nVidia 2015 mobile speculation thread

I saw this this morning. It is good, although some are trying to spin it. They also canned the vp for pr and are putting it under the head of the Radeon Group! As Greenland is taped out, you won't see the fruits until 2017 of an unrestricted design or compromises made for other goals. I still bet Greenland will be a wet dream and will easily go toe to toe with Nvidia (look for articles about amd fury x supercomputing monster - http://www.vrworld.com/2015/09/07/amd-r9-fury-x-potential-supercomputing-monster/ ). No one else comes close to those numbers (like bitcoin was with 290x multiplied).

 

awesome article, i had no idea the fury x was such a number crunching monster! great news for AMD getting sum market share in the professional high-profit-margin department

Sent from my Nexus 5 using Tapatalk

 

Good news but 2017 is too little too late. They need to get out competitive products now.

Sent from my VS980 4G using Tapatalk

 

such as the fury nano in mxm. its like....captain obvious knockin on your door, AMD! do it, dammit!

Sent from my Nexus 5 using Tapatalk

 

As I said, fury series I do believe is competitive. Greenland definitely will be as well. 2017 is where I see a pull away.

 

Well Alienware/Delll posted the R9 m395x driver for the 15 R2. I decided to download it and check its ID. You can see them below.
AMD6921.1 = "AMD Radeon R9 M390X"
AMD6921.2 = "AMD Radeon R9 M390X"
AMD6921.3 = "AMD Radeon R9 M390X"
AMD6921.4 = "AMD Radeon R9 M390X"
AMD6921.5 = "AMD Radeon R9 M390X"
AMD6921.6 = "AMD Radeon R9 M395X"
AMD6921.7 = "AMD Radeon R9 M395X"
The R9 m295x is no where to be found because the m390x is the m295x (I've looked at the 15.7.1 driver I have for my R9 285, and only one entry of the m390x is present, which is the italicized one above). So essentially, the R9 m395x is a higher clocked R9 m390x / m295x.
Hopefully it's better than the GTX 970m and close to the GTX 980m. While older drivers had up to 6921.6, 6921.7 is a new entry (I have the old Omega driver for some reason, and it has 6921.1 - 6 as R9 m295x). I'm hoping we see the 6921.7, but I could be wrong.
Has anybody checked to see if new "OEM" drivers (might be wrong term, but it's the best I can come up with right now) have appeared for Nvidia GPUs? Dell's drivers for the upcoming laptops are based on an old 348.34 driver and recent 353.54 driver (different OS drivers), so I don't think they'll have anything to show.

 

If the R9 M395X is Tonga, and is even a full R9 380 in mobile form, it's still barely better than a GTX 960, which means it'd be at best ballparking the 970M, and likely with a larger power requirement.

It's the same thing I've been saying about AMD's current lineup; there's a HUGE jump in required TDP for even their "midrange" cards. The R9 390 competes with the midrange GTX 970, but even with constant voltage on the 970, the 970 is still 25-50W less demanding. Those cards can't be suited for mobile chips, and the R9 280X GCN 1.0 cards have been axed for Tonga in the current lineup, so... it's problematic.

AMD doesn't really have non-Fiji tech that can fit suitably in laptops. Even the Fury Nano with its fluctuating clockspeed requires too much power to suitably fit in a laptop format. If they had a decently cut down Fiji card (let's say 3072 cores instead of 4096) and applied the R9 Nano tech to it, we might get it down to a ~120W envelope suitable for laptops, but then what'd the power really be like? It'd have to have a neutered clockspeed as well as the neutered core count, and could very well end up trading blows with the 980M at the end of it all, while still drawing more power and running hotter (limiting OCing). And that doesn't even handle the upcoming 990M which we basically have proof is full GM204. Even if the clockspeed is knocked down a few pegs, a 680M --> 780M type increase might very well be expected, and that'd mean that AMD's new card would instantaneously be obsolete.

2017 is WAY too late for a new card as @HTWingNut said, but what do they have now that could hope to be competitive?

