Mobile Polaris Discussion

Well those were supposedly the small dies but there's no way that they came so quickly either way. I do love how people read into these shipping manifests though......

 

I done some napkin math with zen and my guess is arounds 4790k level with 8 core zen, maxed out at skylake level. And thats presuming the 40% IPC improvement is a general improvement and not AMD making things up. Too bad nothing really came out of APUs and HSA. AMD was banking on that....

I feel like whether nvidia stomp them or not entirely depends on how well gp104 is. I doubt anyone predicted gk104' performance beforehand. This is all assuming Vega 10 is designed towards a semi generational leap. Who knows, it might be a 550mm2 chip.(unlikely). Their code name leads me to think there is space for a vega 11 in the future.

If GP104 is a good card and can compete with Vega 10, all is good on AMD front. It will be another 7970 vs 680 scenario and polaris 11 would be a decent card.

If nvidia had to rush out GP100(very unlikely), AMD is screwed. You will end up seeing a crippled/full gp100 smashing amd at high end and a gp104 smashing amd at mid end. Unlikely because its ngreedia and if they can milk pascal for two-three generations, they will.

Tldr: AMD is banking on ngreedia being ngreedia.

 

Didn't Nvidia usually release entry and mid-range products first before coming out with their high-end models afterwards?

AMD did the same just recently by releasing the re-branded 300 desktop series before coming out with their Fury line - Fury Nano being the exception, since it seems like a specialized product (one which I think would be amazing in mobile if Nano 2 is made with Polaris - which would effectively mean overclocked 980ti performance levels in 100 to 120W).

So, for the moment, I don't think we have the whole information.
What wed might know is that Polaris 10 and 11 target entry and mid-range level users... but not high-end.
I don't think Polaris 11 is stated anywhere to be 'high end'.

Oh and for the $1700 gpu... we are likely looking at a pre-production model which could be used as a template for mass production later on - so the end product probably wouldn't cost $1700, but would likely be lower.

One other thing to take into consideration is that AMD is being very tight with details to anything referring to high-end and their upcoming Zen.
They might be releasing just enough info to get attention, and are likely being more careful of what they release.
We did get some very interesting tidbits of information regarding Zen which is likely to be used in servers, but we still know nothing conclusive of the mobile and desktop versions of the product in question.

AMD themselves stated that Polaris is undergoing further optimizations/refinements in these months before final release.
So, I don't think we can say for certainty just yet what will happen - at least not until the actual products are released and we know for certain that no other models exist or are under way.

 

Depends on how you look it, no. Nvidia usually is pretty top down. So is AMD.

Thats not due to yields or anything. Just AMD not wanting to build another die.

For a fresh out of wafer testing die, $1700 is too low.

More like cleaning it up enough so OEMs would actually touch it. Especially it seems like the polaris 10 is bound for mid end OEM laptop market and OEM PCs.

 

Comment 666...

Also no way 4790K can compete with 8 core Zen in anything other than poorly threaded benchmarks, there is a chance even a quad core Zen will outperform it provided similar clocks

 

Let's hope so or AMD is done for at least in the CPU arena...

 

I write this here but somebody can create a topic about it: HP will release notebooks with AMD FreeSync. HP Envy 15z (also contain A-series APU.

 

Speaking of AMD and laptops...
A comprehensive article was released on how OEM's handicap AMD APU's.
http://www.anandtech.com/show/10000/who-controls-user-experience-amd-carrizo-thoroughly-tested

Single-channel RAM sabotage:
http://www.extremetech.com/computin...igns-single-channel-ram-sabotage-amds-carrizo

OEM positioning AMD systems against Intel's unfairly with unrealistic pricing and low grade components:
http://www.extremetech.com/computin...em-positioning-is-still-damaging-amd-products


Well at least this issue is FINALLY being addressed properly... but how much this results in better AMD products from OEM's remains to be seen.
I think OEM's should be fined billions that need to be given to AMD as intentionally damaging them (could have been done via Intel - yet again, even though the evidence of their involvement seems low this time around).

