AMD Polaris 10 "Ellesmere" as Fast as GTX 980 Ti: Rumor

There's a frightening thought.

 

Well better tune down the hype. Else if they price this at 550 or even 500 some people will be disappointed with it then somehow it will be regarded as meh by many people lol.

Personally i imagine AMD will want to make profit as many as possibble. So the best thing for amd to do is wait for nvidia to launch first (to set the price point) and then try to counter nvidia offering appropriately. Maybe they want to undercut nvidia a bit. If they undercut nvidia too much then nvidia might retaliate with with price war. With financial situation of both company which one have more 'ammo' for long term price war?

 

We could use a price war for a change...

 

$300 is just too low, even if the chips literally cost 1/2 in materials to make, all other expenses remain the same. $450 sounds more like it and even that is low.

 

If the rumored new PS4 uses the same Polaris chip as well as it's current "APU", then it might be possible that it does cost very little, if they have a target sales figure that is already made thanks to Sony. Might give AMD fans a reason to pick up the new PS4, too. Win-win. AS I have said before, it wouldn't be the first time such a console design was sold.

But then, I smell a hint of guerilla reverse-marketing.

 

Well, its still cut down some, so perhaps at nearly 980Ti speeds cut down they have and are binning cores for a faster $599 part, but I agree, perhaps $399 for the midrange card, $499 for the mid-high, and $299 for the mid low end card. Afterall they do have to compete with a 980Ti in performance and price, and they are currently around $650 so a card that performs as well within a few percent and costs $100 less seems about right.


If they are releasing in June, the chips are off the fab line and to the PCB line already. Wonder if there will be anymore leaks before then?

 

If it can match 980ti performance it will surely not be $300. Especially when you look how unimpressive the 1080 looks on paper. This actually looks like a good idea for AMD to delay the Polaris 10/11, and let nV set the price first.

 

Consoles have never been top end parts and they never will be. The volume of sales of a console is too high for it, the number of high end chips at a good yield is too low, and both these facts drive up the price. The only unique selling point this current gen of consoles really have is that they are *much* cheaper to get into than they were previously. Ergo, the price can't be too high. The whole reason the current PS4 is what it is, is because AMD had to burn old HD stocks. The actual price of a 7870 at launch of the PS4 was around the 150-200 dollar mark. No way they will double that (and even then you're not getting into 980ti territory).

We are going to be looking at upscaled 1080p probably.

 

980 Ti performance-ish levels for ~$300'ish? I'll eat my shorts if that's true. Then order at least two of those cards.

 

PS4 has a custom SoC, i.e. the cpu, gpu etc share in the same silicon...


On a side note, AMD getting the console deals (for the current and next gen) and pushing with mantle was the worst possible thing that could have happened to nvidia. With low level apis etc and the need to extract every bit of performance of a console, developers will make games aggressively optimized to multi threading and GCN architecture using the low level apis. Given the state of the console ports to pc, nvidia will start to struggle in a not distant future... Keep in mind that the console market generate several times higher profit than pc. If you look at this AMD will have nearly 90% of the gaming market if you take into account consoles... It makes sense to develop for AMD's hardware doesn't it?

 

I don't see it to be true. With the release of their dual gpu at 1500 bucks and its not even Polaris based.
Nawh, not gonna happen.

 

Exactly, the problem here is their about to launch a dual Fiji card for $1500.... although i guess having that topped by a couple of $300 cards would be funny.

If it offers GTX 980 Ti performance (which is sweet BTW ), then it's gonna be priced accordingly.

 

Well, while I am not one to believe that is 100% accurate it would not be a stretch for somewhere in the $350 range for the R9 490 depending on how they are doing the cards this round. If there are to be at some point an R9 Fury (Whatever it is to be named this round) level card after the 490 and 490X, it would not be a stretch to see a price around there with the 490X being around $500. Though I have my doubts its not a stretch to believe they would do that since this round they kept the other cards in the lineups separated by around $50 increments and pretty low.

Depends how they price the next flagship and how much off the 490 is from the 490X. But they could do it like they did with the 290 and 290X where the difference is less than 10% yet the price is much lower.

 

Another website trying to get free hit's on his page.
Since when AMD even shares their final pricing with their AIBs 2 months before the announcement? Not even before 2 wks, they generally give a reference price 1 or 2 wks before so AIBs .
doesn't even make sense as they also are waiting for nvidia to reveal their hand (pricing) as they are planning to announce their new stuff before AMD.

 

4th generation GCN is not going to have large changes except for the shrinking itself, so I'm wondering how ~2.560 of these cores can outperform 4.096 cores from Fiji while still operating at a lower frequency. It would be nice, but I'm not convinced.

Does anyone remember how the Fiji rumors turned out? It gave GTX 980 Ti a beating and was a great overclocker, oh wait...

It probably just means that it might in 1/20 cases come somewhat close. It wouldn't surprise me if it's 30% slower at average.

