Mobile Polaris Discussion

They also need to improve linux drivers. I'm not budging from NVidia if AMD continues to ignore linux.

 

http://www.pcworld.com/article/2956...-graphics-driver-to-catch-up-with-nvidia.html

Then there's this one from this year.

http://www.phoronix.com/scan.php?page=article&item=amdgpu-2016-details&num=1

 

Yeah read a few tidbits they were thinking about doing something about it. But c'mon, we've been told ATI/AMD was working on linux drivers for decades.

I am amused that a splash of NVidia logo appears before my login does on Ubuntu right now...

 

They're not thinking about it, it's happening. Linux kernel 4.2 and up. Tonga GPUs and up.

 

Here ya go

http://www.phoronix.com/scan.php?page=article&item=amdgpu-45-next&num=1

 

I'll still wait for results of Polaris.

 

Considering that AMDGPU is made for Tonga, Fiji, and Arctic Islands and beyond, making everything prior a legacy card, it wouldn't make sense for AMD to not continue to optimize their drivers. They already put the investment in to have the framework built into the kernel itself after all. Nvidia is nowhere near that point......

 

AMD had mantle and yet... it's total failure. I'm not holding my breath because even though NVidia seems behind, they have a knack for succeeding and prevailing.

If AMD can produce something better than Pascal for a discounted price, then I'm all for it, but not holding my breath. I don't expect AMD to succeed at anything anymore.

 

It's not a total failure. It worked in the games that used it. The only reason it wasn't a bigger deal is because nVidia pays off developers not to use it.

 

Mantle did not fail. It evolved into DirectX 12 and Vulkan. Speaking of which, AMD already has Vulkan drivers for Linux.

 

haha, people trying to think about a graphics card that's over 6 months from release and trying to comment on what is has and also on the power draw of it.

I am fine on what I have now, 2 laptops, one with 970m sli and other with 860m.

 

I really don't know why you lot are trying to make an effort to please that NVIDIA troll Zymphad.. It isn't worth it.. He is just so so biased against AMD..

Sent from my LG-H811 using Tapatalk

 

That is what speculation is. If you don't like it so much then make it so we wouldn't even know about your existence in this thread.

 

he has the nvidia logo as his avatar ffs lol

regardless, I don't expect HBM on MXM this generation, from AMD or nVidia. Sure would be nice though

 

I really don't understand why people feel the need to be biased towards Nvidia or AMD. Do you not understand how good it is for we the consumer if both of these companies keep up with each other.

I could honestly care less who makes my GPU. I will buy whatever gives me the most bang for my buck and for the past 15 years, I've bounced back and forth (not so much in the past 6 years).

I have been using Nvidia GPU's in my laptops for the past 6 years but would love nothing more than AMD to create a mobile powerhouse on par or better than Nvidia offerings.

 

Its ok, I will continue reading and get a good laugh.

 

same here. The issue is that Nvidia keeps making good upper medium and upper tier graphics card. My current card is the 780m-970m (upgrade), the last before that was 7970m, then 6990m.

 

This your first speculation thread on NBR? This happens all the time and for some of us it´s part of the hobby.

And while I wont go as far as others and say release is imminent, 6 months is actually not that far away considering releases are often preceded by leaks, benches etc.

 

I don't think that the 'speculation' is off track too much, because we are trying to make reasonable extrapolations with baseline numbers that we saw in the demo-ed Polaris version - so we have baseline data (which is slated for additional improvements).

 

AMD Radeon Crimson on Linux sucks nuggets. HAHAHAHA! This review makes me smile.

 

AMD stated gains in Linux were lies. Love it. Are AMD engineering incompetent or are their marketing off their rockers? That was the question asked.

 

So, finally some new info about that other, faster, GPU that AMD plans to launch in H2 2016 - Polaris 11 - coming from analysis of AMD presentation at VRLA expo:



https://forum.beyond3d.com/threads/...ors-and-discussion.56719/page-33#post-1893328

AMD apparently aims to target minimum VR specs for Oculus Rift CV1 with a new Polaris GPU coming in H2 2016.

