Polaris 10 vs R9 390/390X @ GFXBench

[tential]

QFT, they were claiming it would trump a 980 TI for $299, now they are in denial mode. OTOH I don't care about these leaks, it will only matter when they are in the hands of professional reviewers.

I'll take it one step further. Due to people getting free samples, and not wanting to rock the boat and continue to get free samples, the real performance challenge is a month or two or three after release when people are doing their own testing, and not regurgitating what the reviewer guides told them to do.

 

[Jaydip]

i would ask for real facts on that post but i know you wont find any simply because you are getting a rumor and spreading it like a virus to suit your views kinda like the ones that kept spamming "pascal on april" then "pascal on may"

I am not spreading anything , I am only saying this rumor can be true for all we know.Last I checked that rumor where NV was 6 months behind AMD was true too!

 

[Jaydip]

I'll take it one step further. Due to people getting free samples, and not wanting to rock the boat and continue to get free samples, the real performance challenge is a month or two or three after release when people are doing their own testing, and not regurgitating what the reviewer guides told them to do.

I think anyone remotely competent as a reviewer throw those guide into trash cans.

 

[Sweepr]

WOW thats sick o_0 :thumbsup::thumbsup:

Just imagine a Ellesmere XT performing like a 390x + 97%, for a mere price of 300$ :thumbsup::thumbsup:

And a Ellesmere Pro performing like a 390 + 86% for what ?? 200$ :thumbsup::thumbsup:

The truth is in the translation ():thumbsup:

--------------------------------------------------------------------

It all depends on how you interpret it..

For those with reasonable expectations, that wouldn't be bad at all.
Much closer than the specs would suggest.

- R9 390 (Hawaii Pro): 2560 SPs, 1 GHz, 384 GB/s BW (512-bit)
- Polaris 10 '67DF:C4' (Ellesmere Pro / R9 480?): 2304 SPs, 800 MHz (?), 192 GB/s BW (6 Gbps @ 256-bit)

*Specs based on the SiSoftware entry.

 

[Det0x]

BTW the same guy that said nearly 4.000 points @ Firestrike Ultra - the data which copycat websites used to base the 'Geforce GTX 980 Ti' at $299 rumours - also said this:

- Ellesmere Pro: 86% R9 390
- Ellesmere XT: 97% R9 390X

WOW thats sick o_0 :thumbsup::thumbsup:

Just imagine a Ellesmere XT performing like a 390x + 97%, for a mere price of 300$ :thumbsup::thumbsup:

And a Ellesmere Pro performing like a 390 + 86% for what ?? 200$ :thumbsup::thumbsup:

The truth is in the translation ():thumbsup:

--------------------------------------------------------------------

It all depends on how you interpret the rumors... But for some people there never is any questions -> always spin negatively about other brands, while hyping your own fav

 

[Vesku]

WOW

Am I seeing a model with HBM. Ellesm. DS = 110% 390X

We also have shader performance regression. The GPU side has learnt a thing from the CPU side. It's about time the divisions shared their expertise. Bulldozer adapted for GPUs.

Ellesm. XT = 97% 390X (2816,176,64,384)

And this is after a node shrink.


YAY. Nvidia forever. They're never gonna catch us now.

Using HBM on a card that is only 10% faster than the 390X is fairly implausible.

 

[tential]

I think anyone remotely competent as a reviewer throw those guide into trash cans.

Almost no reviewers do that though sadly. You can expect mist to follow them. Or to push whatever hardware they like the best. You rarely see games that favor both vendors, neutral and have dx11/12 results too.

 

[Arachnotronic]

I'll take it one step further. Due to people getting free samples, and not wanting to rock the boat and continue to get free samples, the real performance challenge is a month or two or three after release when people are doing their own testing, and not regurgitating what the reviewer guides told them to do.

Most reviewers return their "free samples" AFAIK. It's the early access bit (i.e. being able to publish reviews at the same time as everybody else to get in on the initial surge of traffic) that's key.

 

[Glo.]

If ellesmere is on R9 390X level it will be as fast as GTX 980 Ti.

 

[maddie]

Using HBM on a card that is only 10% faster than the 390X is fairly implausible.

One would think that of the entire Chiphell post.

 

[S.H.O.D.A.N.]

HBM and not DDR5X for a card with 390 levels of performance? How does that make any sense? What kind of SKU would that be and how would you position it in the market alongside HBM Vega?

 

[Glo.]

HBM and not DDR5X for a card with 390 levels of performance? How does that make any sense? What kind of SKU would that be and how would you position it in the market alongside HBM Vega?

