Mobile Polaris Discussion

HBM2 isn't HBM so there is a slight chance.

 

That's my prediction as well and I think that I said it earlier. There would be an HBM model, the thing is which one and would it fit on MXM.

 

Still waiting on mobile announcements....hoping for HBM MXM Polaris card. Will definitely replace the nGreedia crap in my systems.

 

Same here.. I've had enough with NVIDIA and their **** drivers... AMD Drivers have always worked, NVIDIA ones just don't most of the time...

 

The nGreedia trash is just becoming overwhelmingly stupid. If AMD releasea a competitive GPU with non full UEFI and 100% working chipset, I will also move to Team Red.

 

A lot of people are fed up with green shenanigans, myself included. We're all chomping at the bit for AMD to deliver.

 

Considering the smaller Polaris 11 is definitely coming to laptops, can someone explain what power does it take to play back 4K VR content with passive cooling?


I did get to see AMD's Polaris 11 GPU running PASSIVELY while playing back 4K VR content last night. Pretty impressive. #AMDCapsaicin

— Ryan Shrout (@ryanshrout) March 15, 2016

http://videocardz.com/58621/amd-polaris-10-engineering-sample-pictured

 

That quote seemed a little opaque. What exactly is meant by "4K VR content playback" - seems unlikely that it could be running a modern 3D game engine rendered in real time at 4k in each eye instead of 1080x1200 per eye. Maybe pre-rendered content or normal VR resolution output to 4k display?

 

I believe it was playing 4K video content, not gaming, so it's good that it uses few watts, but I suppose this means a lot more for ultrabook type machines.

And as far as I remember, polaris 11 is the "weak" card, and we will get polaris 10 or a variant of. Polaris 10 seems to be near Fury X performance, so that's nice.

 

It would appear that one of the posters said the following:"
Reports have come in, Polaris 10 was doing ultra settings @1440p @ constant 60fps on hitman. That is faster than Fury X. Because of the 60fps cap we won't know its peak potential and this was done on purpose to hide their hand."

If this turns out to be accurate, then I think we might be seeing a bigger performance gain on laptops than we originally thought with Polaris.

AMD did say they are 'targeting' desktop 970 performance as the minimum for VR... but, could this also indicate that 970 levels of performance aren't the maximum of what we can expect from Polaris?

 

Though I haven't really read anything about Polaris 10 going for mobile. I think this chip will replace 390X or Nano on desktop side and Polaris 11 for mainstream mobile, i.e. ~970M performance There is a chance though we might also see Polaris 10 as a cutdown variant on mobile, since according to the press, it's about the same size as Nano.

 

Correction Polaris 11 will be 970 level of performance

Sent from my LG-H811 using Tapatalk

 

AMD would be downright foolish to NOT integrate full Polaris 10 as a higher-end mobile chip.
And honestly, right now, they are very tight lipped about Polaris as a whole.
They are targeting 'minimum VR levels' which is GTX 970 level of performance... and that's Polaris 11 (which can run without active cooling while playing back 4K video content).. nothing was said that only Polairs 11 will be in laptops.

If Polaris 10 has the performance better than Fury X, then cutting down the TDP to just smaller levels by using the same/similar technique they used with Nano (which sacrificed only 10% of performance overall while slashing down 100W) should be doable while suffering maybe 5% loss in performance... possibly less... and bringing it in line with Fury X performance-wise, or keeping it above that still (depending on how fast they make Polaris).
We don't know what kind of results we can expect as of yet as AMD said that the final product should be even faster (and the Hitman demo was apparently locked at 60 FPS... so we don't know what the upper ceiling of performance Polaris 10 has).

So, theoretically, we might see 980ti levels on mobile Polaris 10 with low power draw (or at least, acceptable enough to be used in high-end solutions, and possibly below 125W TDP.

Polaris IS 2.5 times performance per watt vs Fiji... that would put it WELL above 980TI levels even when doubling Fury Nano performance.
Cutting down the TDP by about 50% while losing maybe 5% of performance (to keep it in 125W range) would essentially put Polaris 10 in the range of 980Ti (if I'm not mistaken).

Either way, we won't know more until later... I'm just saying that the potential seems to be there.

 

Polaris 10 has to be the high end mobile GPU variant. Polaris 11 was demoed vs a GTX950 wasn't it? As far as I remember, Polaris 11 was for the thin and light notebooks.