 

The r9 m295x / r9 m395x have been / is full Tonga, but with a 256-bit Memeory bus (I think that's what it's called). Apple's R9 m295x did not even match the GTX 970m, so it may be unlikely if the m395x will beat the GTX 970m. Or it could, but it would need a higher clock speed than Apple's R9 m295x.
As for Fiji, I wouldn't be expecting it or HBM on laptops until late next year of 2017. It's not surprising Tonga is used again since it was known the r9 m395x was similar to the r9 m295x. Just at least it isn't a re-name. But for HBM to be used in laptops, it'll be interesting when it happens.

 

Tonga on desktops has a 256-bit memory bus, so there's no change.

 

Next year they will basically have almost the entire line, if not the entire line, on 14/16nm. The fury line up is similar to a 28nm version of what they have coming, except they'll have hbm2. It's still Q3 or Q4. As to your assertions on the card, the nano is a full fledged x with reduced clocks primarily. Oc wouldn't be as good, but you could take the fury, which is a cut down core, could be underclocked down to be in the thermal envelope. More power draw, sure. Don't know what dx11 performance would look like. Powerful number cruncher, definitely! So most would still go with Nvidia this year unless you had amazing cooling. Other than that, the power draw would be the same as an oc 980m for somewhere between the cut down clock and full clock. Should be possible with 330 plug.

 

Umm... I never made any comments on the R9 Fury Nano. Did you quote the wrong person?

That's right. I was thinking of the rumored R9 380x.. Don't know why, though.

 

Yep, exactly! On that topic, it looks like the Fury Nano is just as power efficient in terms of fps per Watt as Maxwell now, see the Power Efficiency graph on Toms Hardware:
http://www.tomshardware.co.uk/amd-radeon-r9-nano,review-33301-9.html

Good news for any potential mobile variant!

 

the fury nano is already in the range to be converted to a mobile gpu. as a rule of thumb, anything in the range of 175-200W desktop wattage can be converted to mxm format with identical specs. weve seen this before with 680M/7970M and their desktop counterparts

Sent from my Nexus 5 using Tapatalk

 

The chances of the Fury Nano coming to mobile are slim to none right now. Don't expect anything competitive in mobile from AMD until 1H 2016 at the earliest, most likely 2H 2016. AMD needs to focus on recapturing desktop market share before they can even begin to repair their relationship with customers and OEMs in the mobile market. People hear AMD and mobile and think of unreliable drivers, Enduro, rebrands, high temperatures, high failure rates. AMD can't afford to take a gamble right now.

 

Yes, we've seen that in the past. My only concern with that with regards to the Fury Nano - have they already done all those power optimizations that they would normally do for an MXM card...have they already done those modifications and just applied them to a desktop card. I wonder how we'd find that out, perhaps take a Fury X and downclock it to Fury Nano levels & then see if it has the same high fps per watt efficiencies - if it has similar power characteristics then I guess we could assume that there might still be some further optimisation to be had through binning (etc) to convert to an even more efficient MXM card.

 

380X is still MIA so Tahiti XT is a possibility, but most likely it will be Tonga XT. The whole point of Tonga was to reduce manuf. costs for AMD and replace Tahiti. Tahiti was one of their most popular GPUs for the longest time and they could achieve similar performance by cutting the 384-bit bus down to 256-bit coupled with new memory compression.

Keep in mind the only reason Fury Nano form factor is possible is because of HBM space savings. Architecturally speaking the Fiji GPU is still less efficient than GM204 and GM200, which is apparent when you look at Fury Nano power consumption when power limit is increased to prevent throttling and when overclocked. Nano is simply a binned Fiji GPU with "intelligent throttling". The bulk of the power savings come from the reduced clock speed.

http://anandtech.com/show/9621/the-amd-radeon-r9-nano-review/18

 

Thanks for that link. They say that they've binned the chips already for the Fury Nano, but they don't say how they came to that conclusion, I mean the voltage difference between the Fury X and the Fury Nano at 1000Mhz is only 1.212V vs 1.200V - a 0.012V (only 12mv !) reduction is not much of a binning effort is it, doesn't seem to be much of a binning effort to me? This leads me to believe they could do some more aggressive binning for any mobile MXM products that could be made.

 

Because that 1.2V is only at max clocks and it's well past the knee of the efficiency curve for Fiji even binned.



Notice how it stays fairly linear from 815MHz/1V to 900MHz/1.07V but then starts to climb rapidly after that. Eyeballing it, the slope from 800-850MHz actually looks steeper than from 850-900MHz. It's no wonder the sweet spot for Fury Nano with stock power limit is somewhere between 850MHz and 900MHz.

