Mobile Polaris Discussion

The tested Polaris part is a desktop gpu, which on its own consumes about 30W (vs 70W for 965M) - in the demo, it was the whole system with Polaris which was consuming 86W (not the gpu on it's own).
That's just over 50% more efficient while providing slightly better performance than 965M.

Depending on how much more AMD improves upon Polaris (considering that this was a desktop part that was tested after all), I wouldn't be surprised if we can get 980 Ti level performance at 100W.

Still, for now we can only wait and see what comes out.

 

That is valid argument but I still think they will bring something faster or make it faster.

 

At the moment nvidia have nothing to announce but rebrands. Good thing too.

AMD need this chance! We consumers should rejoice.

I was forced to NV but after so much trouble. Pulled OC support, bad driver support etc I think it foolish to have ANY trust and or faith in nvidia to keep the gaming enthusiast community alive and happy

Sent from my SM-A500FU using Tapatalk

 

here's a bit more news regarding Polaris:
http://wccftech.com/amd-unveils-polaris-11-10-gpu/

 

Note: the whole system with Polaris gpu was consuming 86W... the Polaris GPU by itself was in the area of around 30W of total power consumption, so that's over 50% more efficient than GTX 950, and it was stated that further efficiency improvements will be incoming before its released.

 

Thanks, I can read.

 

I so far like what I am reading. Was a former AMD customer until 2007-2008 after my HD6870 desktop card started faulting out. This will save so much in costs down the road...

 

http://arstechnica.com/gaming/2016/01/amd-confirms-high-end-polaris-gpu-in-development-for-2016/

"We have two versions of these FinFET GPUs. Both are extremely power efficient," said Koduri. "This is Polaris 10 and that’s Polaris 11. In terms of what we’ve done at the high level, it's our most revolutionary jump in performance so far. We've redesigned many blocks in our cores. We’ve redesigned the main processor, a new geometry processor, a completely new fourth-generation Graphics Core Next with a very high increase in performance."

"We believe we're several months ahead of this transition, especially for the notebook and the mainstream market," said Koduri. "The competition is talking about chips for cars and stuff, but not the mainstream market."

 

Not to mention that Ngreedia is again blatantly lying about their products by fake introduction of new Pascal chips, while in reality showcasing two Maxwell gpus. God, they are hopeless!
http://videocardz.com/58116/did-nvidia-show-maxwell-instead-of-pascal

 

Old news but yes if Polaris actually ends up being good and even more unlikely IF it ends up coming to MXM then it will be a sweaty time for Nvidia.

Q2 release for MX parts means probably Q4 earliest release for Pascal (if they are even ready). AMD could exploit this situation if they want. Will they? Doubtful. AMD have resided in the shadows since the 8800 GTX. When GCN was new they almost came back but not quite...that was 4 years ago!

 

Why is it unlikely that Polaris would come to MXM?
Polaris was announced for laptops... so that more or less implies laptop graphics directly.
Although granted, we don't know which direction the market will go, so for all we know, MXM might be discarded in favour of integrated GPU's even on the high-end.

AMD also announced that Polaris will be released before return to school shopping season, which is early summer - I think that actually says something (along with the premise that they were probably pouring everything they have into Polaris and ZEN - so I doubt they would want to delay anything this time around).

 

Because this:

I don't like how AMD has been highly focused on APUs in the mobile sector and more or less entirely ignored the MXM space for the past 3 years. Tonga was completely absent from MXM, even though it had sizable gains over pitcairn

 

Wasn't/isn't mobile Tonga the M295x?
Also, whose fault is it exactly for its absence in MXM?
Dell seems the only manufacturer that has it.
Lack of availability might be due to AMD's poor financial standing. They're in trouble and need a big win in performance and power efficiency, hence it might be that AMD simply shifted focus towards releasing desktop HBM (plus, when you take into account very small uptake of APU's over the years from OEM's and them using them in relatively underpowered systems with poor cooling... you get the idea - but AMD wasn't directly responsible for this because they cannot force OEM's to use their hardware).

