

Precisely , Roy sometimes should keep his mouth shut sometimes.
But I guess he's just doing it's job, obviously they had to have in consideration Nvidia will replace 970 at some point and 970 replacement for sure will target 980+ performance.

Now the question is the price.

 

[Mahigan]

I think this is probably right and goes to the heart of the recent statements by Roy Taylor and Raj that AMD is targeting the 970/290 level of performance, or a bit over that, at a lower, mainstream price and lower power point, and that, to paraphrase Roy Taylor, "all we've heard about Pascal is that it is a high-end chip between $600 and $1000."

The signs point to these chips competing in entirely different performance brackets. If that is the case, then the compelling question is will Nvidia price their higher performing GP104 variants competitively on a $/performance metric. Or (more likely), will Nvidia price its higher performing chips with a premium because there will not be much, if any, competition (at least for the higher end variant of GP104 with the speculated GDDR5X memory).

I think you're not understanding what Roy is saying. He's saying that they're targeting the 390 market with a product that performs like a FuryX/GTX 980 Ti.

Meaning that AMD wants to bring GTX 980 Ti performance at a $349 price point. That's with Polaris.

Vega will target the high end but it won't arrive until Nov 2016-Jan 2017.

 

[Kenmitch]

Will Maxwell be just as forgotten in 6 months?

Nope Agent 86 is immortal.

 

[DooKey]

I think you're not understanding what Roy is saying. He's saying that they're targeting the 390 market with a product that performs like a FuryX/GTX 980 Ti.

Meaning that AMD wants to bring GTX 980 Ti performance at a $349 price point. That's with Polaris.

Vega will target the high end but it won't arrive until Nov 2016-Jan 2017.

The high end release is the battle of the HBM2 manufacturers.

 

[FatherMurphy]

I think you're not understanding what Roy is saying. He's saying that they're targeting the 390 market with a product that performs like a FuryX/GTX 980 Ti.

Meaning that AMD wants to bring GTX 980 Ti performance at a $349 price point. That's with Polaris.

Vega will target the high end but it won't arrive until Nov 2016-Jan 2017.

No, you are misunderstanding him. You are focusing on the performance level of 980ti/Fury. That is not what Roy is talking about. Roy is unambiguously talking about the install base (TAM) of VR-capable cards, which, at a minimum is 970/290 performance. He expressly dismisses the notion of expanding the TAM by introducing higher performing cards by addressing the Pascal rumors (note: is he spending time on these forums?) of higher price (and higher performance).

And his argument makes sense. Folks (at least in the US) have had access to VR-capable cards for the $350 you propose for 2 years. But, as Roy notes, that price level has only result in 7.5 million VR-capable cards being on the market. AMD's solution, per Roy's statements, is to sell at least the minimum performance needed for VR at a lower price. This will expand the market to lower price points. Releasing a $350 Polaris that performs like 980ti does not provide access to the majority GPU customers who don't buy $350 or more GPUs and, therefore, does not expand TAM for VR.

So, while Polaris may offer performance comparable to 980ti/Fiji (that's the current rumor that serves as the title of this thread), that is not at all what Roy is saying.

 

[digitaldurandal]

The card is such a small part of the VR TAM barrier right now.

I would be interested to see data on how many users with Polaris purchase VR.

I don't think most users who want VR and are upgrading their GPU are going to go with Polaris. There MAY be a group of VR users in the future that bought Polaris for other reasons and later decided to upgrade.

Currently VR Hardware is between 600 and 800 dollars plus the GPU. One of them requires you to setup a space in a room and bolt hardware to the walls. If I undertake that expense I am not going to undermine it by going cheap on the hardware.

I think most VR users are spending 450 or more on their GPU easily. If I was looking to get into VR ASAP I would look at Pascal. Sorry Roy Knowing how these things go though, I will wait at least a year for v2 of VR hardware and to see how everything plays out with the various headsets.

Maybe I am not representative of the TAM though... so.. we shall see.