 

Yep, I'm aware of that, but that doesn't prove that the chips are binned specially for Fury Nano - as I said there's only a 12mv difference between the two, how do they know they're binned?

 

Sorry, meant to quote D2

 

So far, it seems so.

The same applies to nGreedia, but somehow people chose not associate it with that. Optimus is not something to write home either, but it is the lesser evil, I'll give it that. Not to mention they are the inventor of rebadging. Also not to mention that they had failure rates with FX series, the 6xxx series, the 7xxx series, and the mother of all failure rates 8xxx series, the list goes on and on, high end 6xxm and 7xxm started popping here and there as well... High temps - 480m, 580m, 880m. And this dreaded drivers. Most of the people here use nGreedia, please do share, are you running latest? Are you ALWAYS running latest? I'll tell you what I see in 99% of the cases - people say I run XXX.XX because YYY.YY suck or screwed something and etc. How very reliable, right? Yet still blame AMD for drivers! Like so very NOT hypocritical, right! I know. What I'm trying to say? It happens, deal with it, just don't point fingers one direction while the other is exactly the same (OK, with slight variations)! I wish we live in a perfect world, but we don't. Oh, and that's how all the "hearing" starts and goes, because "Why do you look at the speck of sawdust in your brother’s eye and pay no attention to the plank in your own eye?"... Just saying.

Isn't this EXACTLY how Maxwell's efficiency works as well? Clock it anything different than the stock controlled clocks and there goes your efficiency!


This is not related to anyone particular, I'll say something that I've said multiple times, but I'll say it even more so, since it is obviously needed - everyone says free market - I get what suits best my needs and what is the most powerful. Fine, there's nothing wrong with that, that's how free market works. The thing is everyone seems to forget about the other side of free market, when looks at the competition and asks why they can't act accordingly. Well, those chips everyone is talking about, they come from the same fabs, they cost THE SAME for everyone, unless Apple, since they make ORDERS (in quantities never seen before, or why there's no 20nm for us), but that's another story. So here we have AMD having chips costing same as nGreedia's ones. Have you ever seen how both go on prices when they hit the shelves? Now, have you seen their market share? Combine all of the above - do tell, where exactly AMD is supposed to take money in order to stay competitive? They do pretty damn AWESOME for what they have!

It is true, they can't, that's why they secured the consoles and that's why they made a deal with Apple. They have to survive somehow. The PC market is MASSIVELY grIntel/nGreeida biased. Even in AMD/ATi's strongest years, they were just a few percentages away from their competitor. You can see how it goes when they are weak...So yes, they can't count on the HUGE fan base that grIntel/nGreeida have and would buy them no matter what they throw at them.

The R9 Nano is perfect for MXM, but is anyone here honestly believing that the PR runner-up (first place belongs to Apple hands-down) would leave the tag line "The first HBM MXM/The first HBM mobile chip" to belong to anyone else but nGeeida?

 

Yes. Yes I am.

The one exception was when overclocking was completely blocked via driver. Since that was solved, I've been running latest with no issues, just as I was before.

 

@triturbo I wasn't saying that nVidia was doing much better lately but unlike AMD they can afford to experiment, tick off customers, etc.

However I have seen 4 or 5 posts in the last few days about dead 6970M and 7970M in the Clevo/Sager forums and I haven't seen a dead nVidia card except for an 880M that might have been killed by Windows 10 in months... From an outsider perspective (I've never owned a mobile ATI chip, just a 9800 Pro, X800 GTO, and 5830 in desktops (the only one that failed was the 9800), I see recent AMD mobile cards as being extremely prone to failure but I don't see the same thing with nVidia cards. That gives me pause and a logical reason to avoid their cards until they prove otherwise. NVidia driver issues I see, especially with the Windows 10 releases and newer drivers for Alienware machines being crippled. My experience with my Clevo has shown that every driver branch I've tried works fine for the most part though and I've not run into the issues many have.

To be perfectly honest, I have had a 7900 GT and the 9800 Pro fail on me out of all the graphics cards I've ever had. I don't count the cards that my Clevo has killed because that is likely due to a faulty motherboard which is why they have had my machine at Sager for almost two weeks.