As for AMD focusing on APU's in the mobile sector... I think they didn't have anything else to push out.
Carrizo saw better OEM adoption vs Kaveri, but the quality of hardware remains lacking in USA offers (more 'decent' hardware offers on Carrizo parts seem to be in Europe - but the news don't really cover this).

There's also a general lack of advertisement and people just don't know about availability of these chips, or they do, but cannot find any info as to where to get them.

If lack of MXM occurs with transition to the newer manuf. process, it could be adopted across the board, or OEM's might treat AMD differently compared to Nvidia (as we've seen that they are more than ready to pull out all the stops for Nvidia and even try to integrate a desktop 980 into mobile, but for AMD top-end M295X, they cannot really be bothered to modify the cooling or provide adequate power bricks).

At this point, its a question of whether OEM's decide to retain MXM as a whole or not, and technology is moving into direction of full integration to begin with (and less modularity).

Right now, I don't think they would gain anything by integrating high-end gpu's onto the motherboard because the thickness is not affected a lot by it - the most impact which can be seen on laptop thickness stems from optical drive units.

Also, if OEM's work on creating a far superior cooling system for laptops, they could retain MXM for GPU's, but we know they like to focus on cost efficiency instead and cut corners.

Anyway, I think we can only wait and see.
I don't think AMD will disappoint with Polaris in terms of performance and efficiency, and who knows.
Next generation after this one (or starting with this one) might not even use MXM any more and could use something else, or go full blown integrated.

 

Still wonder what was/is the reason for not seeing the R9-M295X/R9-M390X from say Clevo. Someone has bribed their loyalty or something? After all Zentrica made an R9-M295X and DELL has W7170m. So it's not like it can't be done. Maybe @Prema can put some light into it. I still have a bit of a hope for MSi GX series comeback, but it's really, really slim. Zen + Polaris + HDR display would be AWESOME!

 

Correction: HBM Zen (which we know is being made) + HBM Polaris +HDR display would be awesome (in laptops of course).

 

Let's speculate a bit on the announced laptop GPU. From what I've read, it will be based on the desktop midrange solution, i.e. if we take the current AMD lineup - it is 380(X). Considering the architectural improvements from node shrinkage and Raja's words of "unprecedented, a very high increase in performance" we can assume that the new midrange solution would be at least 40-50% faster than 380(X). If I'm not mistaken, this would put it at the Fury(X) levels of performance (pls correct me here, I'm really unsure about this). Consider, the mobile chip will be a chopped down version of the desktop counterpart, then its' actual performance would be at Nano level. Thus we may finally see the Nano-class card in the mobility sector with half the TDP or basically 980 performance level.

This might not be the big comeback we are all expecting, but it would still be a significant step in the right direction. Personally, I would already be satisfied (barely) even with 980M level of performance, in case it is cheaper of course.

 

I thought the laptop GPU was at GTX 950 performance level, which is significantly slower than R9 380X, no?

 

Indeed, though the showcased GPU was a desktop card. I really hope it was an entry-level card though, since it does not anyhow fit in the mid-range segment of the new generation cards and claims of record performance gains. In addition, a 30W TDP is a ridiculously low power consumption even for laptops, leaving a lot of headroom for more gains.

 

That's how I understand it as well, and I think it was in one of the articles - mid-range chip for small notebooks or power efficient desktops. We are yet to see what the top of the line could offer.

 

If I had to guess, it's because they're pointless. More expensive than 970M and at BEST equal to the 970M. Why bother?

Also, no gsync. And there's no freesync panels for mobile that I know of.

 

Iam all for fair competition. Unfortunately AMD has been inferior to nvidia as far as gaming is concerned.

This is why when nvidia decide to launch a ridiculous 150-200w bloated mxm based 980 OEM happily jump to it.

They could have done same treatment for the M295x and clocked it up to desktop speeds and trumped the 980M but it just didnt happen.