Only two AMD Polaris GPUs that they spoke about planned for H2 2016: Polaris 10 and Polaris 11.

Polaris 10 is that GTX 950 performance level GPU for "console level performance in slim notebooks" (demoed running Star Wars Battlefront at CES).

Which leaves Polaris 11 as that min-VR-requirements performance level GPU. So the performance should be roughly at or a bit above GTX 970 or Radeon R9 290 levels of performance, just with lower power requirements and cheaper price (under 350 USD level).

There will be other faster Polaris GPUs too, but these two (Polaris 10 & 11) should be closest to actual non-paper launch (on the shelves soonest at August-September 2016).

 

Hmm, a bit disappointing since the top level performance is around a overclocked 980m. This doesnt encourage nvidia to try...

It is interesting though if under 100w.

 

Very disappointing. Still no word if there will be a MXM option. Until I read something definitive, I'm assuming AMD hardware will all be BGA rubbish with Enduro or whatever they call it now POS switchable graphics system, blame on Intel as well for that disaster.

Simply put, there is zero enthusiasm from me on AMD hardware until I see a laptop that has 40% more power than my 980M overclocked and has no switchable graphics at all, not even an option for it and MXM.

 

Wow... How seriously disappointing... Can't say that I'm surprised but still.

 

What is disappointing? That you can have a VR ready GPU in a ~80W envelop? That even toastbooks would do VR by the year's end? Now, imagine the bigger chips.

 

What's disappointing is that AMD doesn't have a high end part coming. Sorry but nVidia is going to have yet another year on top if AMD is just going to push Polaris 11 to GTX 970 levels. That makes me wonder if they are having problems with 14nm FF.....

 

You want them to give the exact specs and performance so nGreedia would know where to dial their performance. We've seen a handful of such examples before, why it should be this way again.

 

AMD only has 10 and 11, they said so themselves, 11 is focused for 970 level performance to offer a product for VR. It isn't like nVidia isn't delayed anyway.....

 

I wish to think they play it conservative and this is indeed what they have, for now. There still is like half an year to mature the process and then draw the big guns.

 

You cant exactly get out another die on a brand new process within half a year.

 

With Pascal nearing, expecting AMD to beat Nvdia at top performance is like expecting a man who can barely walk to beat Usain bolt.
So is polaris disappointing to us - yes, is it a logical step in right direction - also yes.

 

Lol comparing amd to nvidia as a cripple to bolt..that is a tad harsh haha

Sent from my SM-A500FU using Tapatalk

 

Please, explain why?

 

When have you ever seen a GPU that isn't using "BGA rubbish"? You MXM card is a soldered BGA GPU with soldered memory in a custom PCIe card format. AMD needn't say anything about MXM as it is just a custom card format.

So you have some phobia against switchable graphics? Can't understand why, one doesn't have to use it.

 

Benefits of MXM GPU over actual impossible to replace soldered onto motherboard gpu.

1.) Cheaper to replace
2.) Universally seem to be better made (OEMs cut corners with true BGA) Effects as follows on said "BGA" gpu's:
a; bad at overclocking.
b; far more likely to fail after a fairly short time.
c; usually equipped with inferior cooling systems.
3.) Something people don't care about but should THE ENVIRONMENT
4.) Far better support from enthusiast community. Good luck getting nice modded vbios' that actually unlock full potential from the bga versions.

So it isn't that BGA couldn't be as good it just won't be as good EVER as MXM based graphics as it will always be made cheaper and as part of a disposable package.

I run a nearly 7 year old laptop. It doesn't even have a trashy igpu and it can keep up with any modern laptop (in games) even though it is more than twice my daughters age!

It is pathetic. Switchable graphics, bga the lot of it. AMD are not considering MXM at the moment for one reason: survival. I hope they do survive as the current crap is all because of the monopolies that intel and nvidia now run. Happy times!

Go troll elsewhere mate.

 

Where exactly is it explicitly stated that AMD doesn't have a high-end part coming?
Unless I am mistaken... nowhere.
All we know thus far from AMD is that Polaris 10, and Polaris 11 exist, and that Polaris 11 is supposed to replace current offerings for minimum VR specs at a lower cost and lower power expenditure (while offering most likely higher performance).