My concern is that apparently only two producs, Polaris 10 and 11, are in the works. So if we don't get polaris 10, we migth be out of luck haha.

Polaris 10 has the potential indeed, to have mobile 980tis with us.

 

Actually... Polaris 10 might have the potential to go beyond 980ti levels of performance - in all honesty, I don't know, but let's try to hypothesize a bit (within reason).

The Fury Nano has a TDP of 175W. It's already a decent performer and about 10% lower performance and 100W less than Fury X (which with the latest Crimson drivers seems to be able to go head to head, and even surpass stock 980ti).

Enter Polaris into the equation. At 2.5 performance per watt, the Fury Nano would effectively be 37% faster than even overclocked 980ti... at least according to the following benchmarks - which I think were done before the driver updates that gave all AMD cards a boost:
http://www.legitreviews.com/amd-radeon-r9-nano-versus-nvidia-geforce-gtx-980_177681/3

So, Polaris 10 Nano desktop version at 175W would essentially be faster than overclocked 980ti by 37% (it seems that Nvidia achieved these overclocking potentials and lower power draws due to leaving enough headroom on both counts at stock levels, but once you start overclocking and overvolting, their efficiency gains drop off quickly and power draw goes up quite a bit).

If AMD could apply same power slashing technique to Polaris like they did it on Fury X which resulted in Nano... then 100W power reduction would result in about 10% performance loss (however, given Fiji's design and it's HBM implementation, whether the same power saving technique could be applied to Polaris is a pure hypothesis given that HBM might not be used in those mobile solutions).

I know this is a very simplified scenario (and possibly optimistic considering I'm using Nano as opposed the Fury X for reference), but even in the case of Fury X (which has a TDP of 275W), if you slash the TDP down by 2.5 times (because Polaris is 2.5x performance per watt), you end up with Fury X performance (which is now thanks to new Crimson drivers on par with 980ti) at 110W... and AMD said more performance enhancements to follow - so we don't know if this would be the upper ceiling, especially when you consider that apparently Polaris 10 already performed FASTER than Fury X in that Hitman demo.

So, a mobile Polaris 10 at 110W (extrapolated from Fury X... not Nano) and at least FuryX/980ti level of performance (or faster) seems doable in a laptop.

I know the power reductions based on Nano don't add up when compared to the Fury X, because that would mean that a 75W Polaris Nano for example would be 25% faster than Fury X/980ti - but why is there such a radical difference between Nano and Fury X in that case?
The Nano was binned to work at lower voltages... or they simply dropped the voltage and performance accordingly.

I don't think we can use Nano as a measuring tool. I think we'd need to use Fury X instead as a guideline/baseline... in which case, we end up with Fury X/980ti (or possibly overclocked 980ti performance) at 110W (though if they decide to put the upper TDP ceiling at 120W, we might just see heavy OC 980ti performance).

Even stock 980ti performance would be enough for me to buy a new system with AMD Polaris in it... now if AMD pushes HBM on it as well, it would be an added bonus.

 

Definitely not 2.5X performance/watt vs Nano, at least not for desktop cards. From rumors, Polaris 10 seems to be a 232mm^2 GPU, and even newer rumors put both Polaris 10 and GP104 at ~ 250mm^2, and at that size we are looking at slightly more CUs than Hawaii so about 48CUs/3072SPs which combined with architectural improvements and higher clocks due to 14nm should outperform a Fury X, but that still leaves a chip of that size with a 175W TDP, 190W tops. And this is where I believe the performance/watt improvement is compared to Tonga (also the fact that the 28nm GPU in comparison on the roadmap is a 2H2014 one) since that's the TDP range of Desktop Tonga. 2.5X The performance of Tonga isn't a bad thing though, in a lot of new DX12 cases 380X is starting to match 970, and also 380X crossfire at 1040MHz does beat a Fury X (not at 4K) when crossfire scaling is good (even when 10-25% performance is lost due to scaling). The efficiency will obviously be better for mobile GPUs, and even 2.5X the performance of a M395X (which is faster than 970M when it doesn't throttle) should be impressive at 100W. Then again I might be completely wrong and it's 2.5X performance/watt of Fury cards but that's less likely.