 

[FatherMurphy]

@digitaldurandal

You are absolutely right. The entry cost for VR is, what, well in excess of $1000 (the VR unit and the PC powered by a 970/290 or above card). I think Roy is, in effect, saying to potential VR customers "we're going to make the GPU an even smaller price consideration" for VR. And that helps AMD sell more GPUs (and gaming CPUs... ahem, Zen...). So, AMD is serious about VR and stands to profit from it, but to do so, they have to sell GPUs in the mainstream price target ($300 tops?) that are VR-capable.

**EDIT: Correct me if I am wrong, but my impression is that, at least initially, VR software is generally targeting the minimum hardware level, similar to consoles. This ensures a consistent VR experience while offering software to the largest possible installed base. So, from a VR standpoint (and that's what Roy was talking about), there is relatively little benefit to purchasing a GPU that performs much higher than the minimum level.

 

[cytg111]

So, AMD is serious about VR and stands to profit from it, but to do so, they have to sell GPUs in the mainstream price target ($300 tops?) that are VR-capable..

- That part is not entirely clear to me? Drop a ~970 card at half price "cause we will sell more cards that way"? - Does not sounds like good business to me.
Maybe a play to force VR adoptation cause they will win it back on VR console socs later? I dunno..

 

[hawtdawg]

omg! a GPU company releasing a mid-ranged GPU that's as fast as the previous generation top GPU? Impossible! Even if it's happened every single GPU generation for the last 15 years new node or not, you'd all be crazy to think it would happen again!! It's not like we have the 390, 280, 7870, 6870, 5770, or 4770 as evidence that this would be possible!

Mad I tell you! you're all mad!

 

[USER8000]

sooo glad i didn't dump my 290 lightnings. I can soldier on now till vegans drop.

if i was vegan i would also drop in about 6 hours due to starvation.

lmao!!!

nope agent 86 is immortal.

lol.

 

[Mopetar]

Not sure if AMD can get Fury performance out of Polaris 10.
Tops at 390X.. :\

They showed a limited demo of Polaris 10 where it had better performance than a 390X, so we already know it's at least that good. It's possible that it might be able to hang with Fury X at 1080 or maybe even 1440, but if you just look at the SP count difference of 2560 compared to 4096, Polaris 10 would need some combination of 60% (realistically it wouldn't be quite as much as fewer SPs are easier to utilize) architectural improvements or increased clock speed to provide matching performance, and that's ignoring any memory bandwidth issues that would stem from Polaris only having GDDR5.

It may have been possible for AMD to design a GPU that could very well have SPs that're essentially 60% more effective through some combination of architectural improvements and increased clock speeds due to the node change, but we already know that AMD has been pushing performance / watt as the main selling point, which suggests that although we may see improved clocks and a more powerful architecture than previously, a lot of the design focus went towards efficiency, which isn't necessarily a bad thing either, but you typically can't gain one without sacrificing the other, or some other thing in exchange.

 

[swilli89]

omg! a GPU company releasing a mid-ranged GPU that's as fast as the previous generation top GPU? Impossible! Even if it's happened every single GPU generation for the last 15 years new node or not, you'd all be crazy to think it would happen again!! It's not like we have the 390, 280, 7870, 6870, 5770, or 4770 as evidence that this would be possible!

Mad I tell you! you're all mad!

I made quick chart to display what's been going on (atleast for AMD's mid range releases) as well as to show what we might end up with in Polaris 10 as far as its relative value is concerned:

Assumptions for Polaris 10:
$299 SKU
~ Fury X performance

Here is the data for the chart:

Polaris 10 looks to be the best improvement in price/performance over the previous gen since 4770. The reason this isn't surprising? Both were on node jumps. 4770 was the first chip on 40nm and Polaris 10 will be the first on 14nm (two node jump).