Polaris 11 could also result in high-end chips down the line (think 980ti like performance in mobile by the end of the year - although this is just a hypothesis right now), whereas earlier releases are likely ones which we were told about to date (and will be seen in desktop and laptop solutions earlier).

I doubt that AMD would limit themselves to releasing entry and mid level range Polaris parts, while not having high-end solutions in the mix.

 

I guess Meqol means that one AMD chip can be both MXM based and soldered to mobo.

 

How? Polaris 11's performance is stated to be around 970. You cant change a die on the fly.

Vega 10 didnt even get any official announcement and I severely doubt we will see that in mobile by the end of the year.

 

First you need to design said die, tape it out, do testing, etc. Since its a brand new process, your yield is gonna be crap. So your first wafer chips may not work properly or not even have enough working dies for testing. This would be very very bad. Usually the first batch of chips are very high value and its vital to get those chips rushed to testing ASAP. If anything doesnt work, its back to the drawing boards again and run another revision of the die.

Just from first tape out to release usually takes at least 6 months on a new process without any problems. We are not even talking about design. You dont rush out a brand new die in the middle of your line up because you need something to fill your high end mobile space. Thats outrageously expensive and serves very little purpose. The segmentation AMD done suggest to me that Vega 10 may not be as impressive as we think.

In other words, we have polaris 10 filling in the lower end and polaris 11 at 970/980 level filling in the mid end. Just looking at the performance gap between mid to high end, I dont think Vega 10 is going to be anything over 30-40% faster than a titan x. Vega 10 is probably at around GP104 level. AMD would probably have another die to compete with GP100 once nvidia is moving to launch that next year. If thats the case, there is barely any space to launch another line based on this new die without cannablizing your own sales. You would literally have 5-10% performance gaps between your products.

AMD is probably using the low end polaris chips in a way to learn how to properly get their big die on 14nm without too much issues. This is a smart decision and nvidia does the same thing now. You dont swing out with a big die on a new process, thats just asking for trouble.

 

You never know what they have in the bag.. AMD isn't stupid enough to reveal all its cards.. Just wait and see

Sent from my LG-H811 using Tapatalk

 

May I ask where was it stated that Polaris 11's performance is maxed out at around 970?

If I'm not mistaken, the presentation stated that Polaris 11 targets minimum VR specifications (which are around 970 levels), however this doesn't mean that AMD doesn't have a higher-end Polaris 11 in the mix.
The initial Polaris 11 could simply be a cut down version of a high-end version which is suitable for VR and earlier release date (around summer).
A high-end model might still be in the works... basically, non-cut version of Polaris 11 (if the 970 level chip is a cut down version to begin with).

Polaris offers 2.5 times performance per watt vs Fiji. I would find it extremely odd (not to mention disappointing) if AMD didn't have a higher-end model in the works, especially with Pascal being released around the end of the year (also looks as if there will be no HBM2 for Nvidia, so they will be going exclusively with GDDR5X).
Besides, weren't we told that high-end Polaris chips will come out near the end of the year anyway in the first presentations at CES?

P.S. AMD might just be biding its time with early Polaris releases (summer) by providing a direct competition to Maxwell offerings and closing a 'gap' before releasing high-end products at the end of the year (which would make sense - release the early models soon, and save the best for last).
And if early Polaris 11 indeed has 970 performance... that would make it about 20% faster than 980m - which isn't bad when you also consider the premise it would also consume less power (if its less than 80W, or around 75 W), then we're seeing 40% power reduction vs Maxwell - and AMD did mention they have more optimizations to do - so they might be able to squeeze another 5 to 10% performance at half the power consumption of Maxwell - which wouldn't be a bad thing at all if they can do it.

The only way I could see AMD not having any high-end mobile releases is if they set their power consumption at around/less than 100W, and I hadn't seen info from AMD indicating that they 'abandoned' 100W and above thermals.

Dunno how much more performance could 25W bring (or 100W Polaris)... though I suspect at least a desktop 980 might be doable, considering that lower manuf. processes have less leakage and far more manageable thermals.