Also don't expect to see another nano until Vega 10 launches (late 2016/early 2017) which will be the true replacement for Fiji GPUs.

 

Considering how 'unique' Nano was, I wrote that we shouldn't use it as a baseline (that's a best case scenario)
That's why I chose Fury X ... which has a TDP of 275W.
Polaris was mentioned to have 2.5 performance per watt compared to current GCN architecture... I would imagine this extends to the LATEST Fury cards, not the lower end line - and I think one of the earlier reports indicated that Polaris efficiency gains should be compared to the Fury line.

I could be wrong though.
Since we do know that Vega is in the works as well, this could mean the following:
1. Polaris 11 and 10 are essentially GPU's that have same/similar performance of Fury cards with much better TDP.
2. Vega is the high end of Polaris line and replaces Fiji/Fury gpu's in the high end sector.

But this isn't unusual when a transition to a new architecture and/or manuf. process occurs.
The high end line from previous generation ends up as the middle ground of the new generation, leaving room for an entirely new line of GPU's on the high end to provide high performance.

If that is the case... why is it improbable to see FuryX/980ti performance in mobile Polaris 10 (which would essentially be a highly binned version and will benefit further from power reductions - leaving potential room for upcoming Vega based mobile gpu's).

 

Seems there may be a third card in the works, code named "C99"?

 

Reportedly this is the SiSoftware benchmark results for a Polaris card, which may be R9 480 or 490: http://ranker.sisoftware.net/show_r...e0d2e3d4e7d0e8cebc81b197f297aa9abccff2c2&l=en

 

Here are apparently leaked specs on Polaris 10:
http://wccftech.com/amd-polaris-10-gpu-specs-leaked/

Seems very promising.
I couldn't find info on it's TDP though, but given what AMD demonstrated already with Polaris 10, I think the mobile variants might demonstrate to be very powerful.

 

Here's another link to leaked specs on Polaris 10:
http://videocardz.com/58639/amd-polaris-10-gpu-specifications-leaked

Seems to me like a preliminary engineering test unit with cut down specs ... not really a 'finished' one.
Anyway... if this chip was demonstrated to run Hitman (locked at 60FPS), then the final product might end up quite a bit more powerful.

However, I don't know how powerful mobile versions will be (FuryX/980Ti levels would certainly be possible)... but if this is a cut down version of Polaris... imagine what a binned version for mobile will do.

2.5x performance per watt is due to architectural changes... the process node reduction should be able to give us more performance enhancements (I think this might have been what AMD talked about - they initially tested Polaris 11 which already behaved quite nicely, and then Polaris 10 [probably the one with the leaked specs I linked to] which surpassed FuryX/980Ti.

Would this mean that high end Polaris mobile products will be coming later this year?
Or... if Vega will also have a mobile derivative which will pack a punch (but won't appear in laptops until early 2017).

 

Actually, this might very well be a mobile version of the chip, considering the small core clock of 800 mhz, the same as my 7970M, which was also a significant jump in performance compared to 69XXM series. I really hope this is a so much anticipated return to high-end mobile gaming by AMD.

 

That 14nm node transition is projecting the next gen leap, if that level of boost is true then the gaming industry will get cinematic levels of GPU rendering, maybe we can get a true Division and Watchdogs from E3...

Now the hard to believe part is that Fury X / 980Ti and 980M , GTX 980 Mobile all getting punched in the guts so soon with that 2.5x perf/perwatt...

 

Belief is irrelevant.
That's what happens with 2.5x perf. per watt coupled with massive reduction in manuf. process (at 14nm there's far less leakage for example).

 

I thought it's vice versa. I stay on pessimistic side and say that at best mobile users get Polaris 10 chip in 2017 if ever.

 

We will have to wait and see.

I can see this cut down version being an engineering sample easily enough... but even if it is a mobile top -end polaris part (which I doubt), it would still surpass FuryX/980ti if Hitman demo was any indication (and if this cut down version was tested on Hitman to begin with).

Usually we have cut-down engineering samples being tested before release dates... those don't necessarily end up as mobile products and have more or less higher stats than the engineering samples.

 

I have high hopes for Zen (6 core , 45w TDP ?) + Polaris being competitive to Intel i7 + nVidia combination this year.