Most likely it could use HBM1.

 

[Sweepr]

Or perhaps this model won't see the light of the day.

Robert Hallock, the Technical Marketing Lead at AMD, explains: "We have the flexibility to use HBM or GDDR5 as costs require. Certain market segments are cost sensitive, GDDR5 can be used there. Higher-end market segments where more cost can be afforded, HBM is viable as well".

If the Chiphell post is false, that puts the credibility of the rumoured 3DMark Ultra score in check too, both come from the same user.

 

[S.H.O.D.A.N.]

Most likely it could use HBM1.

And be stuck with 4GB in a world of 8GB? If Nvidia can get GDDR5X for their 1080, then AMD can get it for whatever 'top tier' Polaris 10 SKUs they have planned out.

Anyway, why bother with expensive, poor yield HBM1 on a midrange product when you can put those modules into higher margin hardware?

 

[tential]

Most reviewers return their "free samples" AFAIK. It's the early access bit (i.e. being able to publish reviews at the same time as everybody else to get in on the initial surge of traffic) that's key.

Ya, when I say free samples, I mean the whole thing, the free samples/early access. As AMD also starts to play this whole free sample/Early access game (look at the recent past especially with fiji), can we really be 100% sure of these reviews? I mean, even the most pro AMD/Nvidia site doesn't want to lose early access reviews which is worth tons in ad revenue.
And then the sites who don't get the early access, they feel snubbed, and have incentive to write poor reviews, or just not so favorable.

Sadly though, with the way the GPU system is setup right now for reviews, there are very few reviewers without massive incentives to be biased, whether people want to admit it or not. (On a side note, this is also why I continuously recommend people to get a GPU+Adaptive sync monitor. For the average gamer, this removes all the "Which GPU is faster" out of the equation, since both GPUs will still provide a smooth gaming experience and you can just instead pick your favorite brand/cheapest option/lowest wattage, or whatever else you care about.).

Correct me if I'm wrong, but Termie buys all his video cards for his reviews right?
That's a key review I will look at as that's also a reviewer who directly tests concerns we have.

 

[maddie]

Or perhaps this model won't see the light of the day.



If the Chiphell post is false, that puts the credibility of the rumoured 3DMark Ultra score in check too, both come from the same user.

I can see where you seem to be most confused.

Most of the people claiming that it might be possible for Polaris 10 to approach or maybe even equal 980Ti levels are arriving at that conclusion by analysis of what we know about present cards, the process node change, together with info and guidance already released. It might be incorrect, but at least it's reasoned. Not some post from Chiphell as proof.

My suggestion is to look at the arguments being presented and countering them with you own reasoning. Your viewpoint would certainly look a lot more professional.

 

[Glo.]

Or perhaps this model won't see the light of the day.



If the Chiphell post is false, that puts the credibility of the rumoured 3DMark Ultra score in check too, both come from the same user.

And Hallock said that when was asked about the small GPU. In essence that is higher end tier than low-end.

But overall, yes, you are right. It MIGHT be fake.

 

[Glo.]

Mahigan explained it before, I can only sum everything up.

Polaris will have new command scheduler, that is able to handle both DX12 and DX11 without making bottlenecks. However, because DX11 is serial API, and GCN in essence is so multithreaded it always will create stalls in execution of pipeline. There are two possibilities that can help here. About one - we are 100% sure it will happen with Polaris, and that is Primitive Discard Accelerator, which we already discussed. Second thing, which we do not know for sure is that patent that allows to power down unused parts of the GPUs(stalled), and dynamically OC the parts that are executing tasks. So it will appear that it is 2304 GCN GPU with 1.5 GHz boost clock. Command scheduler also is designed to schedule tasks to work intelligently on whole GPU, not one part, thats why it will be possible to ultimately fully utilize the GPU on DX11 scenario. It is again a trick, just like PDA. But it will make Polaris much faster in much smaller thermal/power envelope.

 

[Sweepr]

Most of the people claiming that it might be possible for Polaris 10 to approach or maybe even equal 980Ti levels are arriving at that conclusion by analysis of what we know about present cards, the process node change, together with info and guidance already released. It might be incorrect, but at least it's reasoned. Not some post from Chiphell as proof.

My suggestion is to look at the arguments being presented and countering them with you own reasoning. Your viewpoint would certainly look a lot more professional.

My suggestion is do a 2-second Google search before throwing personal jabs at other users because they might not agree with your performance predictions. The recent articles were obviously based on that post. And unless you have insider info, with so many unknown variables, your 'viewpoint' is nothing more than guesswork.