 

http://wccftech.com/amd-polaris-architecture-vr-minimum-spec/
I assume this Polaris 11 is the GTX 960/R9 380(x) replacement, price-range wise ($200 to $250)? A thought, of course.
I wonder where Polaris 10 will land.

 

If you look at the quote from AMD itself they say we should have VR capable GPU that cost less, use less power, and are faster than 290X.

This can easily be a cut down Polaris 11 GPU though, I doubt a full chip will be that cheap, so It might cost about as much as Tonga now while the full chip may be faster than 390X and maybe closer to Fiji and cost in the $350 range.

 

8350 gets around 663 in cine r15. Assuming the same clock speed, which is not going to be true at all, 1.4*663=928. I am thinking the performance ceiling to be at skylake level and performance floor to be at sandy. Also they building it on LPP 14nm.

I dont know where you are getting the 2 module zen beating 4790k from.

 

I agree completely. This has been going on far too long. So many opportunities lost because OEM's just did very poor design decisions very likely intentionally So many machines that would cap the performance of the AMD chip, but worse yet, only offer a single RAM chip slot for a machine that could perform 35-40% better in graphics performance if it had dual channel RAM. I used to support AMD, and liked what they were trying to do, but Intel and OEM's just purposely dogged the design and performance of a machine making it seem less of a "user experience" than a comparable Intel product.

Problem is that it's probably a little too late. Unless they got a multi hundred million to billion dollar cash infusion soon, AMD will just turn to specialized designs for consoles and other niche markets. Probably stick with the desktop GPU race because they're at least competent in that area. I'd much like to see Samsung buy AMD though and enter the x86/x64 marketplace. I bet they could do some serious damage to Intel and at least bring AMD product up to where it needs to be.

 

Polaris could potentially be a comeback in terms of mobile; I mean look at the Nano, it's not like it's completely out of the realm of possibility for them to hit back there. Not so sure about Zen though.

 

You shouldn't do pointless calculations like that, and you shouldn't compare Zen to Piledriver based 8350. AMD said Zen should be a 40% improvement in IPC over Excavator not Piledriver, by my calculations (and plenty of others who reached a similar number) that should put it closer to 65% improvement over Piledriver. And one thing you should never take seriously is Cinebench numbers, especially R15, that thing is a joke for a benchmark. And also Zen doesn't have modules like Bulldozer derivatives, it has SMT which IBM uses and so does intel. The top desktop Zen on AM4 socket will be an 8 Core/16 Thread Chip, not even 6700K or Kaby Lake equivalent has a hope of competing with that. 8 Core Zen will compete with Broadwell-E, and how well it does will depend on its clockspeeds, something no one besides AMD knows at this point. It seems like AMD has the efficiency advantage due to only 95W TDP compared to 130ish for Broadwell E, but Broadwell E has core advantage for the top end at 10 cores.

 

Back on topic, 3DCenter found something interesting.

http://www.3dcenter.org/news/amd-gr...-232mm²-chipflaeche-unter-der-14lpp-fertigung

Seems like at 232mm^2 this might be the Polaris 11 GPU, like I estimated a few pages back that it may be ~ 250mm^2 chip. Also AMD's statement about their VR capable chip being less expensive, consuming less power, and faster than 290X goes in line with the chip size as 28nm to 14nm LPP is quite a nice shrink (much smaller than what we would have at 20nm) so if we can get at least double the transistors per mm^2 even with the fact that Hawaii was a second gen 28nm GPU and denser than Tahiti, and the fact that Hawaii has 6.2B Transistors at 432mm^2 means that a 232mm^2 Polaris 11 chip should pack more transistors than that, not that necessarily we will see more Compute Units as AMD should have beefed up the compute unit itself and not to mention all the new blocks they added to remove bottlenecks like redesigning the Geometry Engines and adding schedulers (for compute obviously) as well as beefier VCE, UVD and display engines. If they replaced the GDDR5 memory controller with an HBM one that would have freed up a lot of space but then they said at least one of their new chips has both controllers so this one can go either way.

http://videocardz.com/58237/amds-project-f-is-232mm2-discrete-gpu-made-in-14lpp-process

 

Okay.