 

There are possible indications of Intel signing a licensing deal with AMD regarding graphics as can be read here:
http://seekingalpha.com/article/3961087-amds-reported-licensing-deal-intel-gives

But, if AMD's Zen and Polaris are successful (and it would at least seem that Polaris and Vega are on a very good track of accomplishing this).. would AMD really have a need to sign any such deal with Intel?

It might however increase AMD profile on the market, especially notebooks that have it's hardware inside (with OEM's potentially limiting their sabotage of said hardware when mobile Zen APU's hit the stride next year?).

I don't know about Zen APU having 6 cpu cores... but it would certainly be nice if they could pull it off.
With HBM being in the mix I think they might as the IGP could end up occupying even less overall space on the die... but the higher likelihood will be 4 cores, 8 threads Zen APU with HBM IGP.

 

Nothing would please me more than to have a strong AMD MXM card to replace all the nGreedia chips in my systems. I just hope they will remain compatible with legacy systems, so we won't have to jump hoops like we had to do with Maxwell in Alienware.

 

Thanks Deks that was an interesting article. Well written too!

 

Was interesting to read but I felt bad taste of cheap sneaky method of increasing it's auditory. This source makes it look like there are 2 other pages full of info and all you need is to sign in or register while after using Adguard I found there is absolutely no useful info on those two.
3rd page was completely empty and 2-nd had this:

Can't have much trust to site which uses such tactics. Still, info seems trustful anyway.

 

Probably not Polaris-related, but AMD "Greenland" (Vega?) will have 4096 shaders according to "leak" on LinkedIn:

http://www.3dcenter.org/news/amds-greenland-chip-mit-4096-shader-einheiten-bestaetigt

 

Hmm... wccftech seems to have an article on Vega here:
http://wccftech.com/amd-vega-10-gpu-4096-stream-processors/

Polaris 10 might seem to be equal if not just better than Fury X line... performance-wise and would basically act as this generational performance push into the middle-ground, while Vega is supposed to take place 'at the top'.

Pretty much expected.
Though, the TDP of Vega on that german site seems interesting... 160 to 180W.

I imagine that we could easily see a top-end mobile Vega in that case... which (hypothetically at least) might confirm what I thought before... that from the Polaris and Vega lineup... Polaris would be the middle-ground, while Vega would be the top-end... and, a similar thing could end up being seen in laptops.

If Vega's TDP from that website is in that range... I could easily see AMD making a 120W version with HBM2 as well - which would probably make it a first mobile gpu with HBM technology... while Polaris would still be using GDDR5.

Plus if these TDP figures are accurate, then AMD really managed to pull off a really good efficiency and likely performance as well.

 

Finally some reliable leaks on the launch dates of R9 400 series. To sum up the article says R9 490 and R9 490X will launch at the end of June, with paper launch during Computex and both will be based on Polaris 10 GPU. This means Fury series will either remain as is (in case they are faster than Polaris 10) or they get downgraded to 480 series (in case they are slower, unlikely IMO). Though nothing is mentioned about Mobile Polaris, we already know AMD will release both Polaris 10 and 11 at about the same time, so we can safely assume Polaris 11 will be mobile, hopefully a lot better than the one demoed against 950.
http://videocardz.com/58894/amd-radeon-r9-490x-and-r9-490-launches-in-june

 

Polaris 11 was not meant to be mobile as I read. It was meant to be worse than 10, but desktop card (s)

 

At one of the interviews Raja said with this power efficiency making mobile GPUs was the most obvious of the solutions. Also in one of the news by VC, you can see there are two cards for Polaris 10 (almost confirmed to be 490 and 490X) and SIX for Polaris 11. Bad news is that mobile Polaris 10 is not coming yet, good news however is that 6 variants for 11 should definitely include at least two mobile. Basically it might look like this:

R9 490X - Polaris 10
R9 490 - Polaris 10
R9 480X - Polaris 11
R9 480 - Polaris 11
R9 470X - Polaris 11 (mobile R9 M490X)
R9 470 - Polaris 11 (mobile R9 M490)

http://videocardz.com/58634/amd-confirms-polaris-10-is-ellesmere-and-polaris-11-is-baffin

 

If the above is true which would be nice I reckon based on what AMD has done in the past the mobile top end card would be based on a slightly down clocked version of the R9 480X as a minimum.

The gap between desktop and laptop card performance has closed quite a bit so this is perfectly feasible.