The Polaris 10 GPU is the successor to the 300-series, which AMD views as its mainstream range. This means Polaris 10 is not the next Fury and Fury X, but rather a 300 series successor, which is up to and including the R9 390X. That equates to the eventual Polaris 10 powered Radeon R9 490X being capable of GTX 980 Ti performance for a price tag in the region of $300-400. Click to expand Polaris Price Performance Comparison The Polaris 10 GPU itself has a maximum TDP of 175W, but AMD claims it will generally consume far less than that. Early benchmarks have the Polaris 10 scoring in the region of 4000 points in 3DMark Fire Strike Ultra, which puts it firmly in the ballpark of the Fury X and the 980 Ti. If AMD can hit the rumoured $300 price point with such a graphics card then it could have an absolute monster on its hands.

 

[maddie]

My suggestion is do a 2-second Google search before throwing personal jabs at other users because they might not agree with your performance predictions. The recent articles were obviously based on that post. And unless you have insider info, with so many unknown variables, your 'viewpoint' is nothing more than guesswork.

Even if some articles are based on a posting. So what?

Someone still has the option to do some analysis to see if it makes any sense, or if it appears to contradict what is already known.

Maybe this line is reasonable to you, but at least you can explain why.

Ellesm. XT = 97% 390X (2816,176,64,384)

The only thing different from a 390X is the memory bus width. AMD went to great lengths on several occasions to impress on the public the many new improved units and we end up with a performance regression. This is also with a 28nm planar > 14nm finfet process change.

That post seems most reasonable to me after all, and if you think everything is based on blind guesswork, then so be it.

 

[Sweepr]

Really depends if Ellesmere XT is 256-bit GDDR5, the clockspeed (ES or final clocks?) and if the specs are correct or not (2816 SPs for the full die or less, 2560 SPs?). The fact is, right now his Ellesmere Pro prediction matches the GFXBench score from the OP. I'll update if better results show up.

 

[Adored]

The memory bus will not be a big factor, Hawaii can't gear near to saturating its bus.

 

[ShintaiDK]

The memory bus will not be a big factor, Hawaii can't gear near to saturating its bus.

Your image proves nothing in that regard. However people overclocking the memory on Hawaii gets better performance. Same goes for Fiji

 

[USER8000]

The memory bus will not be a big factor, Hawaii can't gear near to saturating its bus.

That is because Hawaii is made for professional compute workloads where that extra bandwidth might help.

There is also another consideration - Hawaii probably has a reasonably amount of transistors dedicated to DP compute.

I expect Polaris 10 to be much lower in this regard.

I still think R9 390 to Fury level performance,but with significantly lower power consumption.

If they get like consistently close to a Fury X or a stock GTX980TI I would be impressed,but I have the impression AMD has targetted quite a bit of the Polaris range to get wins in laptop and prebuilt desktops,and we don't know if the 14NM node they have can clock as high as the TSMC 16NM node Nvidia is using.

 

[RussianSensation]

Lets say its a desktop part equal to a 390. What is it supposed to cost? I bought a 390 for $230 6 months ago.

I mean... unless its $199 then what is the freakin point. OOO it uses less power go green! Who ****ing cares. Give me performance.

For those savvy tech gamers who wait for deals or have great deals available, it's been possible to get $200-249 R9 290, $255-280 R9 290X and $230-245 R9 390 for months, even years for R9 290 series. That means for these gamers Polaris 10 shouldn't matter. In reality it shouldn't either way because it would be akin to upgrading from a 6950/6970 to say a 7850/7870. The proper upgrade for higher end level chips such as Hawaii is Vega. At the same time, prices will reflect all of the above. Polaris 10 won't be a $550-650 product.

Also, look at regular MSRP and prices worldwide. Most of the world doesn't have these deals. Right now for AMD they are officially selling 390 for $329 and 390X for $429. If AMD beings Polaris 10 to $179-249 price levels, that's a massive improvement from what the rest of the world has. Look at how many people on Steam have 960 and those cards sold for $180-220 for almost all of 2015 and for most of the 2016 until now. Imagine getting 390X performance for $249, then with rebates it could be $220-230. That's easily a 70-80% improvement over a 960 as far as GPU power goes. Look how well 960 sold and that card was overpriced trash. Getting 70-80% performance for these gamers is a big deal given how weak the $200-249 market segment is. Finally look at where the competition has 390X level of performance. 980 generally sells at $420-460 levels. Bringing that to $249 MSRP is a huge improvement, miles better than 960 replacing a 760.