I agree that competition is a good thing and anything that can make nVidia stop with their business practices that are shady to say the least but I don't trust AMD reliability based on what I have seen and I know that I am not alone in that thinking - especially since a lot of people who share that same opinion have actually experienced failed AMD cards.

AMD needs to rebuild trust before they launch another mobile chip.

 

I'm waiting for the official result, but my 970m may have died. I've also seen a 970m died in one of the alienware cases, that 880m you mentioned, and I forgot the other card that died. Just to give a little more. Then control for age of card, market share, failure rate, and you get your answer. Those amd cards mentioned are pretty old in comparison, but these other failures could be manufacturing defects (hence needing to control by percent). This is why I try to speak truth on both. Amd has nothing out currently competitive on the high end mobile. Period. Nvidia hasn't had a great driver since December or before. Amd takes a couple months to optimize drivers after release. Each has ups and downs. I agree people give Nvidia a pass. I think sometimes people are hard on amd. But that is what I think, not facts like above. I'm a fanboy of amd tech, not necessarily their products. Their tech is phenomenal (pun in there somewhere). Execution has been wanting. Nvidia was short sighted on where the market was headed, but has enough cash to figure it out quickly.

With that said, I'll side with ethrem in so far that this is becoming a partisan peeing contest. This is an Nvidia upcoming card forum, so discussions of amd should be of the type surrounding market changes and how it will effect Nvidia's decisions and offerings. I'm guilty of going offtrack as well. But we'd all love to see a 990m (980 full) square off against a fury (Fiji based) mxm. Sparks would fly. In fact, the 990m with a full 980 chip may have been fear of a Fiji based mxm coming. We can't know until things are released. I do think recent changes in amd, both structure of the company and product design, have Nvidia wanting to hit it hard. But, as designs on both sides are taped out, we must wait to see the effects on recent blows for almost two years. But I'm hoping the swings now have spurred a renewed vigor in both sides!

 

It is, but it is still more efficient, and when overclocked, which Fiji can barely do

 

There's a 12mV difference at the top. If a Fury X limited to 175W is capable of the same voltage/frequency curve as Fury Nano, then you're right, the Fiji GPUs used in Nano are not binned. But binned or not, it's obvious efficiency goes out the window for the Fiji architecture at 1000 MHz. So the 12mV difference in max voltage doesn't prove much since Fury Nano was never designed to operate there anyway, and when it is forced to, it is essentially a Fury X.

 

I don't think those Nano's are binned, so I reckon AMD could make a more efficient binned version for MMX at some point.

 

Yep, I've never had problems with NVidia drivers either, I use the latest.

 

Again, you don't know that. There isn't an apples-to-apples of a 175W Fury X vs. Fury Nano in regards to voltage/frequency scaling.

 

Agreed, I don't know that - it's just what I think based on the comparison of the voltages (at the top end).

 

That's exactly what I meant with "screwed something", so that's a no in my book.

There are, I haven't said nothing about it, but first - they are older, secondly it seems odd, but I haven't seen much if any 680m or 780m dead topics around, but I've seen quite a few GPUs, some of which were for sale over eBay. So give them some time, they will pop, definitely 780ms. Also there were failing 980ms (not sure about 970ms though) since day one, problem seems to be resolved afterwards. Haven't heard of such AMDs. As for Windows 10 - that was quite the nightmare, especially for @Mr. Fox and a few other users, who ended-up with useless displays on their Alienwares.

I do realize that I sound like an avid hater, but I had nGreedias before (one of which was out-of-the-box defective FX 5500), and a couple of months ago I was about to get hynix (I do HATE gnusmas... passionately) vRAMed 780m, even after the clock block and the 970 fiasco. Then GameSux happened, well that one really pissed me off. Again, everything is not perfect, GPUs fail, drivers suck, it happens, but these 3 things are INTENTIONAL!!! That's the difference! What's REALLY frustrating is that people give A LOT more fail tolerance to nGreedia, than AMD!

Well I for one never had an ATi/AMD to fail on me and I've had/have a few *knocks on wood*

They have nothing out, I've already posted what I think is the reason. Their tech is indeed phenomenal, but as I said in another thread - ideas belong to creativity, implementation on money, and as I said, AMD are obviously lacking the later. That's exactly how both grIntel and nGreedia operate most of the time - brute-force-money-throwing, because they have them (money) by warehouses.