Some other takeaways:

• 380X looks pretty good here in comparison to 290X as well, but it had two full years to release! Most of these other mid range chips came out 6-12 months after the preceding flagship, Polaris will be about a year after Fiji.
• 280X is simply a rebranded 7970 on the same process, AMD was feeling the sting of an old 28nm process here.
• 7870 was an awful deal comparatively. However its GCN architecture prevailed and this chart would look much different today so perhaps it wasn't such a bad investment.
• The Polaris 10 die size reduction is the largest ever for AMD, this DEFINITELY will help them price aggressively.
• Its also not really fair to compare to 4770 as that was when AMD was coming off an odd year with the 3870 due to its very small die, it was almost a midrange chip itself as it was priced at $219 compared to the 2900xt at $399. More telling, the 3870 was 192mm compared to the 2900xt's 420mm! This totally invalidates the numbers for the 4770 as it is a total isolated case. The 3870 was a true midrange chip so 3870 to 4770 we are looking at two generations of mid-range cards

These facts above, together with the assumptions I stated, has the potential to make Polaris 10 the most compelling mid-range card AMD has ever released.

 

[Irenicus]

Yeah... typical hype. Let's see and if AMD doesn't deliver, what a fail? lmao

I'll forgive your knock on AMD since your avatar alludes to the greatest superior DOOM alternative game style that never caught on. Magic Carpet, the only true god game ever made, that made all doom games look like pacman in comparison.

 

[SimianR]

No one really knows for sure where this thing will land performance wise, but if I had to guess I'm going to lean closer to 390 performance than 980 Ti. I think 390 - 390X performance. I guess we'll see.

 

[swilli89]

No one really knows for sure where this thing will land performance wise, but if I had to guess I'm going to lean closer to 390 performance than 980 Ti. I think 390 - 390X performance. I guess we'll see.

Well don't forget we are looking at up to 3 SKUs based on Polaris 10. I think the lowest SKU will be around a 390 but at $199 and the higher end SKU at $299 nipping at the heels of 980ti/FuryX.

 

[RussianSensation]

Well don't forget we are looking at up to 3 SKUs based on Polaris 10. I think the lowest SKU will be around a 390 but at $199 and the higher end SKU at $299 nipping at the heels of 980ti/FuryX.

Not one of the comparisons in the chart you provided Can be used for 2 reasons:

1) Fury X is the largest flagship ever from AMD. At no point in ATI's or AMD's history could a 220-240mm2 new node chip reach the last gen's flagship when the flagship was 596mm2. Even the move from highly inefficient VLIW-4 40nm 6970 to 28nm GCN 7870 allowed 7870 to barely beat a 6970. However, 6970 was only 389mm2, not 596mm2.

2) Fury X uses the most efficient memory controller and memory type in AMD's entire stack. The HBM1 memory controller alone is smaller than a typical GDDR5 controller AMD has built since 2012. That means unlike all those next gen mid-range cards, Polaris 10 will have a far inferior memory bandwidth compared to Fiji. This is not even debatable because even if Polaris 10 has 100% memory efficiency, since the fastest GDDR5 is 8Gbps, it will never even touch Fury X's real world 387 GB/sec throughout.

In addition, Fury X probably saved 20-30W alone due to the AIO CLLC. That means an air cooled Fury X is not a 280W but a 300-310W card. That means an air cooled Polaris 10 would have almost 3X perf/watt, yet AMD just officially revised it down to 2X perf/watt:

"AMD demonstrated its “Polaris” 10 and 11 next-generation GPUs, with Polaris 11 targeting the notebook market and “Polaris” 10 aimed at the mainstream desktop and high-end gaming notebook segment. “Polaris” architecture-based GPUs are expected to deliver a 2x performance per watt improvement over current generation products and are designed for intensive workloads including 4K video playback and virtual reality (VR). via AMD"
http://wccftech.com/amd-polaris-10-desktop-polaris-11-notebook-gpu/

That sounds like AMD's 2.5X claim is more UP TO not on average.

Polaris 10 could reach Fury X but not at 110W, maybe at 150-175W and in select games of peasant 1080p resolution where Fury X is CPU bottlenecked even by a 4.9Ghz I7-4970K.

Polaris 10 with 2304 SPs and 911mhz clocks is also rumored for PS4 Neo. If Polaris 10 had 30-40% higher clocks than Fury X while maintaining high efficiency, why in the world would Sony give up a whopping 'free' 300-400mhz when today's PS4 already used close to 180W of power? The answer is because a 110W Polaris 10 likely has to be clocked low to maintain the power efficiency. Even the leaks had it 850-1050mhz.

