nVidia 2015 mobile speculation thread

Hasee, a big chinese notebook manufacturer, had a conference in china few days ago.
Wu Haijun, the CEO of Hasee revealed that GTX 990M is coming in Q4 (September - December). Probably close to September to not put it too close to Pascal (June 2016 ish).
And the biggest news is that GTX 990M will be just about as powerful as GTX 980M SLI!

I have an idea what GPU we are looking at, I was playing with the idea earlier in this forum, and I think it is GM200 with 2500 cores. That should equal 980M SLI (3072 cores minus SLI scaling)
GTX 980 with 2048 cores have a TDP of 165W. Meaning a desktop GPU with 2500 cores will be around 190W. GTX 680 with 1536 cores was 195W, and we got GTX 780M with 1536 cores from that one.
Yes I know GM200 is 384bit, but if they can`t that to work with GTX 990M, they have disabled 4 memory controllers and it will be 256bit.



From the conference:






From the Hasee livestream (notice it says 990M = 980 SLI, but that must be a typo since the chinese that watched the livestream is blowing up the notebook forums right now with "980M SLI!!!")





Laptopvideo2go also noticed a new entry in drivers a week ago. I have been meaning to post it earlier but was lacking content to put it context. The codename is weird, not seen before, so it could be this massive card from Nvidia




So this card should be worth waiting for if you can

 

A single mobile GPU based on Maxwell that performs like a GTX Titan X? Sounds way too good to be true.

I'd be able to believe that if it were Pascal. But I suppose if it is true, that's damn impressive, and I want one.

Notice the big if...

 

990M would have to outperform Titan X because 980M SLI already does. Disable all those ROP & MC partitions in GM200 to make a 256-bit mobile GPU and you no longer have anything close to resembling Titan X performance. And then there's the little matter of fitting a 600mm^2 die inside a notebook.

So technically, not only would 990M have to be an overclocked full fat GM200, it would have to be somehow scaled down from 250W to 100-125W at the same time.

This is about as believable as the Loch Ness monster, although if Cloudfire is going for comic effect...

 

With these HQ cpu's i dont see any point to more powerful gpu's right now. The current lineup of intel CPU's are bottlenecking the hell out of them.

 

It's fun to think about, though. Even if this isn't true, we can expect to see similar performance output from the 1080M (Pascal).

Just imagine that, guys. It's going to be great.

 

That was my immediate reaction as well. I wonder if Skylake HQ will be fixed... It actually looks like it might be worse.

 

Manufacturers have the power to fix the HQ CPU's. MSI did, the GT80 titan I have with the 4720HQ overclocks and does not throttle at all like the other laptops I tested. BUT, even with the overclock and the elimination of throttling, my 970m's in sli get bottlenecked alot in some games. What we need is an unlocked CPU with a big TDP...then bring on the 990m or whatever

 

The GT80 is a weird case. Since it is such a large machine and has the cooling capability, MSI unleashed the core. The problem is that signs are pointing to Skylake being higher TDP than Haswell at the same clocks (the 4GHz desktop part is said to be 95W TDP when the 4GHz 4790k is rated for 88W TDP) which is worrying for mobile to say the least.

 

GTX Titan X is probably 10-15% faster than 980M SLI in average over a big set of games. 3072 cores - 15% is 2600 cores. Pretty close to the 2500 core GM200 it could be.

Im not saying GTX 990M is 980M SLI. Hasee is. Read the first post.
The TDP will go down to around GTX 780M if they disable memory controllers. Its only 500 more cores than GTX 980.

Its a pretty wild thing to think about. I was certaim we would see a full GM204 instead.

 

I don't see how this is possible. GM200 is too big for mobile and it makes absolutely zero financial sense for nVidia to release something so powerful now that AMD has bowed out of the race. It just would make expectations for Pascal even higher. I really can't see such a performance jump without competition being a driving force.

 

And this is what is going in my Clevo P750ZM refresh, what I was waiting for.

 

If you paired both Titan X and 980M SLI with an identical CPU, say a 4790K, 980M SLI would just edge out Titan X. A lot of mobile CPUs, esp. of the BGA variety, limit 980M SLI at higher frame rates. 980M SLI also wins at 4K since that is much less CPU-bound.

 

No it wouldnt. Single card always win over SLI when they have the same core amount.