Firstly, apparently the IPC that's a 40% boost isn't over Piledriver (8350) but rather over their newest APU, which already has a bit of a benefit over Piledriver (though it's still quadcore at low speed, so it's not a high performance part by any means). So we have to wait for Zen, but running calculations off Piledriver isn't gonna make much sense right now.

Secondly, I hope everyone here remembers that the 970 and 980 might be "midrange" now, but for a die shrink + architecture change to keep that, it has to be in a rather entry-level card. If polaris 11 is just a GTX 980-esque card, it really should be extremely low power and we should have something mindblowing past it. 980Ti-level performance should be happening with the card of the price range of the R9 380X right now, if they release a counterpart to the Fury X for the ~$600 price range.

 

I think I already mentioned before a few times that when we take into account efficiency levels Polaris exhibited, AMD SHOULD be able to make a top-end mobile GPU that performs close to, or as well as stock 980Ti at 100 to 120W TDP.
Now, if they make a Polaris Nano on the other hand for mobile and HBM2, it could easily reach 980Ti high overclock levels within 120W TDP.

I can understand HBM2 being reserved for top-end models, and I can see AMD using it like that, but alas, we don't know if HBM2 will be used on notebooks for top-end models (at the very least, we hadn't heard anything).

 

?

Sent from my SGH-M919V using Tapatalk

 

HBM2 for notebooks would be a furious AMD comeback in mobile sector, great marketing win, performance boost for mobile users in general (Nvidia would have to accelerate too) and all-in-all the best step AMD could do... more like a jump.

That is why I am sure it's not gonna happen

 

So far i think this is how Nvidia is reacting to polaris.

 

I wouldn't be so sure.
AMD already demo-ed a low end Polaris part against a current entry/mid range 950 desktop GPU that consumes mere 30W (vs 90W for 950)... and AMD said they are making more optimizations.

Nvidia will also be making Pascal on 16nm, and seems to be primarily focusing on GDDR5X for their Pascal line-up.
Only the high-end was mentioned that it could/will be using HBM2, however, there's something that might get into their way.
Namely, since Samsung will be mass-producing HBM2 this time around, and they entered into a patent dispute with Nvidia some time ago (which concluded in Samsung's favour)... this might affect Nvidia's access to HBM technology.

In this instance, Samsung may be more inclined to give AMD priority access because they developed HBM technology with Hynix, and indeed, AMD was due to have priority access to HBM2 as well due to early reports.
I wonder how this patent issue will affect Nvidia's standing with Samsung and whether it might end up benefiting AMD (and if it does... if and when AMD releases high-end Polaris - most likely near the end of the year.. we might see a mobile high end Polari with HBM2 - or at least, it would be nice to see that).

The massive Zen server APU's will reportedly be using HBM2 too - so it seems that was a good call on AMD's part (actually, consumer versions should be same, except that they might max out at 16 cores/32 threads).
It wouldn't surprise me if AMD pulls HBM2 for mobile gpu's... they need good wins, and I think they need to hit with every current advantage they might have.

 

I don't know much about the architectural differences, but from the looks of the Nvidia & Intel's grip, with enjoying years of profits, huge funded R&D, It's a hard game for AMD to catch up two biggest rivals in one shot.

I really want to see AMD comeback with Radeon atleast on par with GeForce on the DT end, this ngreedia's driver mess is too much these days & on mobile end it's rotten from the post 347.88 releases they are really slacking off. Yet my guess is again nvidia will be ready with the GDDR5X for mobile while AMD may take a shot at low-mid end mobile chipsets with GDDR5 only. I doubt they will make HBM2 onto MXM too which nVidia's own, has an upperhand plus considering the top end 980M and 980 mobile & beat them both, plus the cash investment, very unlikely.

As for comparing against Intel, I wont believe unless they beat an i5 for the first showcasing of their Zen, Intel is generations ahead AMD

 

Not really difficult for AMD to catch up.
Nvidia said they will at most be getting twice performance increase per watt vs Maxwell.
Polaris was stated to have 2.5 times more performance per watt vs Fiji.