Even if you compare 680M to 7970M the 680M trumps it easily.

During the past decade only the 5870M was a leader from AMD (last ati branded release). Sad but true.

I hope Polaris is a comeback. Only time will tell...

Sent from my SM-A500FU using Tapatalk

 

Because it's better multi-tasker. You can get your pro fix without reaching deep into your pocket or waiting W7170M to hit the Bay. If ever. Dead set that they sold their backs, waiting to be proven wrong. Just sayin'. Last time I've checked diversity is good. I see that nGreedia would give you ton of it this year.

 

Well, if AMD based notebooks don't have Freesync... for a moment, a feature that was shown by AMD like 3 years ago or so... then why would I buy it at all??? However, I guess notebooks will have it... I hope so.

 

Not that I enjoy the delays, but when you deal with standards it's usually slow. That's one of the plus sides of proprietary, that's of course if you have the money to push your proprietary ****. Also the reason why FreeSync came out so late in first place.
Don't know about AMD, but I can see a lot of nGreedia users that have G-SYNC hardware but don't have G-SYNC and wont be able to change that, unless they get a brand new machine. I hope it wont be the case with FreeSync, because standards.

 

Sorry, but WHAT?
7970m P8823
7970m FS6178

 

That's a ridiculous overclock right there...

 

You hurt my feelings by writing so, but I appreciate your honesty.

 

This is something people usually miss when writing about AMD's overall gpu performance.
They completely ignore its professional software capabilities while focusing solely on gaming and then they judge it solely on that - seems rather one sided and unfair.

Granted, given AMD's lack of releases over the past few years in the gaming area is hardly commendable, however, the M295X seems like a decent enough gaming chip, even though it is less efficient than Maxwell.
It provides gaming performance of roughly GTX 970m while also being FAR SUPERIOR in pro software.

Now, the OEM's could have used it and made accommodations for it's 125W TDP - the 980M has the same TDP and they didn't mind providing adequate power bricks and adjusting the cooling for it.

Could it be recommended to consumers? Of course... probably those that like AMD and also want to work in pro software without shelling out the big bucks but also laptop gamers who want a relatively decent performer (the recent drivers would have likely ensured this).

Still, M295x is basically the ONLY mobile Tonga release that we've seen from AMD lately, and I would imagine given their overall finances and focus on other products, its hardly surprising they hadn't had anything new to release.

At any rate, Polaris should be interesting.
I'm looking forward to seeing further improvements they make to it and whether or not we get HBM on high-end mobile gpu's.

 

Apparently all the sites claiming they saw the BIG Polaris GPU at CES are wrong. The low power Polaris they demoed and the other chip shown to press are codenamed Polaris 10 and 11 (not sure which is which, some say 10 is the small one). Now while the small one is indeed the smallest chip the big one isn't the biggest. According to Charlie from Semiaccurate who saw the chips himself it is just a bigger Polaris GPU and NOT the biggest. He also says he has info that there is a bigger chip which AMD hasn't talked about or shown yet. I know Charlie isn't always right but a lot of the time he is and this time it makes sense.

Anyway the small Polaris is like a Bionaire replacement, the midrange one Tonga replacement, and the top one Hawaii replacement (not saying Fiji because I don't expect to see a 600mm^2 14nm chip anytime soon) with over twice the performance of each (maybe less in case of small Polaris since that one is a really small chip). This means the midrange one, let's call it Polaris 11 for now, will probably end up as the High end GPU for mobile. If this happens we are probably getting twice the performance of 380X in notebooks thanks to lower voltages of 14nm LPP for similar clocks and factor in the huge efficiency gains from architecture changes and especially the process shrink.

 

May the Force hear you on that! Because in that case we're getting heavily OC'ed 980Ti performance on mobile, which is too good to be true! I personally really doubt that. 980 performance would already be super good.

 

I don't remember anyone claiming they saw a big polaris chip demoed. Amd specifically stated that the tested part was entry level.