Finally, the biggest red flag of them all. If Polaris 10 is sooooo amazing, then Polaris 10 Radeon Pro Duo 16GB 350W TDP with air cooling could easily cost $999. So instead AMD spent $$$, time and designed a complicated custom made cooling to launch a $1499 card April 26th? What's the logic in that? Radeon Pro Duo is being pushed for VR development.

What's better for VR development of next gen games? A next gen GCN 4.0 architecture, with the latest features, 8GB of VRAM per GPU, and much more affordable price or the $1.5K Radeon Pro Duo?

It's the same as NV releasing a water cooled GTX580 Pro Duo, but then 1.5 months later here is a GTX690 (Polaris 10 x2).

 

[JDG1980]

Fury X is the largest flagship ever from AMD. [...] Fury X uses the most efficient memory controller and memory type in AMD's entire stack.

Fury X is also severely bottlenecked by an inadequate front-end. Despite having 45% more shaders than Hawaii, it outperforms Hawaii by only 25% even at 4K, and a measly 16% at 1080p.

Polaris 10 could reach Fury X but not at 110W, maybe at 150-175W and in select games of peasant 1080p resolution where Fury X is CPU bottlenecked even by a 4.9Ghz I7-4970K.

There is absolutely no reason for AMD not to release a Polaris 10 SKU with a 150W TDP. That still fits within a single 6-pin connector. People with old/weak PSUs or doing OEM upgrades care about that, but they don't care about the difference between 110W and 150W.

And even if AMD just leaves that headroom on the table in their official releases, nothing stops AIBs from adding it back. Of course, that would hurt AMD in the ranking charts which mostly use reference cards, so AMD would be foolish to do this.

Polaris 10 with 2304 SPs and 911mhz clocks is also rumored for PS4 Neo. If Polaris 10 had 30-40% higher clocks than Fury X while maintaining high efficiency, why in the world would Sony give up a whopping 'free' 300-400mhz when today's PS4 already used close to 180W of power? The answer is because a 110W Polaris 10 likely has to be clocked low to maintain the power efficiency. Even the leaks had it 850-1050mhz.

We have no clear idea what the final shipping clocks of either Polaris 10 or the PS4 Neo APU will be. What we do know is that FinFET has enabled substantial (~40%) clock gains over planar in products as disparate as Apple's A9 SoC and Nvidia's GP100 computing chip. That corresponds with the official literature which states that FinFET enables either substantial power reduction at the same clocks, or substantially increased clocks at the same power consumption.

Finally, the biggest red flag of them all. If Polaris 10 is sooooo amazing, then Polaris 10 Radeon Pro Duo 16GB 350W TDP with air cooling could easily cost $999. So instead AMD spent $$$, time and designed a complicated custom made cooling to launch a $1499 card April 26th? What's the logic in that?

AMD has a lot of Fiji chips they need to get rid of. (They're even throwing in a third chip as a freebie!) I think what you're overlooking is that AMD's GPU "strategy" for the 2nd half of 28nm wasn't really a strategy at all, but a lot of confused flailing around. Almost certainly, Tonga and Fiji were supposed to be on 20nm, but when that failed, AMD had nothing to fall back on. They had to be hastily back-ported to 28nm with mediocre results, while the rest of the lineup was filled with straight rebrands. They still needed to do Fiji so they could gain experience with HBM, and that meant they needed to come up with the best way possible to pitch the chips in their inventory. Note that the 7990 wasn't released until about 16 months after the 7970, and the R9 295 X2 not until 6 months after the R9 290X. Even if it only took AMD an additional 6 months to validate a Polaris 10 dual-GPU card, that would mean a release date around Q1 2017, and by that time Vega would make it obsolete. Better to go with what they already engineered, and use Radeon Pro Duo to blow out the remaining Fiji stock to the select class of customers who can make use of its high shader power and aren't as bothered by the other drawbacks.