Plus, AMD is working with 14nm... Nvidia with 16nm, so I would imagine they are doing all right and theoretically shouldn't have issues in terms of performance this time around.
Polaris is a huge jump for them... but AMD needed to make it in the first place, and I think they knew/expected what Pascal might bring, and focused Polaris towards that end.

As for Zen... it was reported it has 40% IPC increase over Excavator (read Carrizo, which actually has better single threaded performance vs Piledriver).
Also, Current AMD APU's aren't really viable to go against Intel because their 4 core APU's behave like dual core Intel CPU's.

SMT inclusion and IPC increases on Zen will address that, and quite possibly allow AMD to surpass Intel.
Add HBM2 to the IGP Polaris part, and you end up with an IGP with massive bandwidth which AMD IGP's need - this will also come in handy for HSA/HUMA use.

So, taking that into consideration, I don't think Zen should be underestimated (we also shouldn't over-estimate it either - but the numbers and estimations are based on what we know of Zen thus far and was officially released by AMD).

 

Has this been linked? Project F according to ex-employee on LinkedIn so grain of salt:

http://wccftech.com/amd-polaris-gpu-die-size-232-mm-2/

 

Some more entries in Zauba:
http://videocardz.com/58308/amds-baffin-weston-and-banks-gpus-spotted-on-zauba

Nothing of a high end it seems, though hopefully that Baffin XT will be an upper mid-range desktop card to replace 380(X) AND high-end mobile.

 

I'm guessing the goal for mobile polaris is to meet occulus vr minimum spec. I somehow doubt they can get performance of 980M under 50W.

If AMD manages that by some miracle, then its hello volume. every mainstream laptop manufacturer would want polaris in their laptops with "VR Ready" stickers

 

My apologies if we've seen these cards already, but today Lenovo announced a AMD Radeon R7 M460 shipping in their new Yoga 510- have we seen this yet? I googled that card but didn't find much....

http://liliputing.com/2016/02/lenovo-launches-yoga-510-yoga-710-miix-310-windows-2-in-1-tablets.html

 

That m460 is a joke! With a 99.9% possibility of yet another rebranded crap. If it beats Intel's iGPU it would already be great for them.

As for the minimum VR spec, that would be a really good marketing move, since not even a 980M fits in there! Unfortunately, they haven't mentioned anything about mobile during their VR targeting campaign, so we can just hope. Considering AMD's attitude towards high-end mobile, I really doubt they will come up with anything better than 970M-980M. If only they could bring a competitive product and attract attention of MSI and ASUS, that would really help, since those two companies constitute more than 50% of gaming laptops market share. Wishes wishes...

 

NotebookCheck folks expect it to be based on Radeon R7 M360:

http://www.notebookcheck.net/AMD-Radeon-R7-M460.160121.0.html

AnandTech folks also expect just a rebrand here:

http://www.anandtech.com/comments/1...-and-yoga-710-convertibles-at-mwc-2016/490019

Mobile Polaris shouldn't come before at least August 2016, so anything before that would likely be just rehash of existing GPUs.

 

"based on"

 

Due to problems with FF on both 16 and 14, I'd expect delays across the board. You'll all be happy to know that AMD expects to compete with the 980 Ti by 4Q..... Without a closed liquid cooler.

 

That's out of context, but anyway, if you follow the link

and look at the "XConnect GPU Compatibility List", you'll see: "Mobile dGPUs Based on the Above"

Something to get excited about?

 

Based on R9 300 Series GPU.

Also I was kinda excited about xgpu until I read Windows 10 in the requirements section.

 

Well, there are Furys above as well and it's not like there aren't 3 series GPUs already (not as widespread as I would like though). I for one still think that first gen HBM would be used alongside HBM2 and GDDR5. After all they've put a lot of money into it, and mid range would be fine with 4GB. Of course, as always - the more, the better, but that's why there's HBM2.

 

Yeah, it's like Seagate Momentus XT had portion of NAND modules which were already outdated in production of actual SSDs.

But I think AMD already gave up to the ODM standarts like MXM heatsinks standarts and other stuff which always slows down progress.