"GM200 too big for mobile"
Where did you pull that statement from? The GM200 die is not too big for the MXM cards. Remove the black tape on the heatsink and you got an area big enough for 600mm2.
Nvidia couldnt do GK110 on mobile because a full GK104 was 195W. A full GM204 is 165W. They have thermal headroom to use GM200 inside notebooka.

"Zero financial sense".?
You are pretty bombastic in your statements today Ethrem
I can give you a reason: GTX Titan X 3072 cores. GTX 980TI is said to be 3072 cores too.
What about all the GM200 chips that doesnt meet criteria to become a TitanX/980Ti with damaged cores? Write them off as a loss and toss it in the garbage or find a product they can use these damaged chips on

 

One can only hope.

 

980M SLI has higher core clock, a wider aggregate bus and more ROPs

Look at this: http://www.eurogamer.net/articles/digitalfoundry-2015-msi-gt80-titan-sli-review

980M SLI beats Titan X in 5/9 games at both 1080p and 1440p despite having 3.8 GHz 4980HQ ( if it doesn't throttle) instead of 4.6 GHz 4790K.

 

It makes more sense to release full GM204 and release the GM200 rejects as other desktop SKUs. Big die size means that systems will have to have cooling redesigns to accommodate the larger size for just one card before Pascal comes. I think that's a dumb move. And again, releasing such a high performance card makes zero sense as it will set the bar for Pascal to be a ridiculously large upgrade or people will buy 990M and skip Pascal until the first rebrand or maybe even the second. If 990M is around the same speed of 980M SLI, Pascal is going to have to offer 4K @ 60FPS max settings to get people to upgrade, hence my "zero financial sense" comment. It's better for nVidia to waste the unsuitable chips than unnecessarily raise the performance bar for Pascal and potentially hurt sales of the new architecture, especially when they still have room to grow with GM204.

nVidia is going to skip an opportunity to milk old tech? In a climate with no competition? I'll believe it when I see it

 

One more exhibit on Cloudfire's "single card always win over SLI when they have the same core amount"

750M SLI (2 x 384 shaders): http://www.3dmark.com/3dm11/8427940

765M (768 shaders): http://www.3dmark.com/3dm11/9141801

 

I like this
"He said Wu Haijun, Mars can also make efficient business notebooks"

Either this whole conference is pure comedy or they have a different definition of Mars.

 

@Ethrem:
GM204 is used by many GPUs ranging from 1024 to 2048 cores. They cover the entire field so they can use broken chips on products. Products below are GM206 and GM107.
The only GM200 is GTX Titan X. Which is a full GM200. You have to go all the way down to GM204 to 2048 cores to make a product. Everything between 2048 and 3072 does not exist.

Do you see the difference? Not using these broken GM200 chips on a product "does not make financial sense"

@octiceps:
I was of course talking about chips clocked the same. You should know that Nvidia cards does not scale 100% in SLI.
Go troll somewhere else with your condensending posts. 90% of your posts on this forum is just childish. About time you grow up.

Im out. What a timewaste

 

Hazzah, I was wondering when we'd start hearing some rumors about new chips around Computex time.

 

I've already issued two infractions, keep the flames out of here everyone.

@Cloudfire Have dies this large been used in mobile before? I see your point but they must have extremely high hopes for Pascal if they feel confident enough to release a single GPU this powerful for mobile.

 

It translated to Mars because apparently that's what they're going to name the laptop. Well a literal translation would be "God of War", and Mars is the God of War in Roman mythology.

But that snippet is full of lulz for another reason: it actually mentions something along the lines of Intel and nVidia no longer have a strong competitor, thus progress is expected to slow down. And then in the same breath apparently 990M will come out in Q4 this year, have 980M SLI performance, and cost more.

Full of lulz.

 

Last time we had a big die on mobile was GTX 480M. 529mm^2 GF100. It was a disaster. Nvidia had to disable and underclock the chip so much to stay within thermal and power limits, it could barely outperform mobile GF106/GF116 GPUs like the 460M/560M at less than half the size (238mm^2) and TDP or AMD's 5870M (166mm^2). 480M was discontinued in a matter of months. There hasn't a big die on mobile since.

 

But that was Fermi. I wonder if Maxwell would do better. Then again, that die is even larger and regardless is going to take up more surface area that has to be dissipated. As I've seen with my own Maxwell cards, they don't run cool in every situation. When they are pushed past their "efficiency" point, they're just as hot as Kepler.