As for performance, I don't see it as impossible getting 980ti level performance on a high mobile polaris part.

Polaris is supposed to produce 2.5 times performance per watt vs Fiji. That means that AMD can fit nano performance into 70w envelope. Give the chip extra 30 to 40w and it should be able to reach 980ti performance, if not even surpass it.

AMD did mention they have more optimization to do before release, so right now, we can only theorize without confirmation. Imejust saying that getting 980ti performance or slightly beyond that should be possible with hogh ens polaris... and proobably more if HBM2 is introduced

 

WATT per PERFORMANCE ratio increases exponentially. We won't get 980Ti performance in 70W GPU.

 

There is no reason why they shouldn't get at least close to Fury X performance, heavily OCed 980Ti maybe not but 980 is way too weak. A 390X is a bit faster than that yet they can easily fit an equivalent GPU in under 100W with just the power efficiency improvements they have claimed. This doesn't even factor in any binning that's usually done for mobile chips, not the fact that the second GPU would be faster than even a Fury X and reducing clocks would make it way more efficient than anything we have now. Think of what happened to the Fiji Nano, they reduced the clocks a bit and the efficiency skyrocketed. Now with the higher clocks that the 14nm LPP is supposed to bring (has to do with lower voltages at the same clock and higher transistor switching speed) we should finally see higher clocked stock desktop cards, 1200MHz shouldn't be an issue I believe, so that hopefully leads to 1GHz mobile chips that are very efficient.

 

First "journalists" have claimed that they saw "THE BIG POLARIS GPU", this has nothing to do with the small Polaris GPU, AMD actually showed a second larger chip which people decided to call "THE BIGGEST" without even getting confirmation from AMD first. I don't remember which site did this but a few others decided to follow them too. You are right that 980Ti performance isn't anything impressive, it loses to Fury X at stock, and Fury X is full of bottlenecks because old GCN wasn't designed to support a chip that huge. AMD itself claimed they have completely changed the architecture to fix all these issues which has improved the efficiency both by performance/Stream Processor and performance/Watt by quite a bit to the point where while not as high as the performance/watt improvement from the node shrink it's still a large chunk of the total efficiency improvement.

Well you are probably right that we won't get that much performance at 70W, I don't expect to see a 300-350mm^2 GPU under 90W (FirePros are like 90-95W) in the first place and 100-120W could easily outperform a Fury X as long as there are no memory bandwidth issues. Now the desktop 480X or 490X (not the Flagship GPU) should use HBM so hopefully the same chip on mobile uses it too.

 

And I didn't say anything of the kind.
I said that R9 Nano performance (which is slightly faster than desktop 980) would fit into 70W (since AMD mentioned that Polaris would have 2.5 times performance per watt of Fiji, and I'm simply extrapolating that since R9 Nano is 175W, a Polaris Nano version would be 2.5 times faster - dividing 175W by 2.5 equals 70W, which means that there's roughly 30 to 50W leftower for AMD to work with so they can cram 980Ti performance into 100W or 120W mobile high end part - that is, if my math is roughly accurate).

So it could theoretically be something like this:
70W mobile Polaris = R9 Nano/ slightly faster than desktop 980
100W - 120W mobile Polaris = desktop 980Ti (or overclocked 980Ti).

It really depends on what AMD decides to do with Polaris and how efficient they make it - they mentioned they still have more optimizations to do, so I'm just saying that a high end mobile Polaris 100-120W part that has 980Ti level performance (or more) should be possible.

Also, we don't know whether HBM will be placed into high-end mobile Polaris parts...
I could see the 70W mobile Polaris with say 4GB HBM, while the 100-120W mobile Polaris could have 8GB HBM.
Other low-mid range gpu's would likely come with GDDR5.

