Pascal: What do we know? Discussion, Latest News & Updates: 1000M Series GPU's

still very much alive, actually:

http://www.computerbase.de/2016-03/moores-law-lebt/

also, its about transistor density / silicon complexity, rather than actual computing power, fine but important difference

Sent from my Huawei Mate 8 NXT-AL10 using Tapatalk

 

2016> Kaby Lake > (X_X)

 

Moore's Law is dead already? That's nearly a decade early...

 

I completely agree with you mate, that's why I've kept this system for so long. It's battery is dead so I keep it plugged all the time. I use it as a home theatre control hooked up to a 60" HD panel for movie sessions with my girlfriend. I play older games on it as well. The laptop has so many scratches that it looks like a cat has been sleeping on it for years. But I love it to bits and I wouldn't certainly discard it. May be I'll resurrect this baby someday.

For now, I'm ready to play the waiting game, till Q3 of 2016 if need be to see how the next generation lines up against the previous ones. If it wasn't for the Pascal hype/Directx12 and VR (Project Cars), I wouldn't even think of buying a new laptop. It's just that VR has got me into a "shut up and take my money" mode.

 

P570WM

 

That thing is like 2 P75xZM on top of each other lol

 

a machine for real men

Sent from my Huawei Mate 8 NXT-AL10 using Tapatalk

 

Good luck finding one of those in the wild, they are equally awesome behemoths.

 

I used to own one

 

Pretty sure Moore's Law technically slows or becomes "untrue" after 14nm because 28nm to 14nm should give them the ability to double performance per core/watt.

However, you must keep in mind that they will never release 2x the performance on the first iteration as they need to profit off of 14nm, which means they'll most likely have 2-3 generations of Pascal, the last being twice as powerful or more. But yes, after 14nm is 10nm, correct? Something like this... It technically shouldn't be possible for pure core performance to double anymore. Other areas can still be improved upon, though. The 1180M or 1280M should be the most powerful of Pascal mobile cards.

 

I think theoretically performance could quadruple when going from 28nm to 14nm, at least in terms of the number of transistors you can fit on a chip:
(28^2)/(14^2) = 4
They missed out the 20nm node - skipped it & went straight to 14nm (in AMD's case) & 16nm (NVidia).

 

Oh okay, well even better. Hope the last Pascal mobile card is 4x the 980M, lol.

 

From GTX580m to GTX 980m, there was a 4X performance increase.

In these conditions, it took them 4 generations and 40 months for this.

There is a chance of Pascal being this much faster, considering the theoretical improvements.

 

Fingers crossed! Although what did it say on that roadmap you or someone else posted up with projected performance improvements through Maxwell/Pascal/Volta/etc? If I remember correctly Pascal wasn't 4x performance of Maxwell even though we're going all the way from 28nm down to 16nm - praps some other reasons why they can't x4 the amounts of transistors on a chip, etc.

 

Supposed new leak about Pascal launch, from AIB manufacturers, putting GTX 1070/1080/1080Ti launch to June:

 

4 iterations and 1 generation, not 4 generations. They intentionally did this because of AMD and Intel starting the re-branding trend. 680M, 780M, 880M, and 980M are all 28nm and were primarily re-brands. I am assuming Pascal will have as many iterations. It has nothing to do with what they're capable of... They just had no competition.

680M is nearly 100% better (2x 580M) in some games, which is why I initally believed the 1080M would be 2x as fast as the 980M. However, it seems less likely now.

 

That is almost like they were doing this on purpose, delaying the release of their best.

Well, as long as Pascal brings a very good quality and with much better oc possibilities, I am all for it!

It is a business after all, and they have to make money out of it.

I just wish they would at least make those cards, at least the 980m - 1080m or whatever they will call it's replacement much more overclockable, to let us tweak to the best extent the performance *evil laugh*

 

Ahem, 8800m, 9800m and 280m say Hi.

 

I just hope they will release full power GPUs now. We need it. 4K needs it. Me playing Neptunia on 3 UHD screens needs it. Going down in the darkness of VR needs it.

 

Don't get your hopes up for seeing high end mobile Pascal anywhere near June...

 

true. Until it becomes a reality and actually available to notebook manufacturers to install in laptops, it might take even longer. That's why I never wait for new tech. Always get the best that's available now unless you have a good system to play with until the next best thing is available.

I have 0 patience personally. That's my biggest weak point.

 

I've been saying it for months now. Don't expect anything high performance before 4Q for mobile and be prepared for Nvidia to bring out the milk machine.

 

That is okay. I just hope they release the desktop versions sooner.
At least, I can make purchase, knowing what to expect at best from next gen of mobile GPUs. I mean, if desktop is 4x maxwell, might as well wait three more months to see what mobile brings. If desktop is under 50% better, might as well buy a gtx980 notebook.