Now, if AMD used HBM2 too, they probably would be able to gain OC-ed 980ti performance at 100W (or 115W), but with standard GDDR5, we're likely looking at 980 desktop level performance at about 100W for mobile high-end Polaris.

 

The very reason I asked you to explain, so we can level.

If I understand correctly you're saying that they can't bring-up a new core for 6 months, because they haven't planned it yet, let alone making it, right? Well, that's where I'll repeat myself - they made the first batch with lower end chips, just to test the waters where the new process and the new design meet. Obviously they have working chips. They said there is some more tweaking to be done, so give them a two more months for that. Then comes the plans for the bigger die - why do you think that there is no such thing already on a drawing board, or on a hard-drive somewhere? Tech corps usually plan their stuff well ahead of time, that's why there are such things as roadmaps. Well, that's why there are other things called delays, but let's focus on what we have for now. So, they have the new low and mid chips, claiming that they need some more work, I'm pretty sure they have the designs for the top, they just wait to study and debug the process/design. I still think that it is possible to see top end announcement within 6 months.

 

@triturbo

The bigger die is called vega 10. I talked about it. Its would the full performance die. However, do not expect that on the mobile market. We have never gotten the flagship die on a mobile since 480m. And no, 980 desktop is not the flagship card.

I never said amd dont have a higher end die. What I said was I doubt there is any other design inbetween vega 10 and polaris 11. As for the design stuff, I wont be surprised they already are working on finishing their ultra big die design to deal with gp100 later. Companies does this years ahead of production.

@Deks
Polaris 11 is its own chip. Sure, the first version might be cut down, but cutting a chip from high performance to mid level directly at release is unlikely. Hence later I said its somewhere around 970/980 to account for that. You do that later to fill missing SKUs in your line(see 6930,7870xt, 465 etc) and a lot ofthose times, the cutted chips are either too good that it cannablize the lower end's sale or too poor performance that its cheaper to just get the mid end card and have the same performance.

 

Meanwhile another GPU has been spotted in Zauba, something very expensive though. Wonder what is it:

http://wccftech.com/amd-flagship-polaris-spotted/

 

The only reason there would be a cut down chip on release is if that spin isn't ready for prime time and they push it out to market anyway which would really spell out doom and gloom for AMD. Seriously, if they are targeting Polaris 11 at 970 levels then the chips must have issues with heat or other unforeseen issues that they don't have time to fix before marketing. It isn't a simple matter of taping out a die. You tape out a die, test it, then fix bugs, respin the die, debug again. You usually don't get a usable consumer product with an acceptable yield until the 2nd, 3rd, or even 4th spin and that all takes time, something AMD doesn't have.

 

So we are supposed to believe that AMD got priority on a process that was just finalized at the end of 2014 and was in the process of being implemented at GloFo pushing out chips in January?! Yeah. Right.

 

I am seriously just thinking that polaris 11 is meant to a mid end card designed around 970/980 performance with an 100w tdp. The top end chip is being refined to deal with whatever nvidia comes up in GP104.

AMD learned its lesson on launching flagship first and having to do another refresh of it in 28nm launch. The mid end market is the most profitable market. Since AMD is in survival mode eith their debt due at 2019, heavily marketing their mid and low end is not a bad decision. The problem with AMD is that both Zen and Polaris need to be massive success. A moderate success will not save them here.

 

That and also if this is their fresh dies, the value wont be just 1700 usd and it wont be trusted to this shipping company.

 

Oh I agree with you completely but pushing out a midrange at 970/980 levels is just asking for nvidia to stomp them. TDP matters in laptops but we have already seen that nvidia can dial back GM204 for mobile if they really want to. On the desktop, people tend to be far less concerned with TDP. I mean I know people who buy 1200 watt power supplies just because they can. Efficiency is the last thing on a desktop gamer's mind.

I hope for AMD's sake that they can push their big die sooner rather than later.

As for Zen... No chance. I don't believe it is possible to catch Intel at this point with their lack of financial resources. We shall see I guess.