Then when they start to milk they can release a Polaris 10 based top end mobile (one can dream right)

Polaris 10 will probably be reserved for desktop as you state Hurik. Competition would be nice to have again! Drivers *might* start to improve from Nvidia side then...

 

Who want's a TOP mobile GPU with 128 bit memory bus? NOT ME!

 

Depends on how efficient the architecture is using that bandwidth. 5870M was pretty efficient with it when it came out. Maxwell improved significantly the performance of bandwidth utilisation. AMD can certainly do the same with Polaris.

Either way let's see performance when we EVENTUALLY see some facts and figures! The wait is getting painful.

 

Maxwell has bandwidth compression, which mainly affects colors, that's why there are issues (both desktop and mobile) with high gamut (more than 6bit) displays. There is similar tech from AMD and I think that it also has it's fair share of issues. The thing is that one of the slides for Polaris has focused on HDR displays, this means Rec.2020 coverage, which is wider than aRGB, WHICH is covered by a very few displays and ALL of them are at least 8bit (the new 4K DreamColor and PremierColor, the older ones were all 10bit RGB LED). But again, that's for aRGB, for Rec.2020, one needs a new display technology and to return to the GPUs, it also means that Polaris would be really high gamut friendly and hopefully no gimmicky compressions would be used in order to get as much performance as possible *cough*Maxwell*cough*. To sum up, I think that if AMD found that 128bit bus would be enough to drive 10/12bit color, I see no issues either. Just for bragging, my display has ~85% Rec.2020 coverage, or why I picked 8740w DreamColor to build a system around (and yeah, 16:10)

 

Isn't that the reason why AMD is so deep in the hole right now? Because each year for the last 7 whatever AMD found wasn't exactly equal to what customers did, isn't
it?

 

They must improve color compression in drivers if they want to do such a reduction in bus system

 

Give me a definite answer what EVERY consumer wants and trust me overclocking wont be one of the popular answers. I love to tweak, but we are few, and even then I wont let a few MHz to decide my longterm usability. For example last year the other team played a lot of unpopular cards, alienating a lot of people, questioning this longterm usability thing. They still have market share, and pretty big on top of that. So you tell me how market works. I can't.

 

Give me a definitive answer where I wrote "EVERY", or when I gave a hint I was talking about 1 specific wanted thing.
Somebody wants lower power consumption than what AMD offered. Someone lower temps. Someone wanted quieter fan and someone no rebrands. But in the end not enough people bought AMD, that's it.

 

AMD hasn't released a MXM mobile GPU since the R9 290MX.. Which was like 2013/2014.. They better do so with Polaris or they are a dead dodo walking!

 

http://www.zentrica.com/product/amd-radeon-r9-m295x-mxm/
I mean.... this exists. And the W7170m in the Dell Precision 7710 is MXM (if the Owner's Manual is to be believed).
But as far as I remember, isn't it the ODM/OEMs that make the MXMs?

 

Yeah, but they don't have huge budget to cover ALL the wishes out there and what you say in the quote is exactly what I meant, people want different things, but not that different to not consider AMD. What power users dictate is entirely different from what casual gamers need. Then comes the thing that a lot of the power users don't game at all... Also quieter fan - they've put AIO, then people laughed that it is a nuclear reactor, it also came with lower temps, so it ticks two boxes. I understand that they don't have much new stuff in the mobile segment, but they do have pretty good offerings on desktop and the only that are 4K capable... yet most people still don't buy. So again, explain to me how market works. People buy what power users overclock to the hell and back, so they can brag that are among the cool kids on the block, but never use it that way. AWESOME!

It is MXM, but so far can't be find anywhere.

ODMs make orders, OEMs make the parts/components/build the machines. From what I see, no one but DELL has AMD parts in the high end, and there are some HPs in the low and mid end. I hope that things would change this year, or a lot of "enthusiast" brands (as few as they are now) would lose the meaning for me.

 

So wait, no Fiji rebrands at all? Ehhh... I'm skeptical. Is there enough HBM 2 to go around for all of that Polaris? Or did I miss something?

 

Last I heard there wasn't enough HBM2 and they would be relegating it to high end desktop.

Sent from my SGH-M919V using Tapatalk

 

No HBM2 for Polaris, only for Vega at the end of the year.