 

Unlike the XTs (the 8Gigs they shipped with was not enough even back then, let alone now), 4GB is still good for WUXGA (1920x1200) and even WQXGA (2560x1600), so again, why not in a mid range GPU? 4K? Playable as the Fury line shows, but you really have to put money into a machine for 4K. A mid-range on the other hand doesn't equal 4K (rather FullHD, or 3K), at least not in my mind.

 

Looks like Polaris 10 will launch in June "targeting minimum VR spec" at affordable price below $349 - does this suggest they will also offer a mobile GPU that meets min. VR spec?

http://www.digitimes.com/news/a20160309PD206.html
http://wccftech.com/amd-launch-polaris-gpus-june/

 

How underwhelming if true. There is nothing special about minimum VR spec, which is GTX970/R9 390 performance. Offering a mobile GPU capable of that... they already exists basically.

We need real improvement, not silly small increments. Wow us. Getting a 980m from amd is fine, but this should not be their end goal.

 

Been a long time since AMD has released any decent mobile GPUs. I'd be thrilled with even GTX 970 equivalent, which would be a bit faster than 980M, no?

 

@ryzeki - What we have on the other side to be so fast to dismiss 50-70W VR ready chip that can fit in pretty much anything (we've seen 50W chips in 13") and wont cost an arm and a leg?

 

Indeed it has been a while. We are hopeful, and yeah GTX970 is 15% faster on stock to 980m. I hope they go further though.

We don't have anything like that in terms of tdp, but we have far surpsassed that in performance. Good for thin laptops. I want actual powerful products i can use.

Obviously there will be stronger products, but how much will we have to wait for them?

I am just underwhelmed that its a product that gives me nothing haha. Obviously it will be a good product.

Sent from my SM-G925I using Tapatalk

 

Depends on how you look at it.
The GTX 970 is apparently more powerful than 980M, however (and, if I am not mistaken), Polaris will probably deliver this performance at a much lower power draw of 980M, essentially and potentially leaving room for high-end Polaris mobile chips that will go up against Pascal.

I think this might be part of their new strategy.
These Polaris chips could effectively help AMD recapture laptop marketshare and then release high-end versions of Polaris later (probably near the end of the year - and possibly in line with Zen 2016 release).

 

Yes, it should. This is why i want a high end gpu. Im not interested in the thin and light polaris. I want the polaris i can use. 980m performance is what i have. Give me twice or more!

Sent from my SM-G925I using Tapatalk

 

I'm not interested in thin and light either, but that's not something to be dismissed easily. It's an actual product that's going to happen and we can base our guesses on it. What do you know about nGreedia? Any performance values? Is there a single GPU showed and tested? What about the rush? You are quick to say that you are tired of waiting, but no one really knows how long you'll have to wait for nGreedia as well. Fun stuff.

 

Nvidia is even worse. They say pascal can bring up to 2 times the performance per watt among other improvements, but they haven't showed anything.

I am tired of waiting because we have had literally the same news from both camps, since their reveal. Nothing new or exciting. Hell, at worst we only get news of delays or only X products releasing, etc. That's why I am like this. Rumors were that the polaris products we will initially get won't be the high end stuff, but for the thing and light market.

I doubt I will be able to upgrade to a polaris gpu. It would be great as I prefer AMD/ATi over nvidia, but at least I hope Pascal does DX12 as it should hah.

 

I will attend CEF Shenzhen next month. You think there is any chance of spotting a Polaris or Pascal?

 

Let us know if you do hear or see anything But I don't know, never heard of CEF

 

Reportedly AMD will demo some Steam VR benchmarks with Polaris in a couple hours here:

http://www.ustream.tv/channel/gQETjZzLnf9?sf22465712=1

March 14 @ 4:00 PM PDT

 

AMD roadmap (subject to change) with Polaris, Vega and Navi:

Spoiler: AMD roadmap

This suggests Polaris won´t be HBM2, did we know this?
And look how close Vega is to Polaris.

http://www.anandtech.com/show/10145/amd-unveils-gpu-architecture-roadmap-after-polaris-comes-vega

 

Yeah just saw it. There were already rumors about neither pascal nor polaris using HMB and instead GDDR5 for their GPUs. I guess this sort of confirms it. This sucks, but at least polaris as an architecture is supposed to bring good performance increase.