 

Jup and it get's even worse for nvidia when we look at some low end chips:

435m > 540m > 630m > 720m > 820m

Yes they rebranded that bloody old fermi chip 4 times!!!

 

According to Hardware Battle the production of 970, 980 and 980 Ti has stopped:

https://translate.google.se/transla...oard.php?bo_table=news&wr_id=18732&edit-text=

 

Yep, I saw this over on Guru3D recently. They're making way for Pascal, a good sign!

 

The real good mobile parts based on GP104 would come next year probably with GP100 for desktop. We got another year of waiting to go.

 

i guess thats good news for our strained wallets

Sent from my Huawei Mate 8 NXT-AL10 using Tapatalk

 

The only good thats come out this slow upgrade cycle is usually you can skip a generation without problem.

 

the more i read here the happier i am that i got the 4yr extended warranty on my machine. at this pace ill have owned it two years already before there even IS anything to upgrade to gpu-wise. and even then, thats not to say it would be worthwile to do so

Sent from my Huawei Mate 8 NXT-AL10 using Tapatalk

 

They're not going to wait another year to release mobile graphics cards, lol.

You must not fully understand business. They can't afford to wait that long... The 980M came out in Sept. / Oct. 2014. Later this year is far more likely than Q2/2017. That's absurd. And don't forget that Samsung and TSMC HBM2 went into mass production in January... Think they're just going to sit on inventory? Absolutely not.

If they launch desktop at Computex, anything is possible. They launched the 780 and 780M at Computex.

 

They sure as hell are going to milk maxwell for all its worth before releasing pascal. They haven't even released the new Maxwell milk editions yet. And with AMD completely absent from the mobile space, they have no reason to push pascal out. There is not a peep about mobile Polaris yet in the high end area, only for mobile APUs

 

Yeah I´m a little worried if we even see a mobile chip at Computex it´ll be a Maxwell 980MX or something.

Edit: Why do you have two Alienwares with 780M SLI?

 

I upgrade computers on the side and I stuff the pulled parts into my computers or antistatic bags.

 

The gt80 screen goes up to 100hz

 

You mean FHD screen, no?

I really hope they will be pushing 4K 100% adobeRGB screens with 100HZ faster, though our current best GPU could not keep up with it, this is probably why it was not released, for nvidia to not lose face.

 

Your choice of words scares me alot. I really don't want to push my pascal purchase to next year.

 

Sure they can, why not?

Sales arent decling. Not with everyone and their dog buying 980 desktop edition. We might see a response to 50w polaris, but as I said, the high end mobile part are not coming anytime soon.

HBM2s are going on GP100 and GP100 ONLY! If GP104 launches on GDDR5/GDDR5x, it is not going to get HBM2 period. The die aint designed for it and nvidia is not going to redesign their already confirmed working die for HBM2 goes through another round for validation etc. We might see a GP204 with HBM2, we may not.

We might see a move in october for high end mobile if AMD really makes some noise on a high end mobile polaris.

 

Keyword: If.

 

GP104 is launching on GDDR5/GDDR5x period. P100 demands is going to keep up on HBM2 for a long time. Considering the prototype GP100 havent even gotten to the corporate partners yet, it aint ready.

GP104 is NOT coming with HBM2, we already seen the picture of the die....... not interposers.

 

What you saw is not mobile, lol. Those were for cars.

 

GP104 die shot was leaked???

http://www.guru3d.com/news-story/nvidia-pascal-gp104-die-photo.html

 

That looks more like a GM107 replacement than GM204 replacement. Looks entirely different.

Makes sense that GM107 would be replaced by something similar. It always is.

 

Ehe yeah, GM107 replacement is not going to be 300mm^2. Thats a massive Gx107 die. Its the perfect size for Gx104. GM204 is outlier.....

 

They eyeballed the size based on that crap image, lol. It could be zoomed in and appear larger.

There's no way I believe that is GM204 replacement unless they are skipping Pascal.

 

.... there is literally memory chips on the side you can compare with........

Its perfect size and the memory config make sense for a 1070.....

Anyways, if the leaked cooler is legit, it is the ugliest coller I have ever seen.

 

No HBM2 for nVidia cards besides Teslas this year period. GP104 is never going to get it either.

 

How can you look at this and say they'll never use HBM2 in mobile graphics? HBM2 allows for faster computation, thus reducing the volume required to process data.

They will but nobody knows when at the moment. It all depends on what they release this May/June.

 

Mobile HBM2 isn't going to happen for nVidia until Volta. So far their only HBM2 capable GPU is GP100 and if they release a desktop variant of that it will be next year at the earliest or if the rumor of GP102 is correct they will get that instead, and then that has to be HBM2 and not GDDR5X (nVidia is always behind when it comes to memory technology, either they are late or once they catch up the bandwidth is lower and on top of that their memory controller isn't as efficient). As for mobile no way in hell a GP100 is getting into a notebook.

 

Maybe GP204, Volta is pretty back end to say the least.......