 

Are there any notebooks besides the P570WM and 18" AWs that can handle a 150W GPU without totally melting down? That's assuming such a 990M can even fit within a 150W TDP envelope.

 

You're still talking a 250W GPU that needs to be cut down to mobile, which is unprecedented aside from one failed attempt.

At that point, a full fat GM204 would be faster. Bank on it.

 

They managed to fit the 195W GK104 just fine and the Titan X is rated at 250W but maxed out at 247W with an average of 224W during gaming when Tom's Hardware tested it so it's in the realm of possibility, just not probability.

 

Even though 680 is 195W TDP it uses around 170W during gaming

 

Yeah sounds like a disaster ready to happen

 

guys, if intel and notebook manufacturers (not MSI) continue the horrible BGA HQ cpu trend, you wont have to worry about TDP on the 990m, its usage will never hit 100% with a cpu running at tdp-throttled baseclocks lol

 

The future is coming

 

Yes it is...

 

2birds1stone

 

if its real I will grab them 2 x 990M
Anybody wants my 970ms...??

 

I don't doubt that 990m will be 40-50% faster than 980m, however I doubt it will equal 980m SLI. And if Clevo or Alienware offer an SLI version... well then we have one mighty laptop on our hands.

 

More like a fireball.

If this is true, its going to be one hot chip.

 

Yeah 880M which was 125W ran ass hot already. A 990M SLI laptop will be a portable nuclear reactor.

 

It is unlikely this is an accurate depiction of upcoming NVIDIA releases. Don't put so much faith into it. Could be a simple typo.

If it is real, great. I may want one... But the odds are against it being so.

 

If that chip comes to light, it will not be overclockable. I can guarantee that one. Especially if it's in a low TDP envelope.

 

And nVidia will have it vbios blocked, system BIOS blocked, and also will neuter the power delivery to just above spec.

 

With maxwell's "low TDP" being really "high TDP but voltage adjustments to keep it overall-y low", you can't go above design spec in such a low TDP environment. Hell, even modified vBIOSes may have problems for the 990M if it's true, and I 20000% guarantee you it'll never be in a SLI notebook... not unless we get 400W-450W bricks the same size as our current 330W bricks.

 

If you want to overclock it you'll need one of these and a 1.21GW PSU.

 

I bet you if look back at the Chinese website the information is coming from you will see post from one of the guys going haha April fools.

 

Exactly. Once I modded fixed nVidia's broken, throttle-tastic vbios, I've never seen my 970s go below 165W TDP even during quiet scenes. Of course to be fair that's because I forced voltage to be a constant 1.26V when under load. But what's scary is that during intense scenes in BF4, with the unlocked vbios both of my overclocked 970s can draw almost 250W! Holy

 

That's not being "fair", that's "how cards are supposed to work". It's all well and good if their tricks WORKED and people didn't need to FIX them. If their minimum voltage was something like 1.15v and max was 1.25 or something, that'd be different; it's far less likely to crash in that case. But the SLI issues are a whole other ball game. I'm quite happy my kepler works at differing voltages in SLI if I tell them. Seems my slave card needs more voltage than my primary, so I set that much higher and I get some nice stability, and my main card keeps a bit cooler. Can't do that with Maxwell unless a modded vBIOS like yours is installed, which is terrible.

 

That would be insane! What's the point of having such powerful chip if it's crippled to 5MHz core (exagaration)? Is this R9-M295X round two with even more ridiculous doses of insanity? I was thinking and actually expecting full GM204 chip ever since the 980m release. Anyway, if they can pull that one off (the die size suggest it's still on 28nm), having more than 5MHz core clock (means something in-line with the desktop counterpart, not exactly 5MHz) and be coolable with a single (normally sized) fan - WOW, just WOW! Overclocking it would be a whole different story, as most of us know, how Maxwell's efficiency actually works. It would be interesting to see.

 

Trust me you would not want to overclock such a hypothetical 990M. What are the power pins on the MXM connector rated for anyway? I mean do you really want to risk literally melting the MXM slot or having it burst into flames?

 

10A x 19V = 190W You'll need quite the cooling for it though.

 

Not if they use Intel BGA trash...

 

Considering Hasee's gaming line is a Clevo reseller, have we heard anything from Clevo yet?