 

Lol you guys are dreaming way too much. In case you have never noticed, as core speeds go up, power consumption goes up dramatically as well. It is FAR from linear. The efficiency gains that they have achieved with this weaker card will definitely diminish as the clock speed is increased. It has always been that way. Also there is leakage to deal with. I think you all have your expectations way too high.

 

14nm LPP has lower voltages at similar clocks, plus Finfets have way less leakage, 1200MHz clocks for desktops was only an average estimate, maybe even a bit conservative. Then again I don't expect to get Nano performance at 70W but above Fury X performance is achievable for top mobile Polaris GPU.

 

I'd expect the AMD part for 90 watt laptops (xps15, mbp15) to roughly equal up to a 970m.

For AMD that would be amazing progress as m375x is only about 60% of a 960m. The new Iris Pro will likely match the m375x.

Right now for AMD's survival they need design wins in the high volume segment.

 

As far as I understood, this is exactly what AMD aims for. Everything what they have been saying so far is consistent with this vision (wattage, performance, form factor).

They go after market segment of notebooks which so far were thermally limited to x50m / x60m, not after gaming notebooks which kinda-sorta noisily warmly managed to stuck x70m / x80m into their thin chassis (MSI, Aorus, Razer, Gigabyte).

So I would expect this new smaller Polaris GPU in: Macbook Pro 15, Dell XPS 15, Asus Zenbook, Asus G501, 15" Thinkpad / IdeaPad, HP Omen, Acer Nitro, etc.

There is a lot of these notebooks being sold, likely many more than anything with GTX 970M / 980M.

Nvidia Pascal x50m / x60m will probably be comparable, but if AMD Polaris gets there half a year before Nvidia Pascal, AMD could score a big win [*].

---------
[*] AMD had already shown a working chip and plans to launch with notebooks at shelves for "back-to-school" timeline, while Pascal is still just on the paper (apparently what Nvidia shown physically at CES wasn't real Pascal but MXM Maxwell), with rumours pointing at mobile Pascal launch soonest at the end of 2016.

 

While most of us are focusing on the future of AMD (understandably), there is something in the present that just popped up quietly.
In one of AMD's mobile graphics card pages, the R9 m395x had 8 GB of VRAM. That never popped up until recently as the Alienware 15 R2 now has an AMD 8 GB option. You'll need to view it in desktop mode and check the graphics card options. I just stumbled upon this fact.
It doesn't mean much, but at least VRAM capacity is in good shape for Polaris. At least for mobile variants.

 

AMD is known for taking old card and rebrending it to twice more memory size because couldn't overclock significantly (those cards are already close to bleeding edge like 390x 8GB when Fury had 4GB). It's the cr@ptic AMD needs to cut for future Polaris rebrands to look serious and to save/better invest money.
So that investigation means nothing for Polaris.

 

Well... adding memory has almost nothing to do with whether or not their memory can overclock better. Look at the GTX 960. It performs somewhat worse than usual in certain games that like memory bandwidth being high. 2GB or 4GB versions don't matter, if bandwidth is too low it's too low.

The reason the R9 390/390X had 8GB was because there were 8GB 290/290X cards and there was no reason to reduce the memory amounts.

The reason the Fury, Fury X and Fury Nano didn't have 8GB of vRAM is because HBM1 wasn't specced for over 4GB of vRAM. Going by that knowledge, you'll never in your life witness an 8GB Fury or Fury X. If AMD is stupid enough to rebrand Fiji and use HBM2 on it, then maybe you might see 8GB. But that would be stupid. Arctic Islands needs to fill the ENTIRE line of AMD cards. End of story. There must be no rebrands in the entire line or AMD is just asking for extra flak coming their way. And they need to have as little flak as possible.