Do you know why AMD is so "Async loving" and getting up to 20% improvement in Ashes of the Singularity? It's because ~40% of the GPU is not actually doing anything because of bad scheduling. Remember that Fiji has 53% more computational performance than GTX 980 Ti, and still it's generally beaten by GTX 980 Ti. The whole point of async shaders is to utilize different resources in the GPU for different parallel workloads, and AMD clearly gets a boost from threads using the same resources which tells us that they don't utilize it well in the first place. So AMD is in essence "misusing" async shaders to compensate for a bad architecture. For comparison, Maxwell is pretty much at 100% percent utilization so doing several computational intensive threads at the same time would not give any benefits.

And the reason why AMD get's a bigger "relative boost" from Direct3D 12 is that Nvidia decided to apply the Direct3D 12 driver side optimizations to all API versions, which in turns results in little relative gain between version 11 and 12. Everyone who follows the technology should know this (1)(2). I don't know if people just fall asleep when we get to the technical details, or if people suffer from "selective memory loss".

 

Here's the Line-up:

- R9-470 (Polaris11-Pro, 4GB GDDR5, R9-380-Performance, $180)
- R9-470x (Polaris11-XT, 4GB GDDR5x, R9-380x Performance, $230)

- R9-480 (Polaris10-Pro, 4GB GDDR5, R9-390/R9-290x performance, $270)
- R9-480 (Polaris10-Pro, 8GB GDDR5x, R9-390x/Fury performance, $350)
- R9-480x (Polaris10-XT, 8GB GDDR5x, Fury-X performance, $420)

Vega will be out around March of 2017, and AMD will respin Polaris with MUCH higher ClockSpeeds.

- R9-570 (Polaris11-Pro, 4GB GDDR5, R9-290-Performance, $200)
- R9-570x (Polaris11-XT, 4GB GDDR5x, R9-290x Performance, $250)

- R9-580 (Polaris10-Pro, 8GB GDDR5x, +20% Fury Performance performance, $370)
- R9-580x (Polaris10-XT, 8GB GDDR5x, +25% Fury-X performance, $450)

- R9-590 (Vega-Pro, 8GB HBM2, +70% Fury-X performance, $600)
- R9-590x (Vega-Pro 16GB HBM2 +85% Fury-X performance, $800)

- R9-Rage-X (Vega-XT, 32GB HBM2, 2.3x Fury-X performance, $1100)

 

Rumour is Polaris 10 die size around +200mm² and Polaris 11 to be around half the size. I think the die size being half the size over NVidia GTX 970 398mm² if true, and AMD wanting to win market share back, we could see the price allot lower than we expect.

GP104 said to be around +300mm²

Edit R9 390 and GTX 970 is around 240 in Europe. AMD said they will target this market, but we dont know with what card. Am dreaming its Polaris 11

 

I think you are closer to reality, but overshooting both price and performance.

The R7-470 is reportedly coming to laptops before desktops, so we may not see it until later in the year. At <50W, it will not be as strong as a R9-380, but somewhere between the 370 and 380 and be priced <$150.

The R9-480 will perform a bit below the R9-390 and GTX 970, and will be priced ~$250. This one will come out this summer, and force modest adjustments in older cards in this range.

Pascal cards introduced this summer (1070 and 1080) will be higher end and won't compete directly with Polaris. The 1070 will compete with Fury though, and force some price adjustments.

So, that's my crystal ball. No big shakeups or big price drops.

 

I remember the last GPU launch. Probably $300 MSRP, $500 street price on launch day with it coming down very slowly over time.

 

In all likelihood these comparisons are in DX12 where we know the 390 & 390X can compete with the 980 TI and they are currently in the $350-$450 range.

 

It's highly unlikely that AMD or Nvidia will sell FuryX/980TI-level cards for less than $400. My guess is that the Polaris XT is close to FuryX/980TI performance wise and will sell for around $450, whereas the Polaris Pro is close to the 390X/Nano and will sell for ~$350. Performance per Watt of Polaris is rumored to be 2.5X that of current AMD's cards. At 135W, we should see +300W performance from this generation, i.e. FuryX performance.

 

I wouldn't extrapolate to much based on PR exaggerations. Maybe 2.5x from absolute worst case to best case. It simply makes no sense for performance to jump that much in one generation. They would have called it 490 instead of 480 if there was a 2x performance increase over 380.

 

I highly doubt the first production chips at 14nm will be anywhere as cheap as the massive times that 28nm has had on market to drop in price, there just isn't anyway that Global Foundries would've licensed all that IP from Samsung (not for free) to sell things (especially the first batches of the ramping yield 14nm) at any "great" price.

 

I believe the perf per watt could jump like that, because of the 28nm was tapped out, if Fiji was on 14nm it'd even have relative jumps like that; unfortunately the ram max in Fiji and poor overclock hampered any major jumps....AMD overpromised with Fiji overclock....IDK why, especially when their shareholders are sue happy about everything....hell they will sue about what they consider a "core" these days.....hell "cores" didn't always have FPU's on them until in the recent decade or so, shared FPU's aren't sight unseen.