Let's talk upcoming then, with a quick glimpse at the past for one last time. When 980m was released, I thought that it was a joke. nGreedia obviously could've released the full 204 chip, but they didn't. Fast forward now - we are talking about two-freakin'-hundred-watt-mobile-GPU!!! I mean, how to take them seriously? AMD has always released whatever they had best on their hands. Anyway I admire the balls behind such decision and I'm eager to see this system in person. To be honest I was dreaming about this very setup (OEM made) ever since I hand-made a hybrid cooling for my 5920G and that was couple of years ago...

 

For "great" nvidia drivers for example the new ones for Metal Gear V are crashing Hitman Absolution....

 

People waking from sleep have noticed serious problems with artifacting too.

 

AMD have always had a perception problem. Even when they beat Nvidia to DX11 by > 6 months, and all Nvidia had to show at the end of it was the pos that was Fermi, AMD still had less market share. The same was true for Athlon vs the P4.

Except unlike Intel, Nvidia are not only greedy, but openly arrogant. They're the most unlikeable company on the planet. I hope that next time I go to buy a notebook, there are more out there with AMD GPUs as an option. Giving Nvidia my business with this 970M felt gross and the terrible W10 drivers are icing on the cake.

 

Athlon and p4 came down to bribing oems not to carry amd. There was an antitrust case on it. But damage done can't be undone by lawsuit.

 

Only because this is an awesome discussion on the development of HBM, the upcoming standard for both sides new cards, am I posting this here. AMD isn't credited for GDDR3, GDDR4, or GDDR5, but it still shows what was developed when to bring us to now.
http://www.kitguru.net/components/g...o-work-on-hbm-technology-nearly-a-decade-ago/

 

R9 Nano in mobile. 2 problems:

- Does current MXM specs support a HBM card? With a huge GPU package in the middle.
- R9 Nano is already a underclocked full Fiji. Ranging all the way down to 800MHz to meet the 175W TDP specs, plus it runs on lower voltage than Fury X. A 100W/125W mobile Nano would have to be what, 600MHz? Would that be stable or financially wise?

Personally I think for a HBM card to go in a MXM card, MXM sig needs to make new specifications for the OEMs to follow (will we get HBM on MXM or will they all be soldered....). And I think AMD rather wait for 400 series that have smaller HBM chips for mobile instead of wasting full Fiji away on mobile and running them extremely low (say 600MHz instead of 1050MHz Fury X). They rather supply the desktop GPUs that can run them on full speed. Remember that AMD have capacity problems with the Fury cards.

 

On the Nvidia side of things, notebookcheck has put a GTX 980 (Notebook) page (the 990m).

 

Do you all think this new gtx(notebook 980) will replace the 980m in terms of pricing or will it be a more expensive config?

 

No such problems here.. I think the problems are really in Optimus systems...

I know.. I would love to buy AMD again however they have not put anything out which is worthwhile spending money on so NVIDIA will get my money as long as this goes on..

 

Desktop users have complained the most, actually.

 

As far as disgusting companies go, Creative Labs still takes the cake for me

 

There is enough space ( src):

Power-wise Nano isn't much worse than MXM monsters like 880M, which consumes much more than 100W.

Personally I have no problem with BGA soldered GPU if there's a Fiji option.

 

Considering AMD's mobile GPU track record since the 6990M, I don't think you really know what you're saying.

 

because 7970m has a very high crap out rate after a year lol?

 

...or two in my case was still an awesome card though, beat my previous 485M by more than 50%, and that in a single gen jump, crazy!

Sent from my Nexus 5 using Tapatalk

 

Pretty much =D.

Tis what a new process + die shrink does I think. The gains for Pascal are likely going to make Maxwell so obsolete it'll be a joke.

I don't think I'll ever get to make a jump like I did before though. Core 2 Duo 1.83GHz + Intel GMA X3100 --> i7-950 + 280M --> i7-4800MQ + 780M SLI

 

Just took another look on notebookcheck's 980 (notebook) page, it said it will release at 9/19, em~


Boring boring Chelsea

 

That's 9/19/2014 according to NBC. Sounds like old news to me.

 

Oh, I didn't notice that, thanks for the reminder.


Boring boring Chelsea