As for the "craptastic rebrands", if you cross-check the specs of the 290/290X vs 390/390X, the cards not only got a bump in base clock speeds, but they actually got a bump in base memory clock from 1250 to 1500 (5000 to 6000). They already have a 512-bit memory bus; they actually match the 980Ti/Titan X's default memory bandwidth INCLUSIVE of Maxwell's memory bandwidth improvements. If you can get their cards to 7GHz on the memory (I'm not sure how their memory OCs work, I only hear of their core being crap for OCing) then they'd actually have 448GB/s memory bandwidth, which beats the 441.6GB/s effective memory bandwidth that you'd get from clocking a 980Ti or Titan X at the generally-accepted limit of GDDR5 of 2000MHz (8000MHz effective). This is inclusive of maxwell memory bandwidth improvements as well.

But yes, again, the new lineup must be 10000000% arctic islands cards. I don't even care if it's a fanless volcanic islands equivalent of the GT610, get it out of there. Arctic Islands or go home.

 

Core overclock, not memory. AMD's top cards usually don't have OCing potential like Nvidia's.

 

Is there reason to believe Polaris will be available in MXM 3.0b? If not, why care? My impression AMD will likely go for low power BG for laptops.

I have doubts AMD will release anything for high end laptop, it will just be for desktop.

Hopefully I am wrong, but that is my impression.

 

They need a high end MXM card and since it looks unlikely for this generation of graphics cards to have HBM on mobile, it would be stupid to not release a high end Polaris card for the enthusiast crowd, ESPECIALLY if they beat nVidia to market!

 

Although they need a new die for it. Even if they start on it now, it wont beat nvidia pascal at this point.

If they clock polaris 10 to the moon, it might have similar performance as a 970m with worse power consumption. If they try to push polaris 11(250+w card), on mobile, it will end badly.

 

You do realize that Polaris was stated to have 2.5 times performance per watt vs Fiji, right?
Bearing that in mind, I wouldn't be so quick to say that Polaris won't be able to be on par or possibly even better than Pascal (since Pascal is expecting 2 times performance per watt vs Maxwell).

Also take into account that Polaris uses 14nm vs 16nm for Pascal.
AMD has an advantage on the manuf. process so they might be able to squeeze out more performance and efficiency on that too, plus, it also depends what kind of other enhancements they made to Polaris.

Plus, as far as I know, low end and entry level gpu's will effectively be Polaris.
I think it was also stated recently that none of the new gpu lineup will feature rebrands.
For high-end mobile Polaris parts, I think we will have to wait until summer or just after summer.

 

On my phone so short reply.

Its not that AMD cant beat pascal. Its the fact that AMD cant beat high end pascal on mobile with the 2 dies they have. Polaris 10 is the low power with the performance of a 965m/960m?. Polaris 11 is a big die high end desktop gpu.

As for 14nm vs 16nm, is not that simple. See discussion between me and ethrem. It seems like GF is using Samsung's process.

If there is a midrange polaris die thats designed for high end mobile, it would have at least taped out by now or amd would mention it. I believe AMD did say that we have 2 polaris atm.

 

Polaris 11 isn't a 250W+ GPU, I would be surprised if it uses more power than Tonga. Also with release drivers and higher clocks (the early sample was running at 850MHz) Polaris 10 should perform similarly or even faster than 970m at <= 50W, the ultrathin variant will stay under 30W for obvious reasons.

As for there being only 2 Polaris GPUs, there are actually 3 based on the same architecture (the largest one will be more compute oriented obviously), with the highest end one codenamed Vega 10 which is still unannounced and the release timeframe is still unknown.

 

Polaris 10 I think was the one demo-ed in a desktop solution.
The GPU on it's own consumed around 30W (separate of the rest of the system), and AMD stated that they still have more optimizations to do - that was just a demonstration of what the part was capable of early on, and thus far, it seems impressive.

My point is that a mobile Polaris 70W part should be able to match R9 Nano performance... and with 100W to 120W, it should be doable to get to 980Ti level (more so if they slap on HBM 2 on both/all high-end mobile Polaris parts) - but, right now, this is mere speculation.
We won't know what's going to happen.
All I'm saying is that Polaris certainly seems like it has what it takes to pull this off... now we just need to see if AMD will make it (or something similarly great) happen